sys/cam/ctl/ctl.c

   1 /*-
   2  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
   3  * Copyright (c) 2012 The FreeBSD Foundation
   4  * All rights reserved.
   5  *
   6  * Portions of this software were developed by Edward Tomasz Napierala
   7  * under sponsorship from the FreeBSD Foundation.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions, and the following disclaimer,
  14  *    without modification.
  15  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
  16  *    substantially similar to the "NO WARRANTY" disclaimer below
  17  *    ("Disclaimer") and any redistribution must be conditioned upon
  18  *    including a substantially similar Disclaimer requirement for further
  19  *    binary redistribution.
  20  *
  21  * NO WARRANTY
  22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  23  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
  25  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  26  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  30  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
  31  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  32  * POSSIBILITY OF SUCH DAMAGES.
  33  *
  34  * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
  35  */
  36 /*
  37  * CAM Target Layer, a SCSI device emulation subsystem.
  38  *
  39  * Author: Ken Merry <ken@FreeBSD.org>
  40  */
  41
  42 #define _CTL_C
  43
  44 #include <sys/cdefs.h>
  45 __FBSDID("$FreeBSD$");
  46
  47 #include <sys/param.h>
  48 #include <sys/systm.h>
  49 #include <sys/kernel.h>
  50 #include <sys/types.h>
  51 #include <sys/kthread.h>
  52 #include <sys/bio.h>
  53 #include <sys/fcntl.h>
  54 #include <sys/lock.h>
  55 #include <sys/module.h>
  56 #include <sys/mutex.h>
  57 #include <sys/condvar.h>
  58 #include <sys/malloc.h>
  59 #include <sys/conf.h>
  60 #include <sys/ioccom.h>
  61 #include <sys/queue.h>
  62 #include <sys/sbuf.h>
  63 #include <sys/smp.h>
  64 #include <sys/endian.h>
  65 #include <sys/sysctl.h>
  66
  67 #include <cam/cam.h>
  68 #include <cam/scsi/scsi_all.h>
  69 #include <cam/scsi/scsi_da.h>
  70 #include <cam/ctl/ctl_io.h>
  71 #include <cam/ctl/ctl.h>
  72 #include <cam/ctl/ctl_frontend.h>
  73 #include <cam/ctl/ctl_frontend_internal.h>
  74 #include <cam/ctl/ctl_util.h>
  75 #include <cam/ctl/ctl_backend.h>
  76 #include <cam/ctl/ctl_ioctl.h>
  77 #include <cam/ctl/ctl_ha.h>
  78 #include <cam/ctl/ctl_private.h>
  79 #include <cam/ctl/ctl_debug.h>
  80 #include <cam/ctl/ctl_scsi_all.h>
  81 #include <cam/ctl/ctl_error.h>
  82
  83 struct ctl_softc *control_softc = NULL;
  84
  85 /*
  86  * Size and alignment macros needed for Copan-specific HA hardware.  These
  87  * can go away when the HA code is re-written, and uses busdma for any
  88  * hardware.
  89  */
  90 #define CTL_ALIGN_8B(target, source, type)                              \
  91         if (((uint32_t)source & 0x7) != 0)                              \
  92                 target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
  93         else                                                            \
  94                 target = (type)source;
  95
  96 #define CTL_SIZE_8B(target, size)                                       \
  97         if ((size & 0x7) != 0)                                          \
  98                 target = size + (0x8 - (size & 0x7));                   \
  99         else                                                            \
 100                 target = size;
 101
 102 #define CTL_ALIGN_8B_MARGIN     16
 103
 104 /*
 105  * Template mode pages.
 106  */
 107
 108 /*
 109  * Note that these are default values only.  The actual values will be
 110  * filled in when the user does a mode sense.
 111  */
 112 static struct copan_power_subpage power_page_default = {
 113         /*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
 114         /*subpage*/ PWR_SUBPAGE_CODE,
 115         /*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
 116                          (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
 117         /*page_version*/ PWR_VERSION,
 118         /* total_luns */ 26,
 119         /* max_active_luns*/ PWR_DFLT_MAX_LUNS,
 120         /*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
 121                       0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 122                       0, 0, 0, 0, 0, 0}
 123 };
 124
 125 static struct copan_power_subpage power_page_changeable = {
 126         /*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
 127         /*subpage*/ PWR_SUBPAGE_CODE,
 128         /*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
 129                          (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
 130         /*page_version*/ 0,
 131         /* total_luns */ 0,
 132         /* max_active_luns*/ 0,
 133         /*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
 134                       0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 135                       0, 0, 0, 0, 0, 0}
 136 };
 137
 138 static struct copan_aps_subpage aps_page_default = {
 139         APS_PAGE_CODE | SMPH_SPF, //page_code
 140         APS_SUBPAGE_CODE, //subpage
 141         {(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
 142          (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
 143         APS_VERSION, //page_version
 144         0, //lock_active
 145         {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 146         0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 147         0, 0, 0, 0, 0} //reserved
 148 };
 149
 150 static struct copan_aps_subpage aps_page_changeable = {
 151         APS_PAGE_CODE | SMPH_SPF, //page_code
 152         APS_SUBPAGE_CODE, //subpage
 153         {(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
 154          (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
 155         0, //page_version
 156         0, //lock_active
 157         {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 158         0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 159         0, 0, 0, 0, 0} //reserved
 160 };
 161
 162 static struct copan_debugconf_subpage debugconf_page_default = {
 163         DBGCNF_PAGE_CODE | SMPH_SPF,    /* page_code */
 164         DBGCNF_SUBPAGE_CODE,            /* subpage */
 165         {(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
 166          (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
 167         DBGCNF_VERSION,                 /* page_version */
 168         {CTL_TIME_IO_DEFAULT_SECS>>8,
 169          CTL_TIME_IO_DEFAULT_SECS>>0},  /* ctl_time_io_secs */
 170 };
 171
 172 static struct copan_debugconf_subpage debugconf_page_changeable = {
 173         DBGCNF_PAGE_CODE | SMPH_SPF,    /* page_code */
 174         DBGCNF_SUBPAGE_CODE,            /* subpage */
 175         {(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
 176          (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
 177         0,                              /* page_version */
 178         {0xff,0xff},                    /* ctl_time_io_secs */
 179 };
 180
 181 static struct scsi_format_page format_page_default = {
 182         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 183         /*page_length*/sizeof(struct scsi_format_page) - 2,
 184         /*tracks_per_zone*/ {0, 0},
 185         /*alt_sectors_per_zone*/ {0, 0},
 186         /*alt_tracks_per_zone*/ {0, 0},
 187         /*alt_tracks_per_lun*/ {0, 0},
 188         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
 189                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
 190         /*bytes_per_sector*/ {0, 0},
 191         /*interleave*/ {0, 0},
 192         /*track_skew*/ {0, 0},
 193         /*cylinder_skew*/ {0, 0},
 194         /*flags*/ SFP_HSEC,
 195         /*reserved*/ {0, 0, 0}
 196 };
 197
 198 static struct scsi_format_page format_page_changeable = {
 199         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 200         /*page_length*/sizeof(struct scsi_format_page) - 2,
 201         /*tracks_per_zone*/ {0, 0},
 202         /*alt_sectors_per_zone*/ {0, 0},
 203         /*alt_tracks_per_zone*/ {0, 0},
 204         /*alt_tracks_per_lun*/ {0, 0},
 205         /*sectors_per_track*/ {0, 0},
 206         /*bytes_per_sector*/ {0, 0},
 207         /*interleave*/ {0, 0},
 208         /*track_skew*/ {0, 0},
 209         /*cylinder_skew*/ {0, 0},
 210         /*flags*/ 0,
 211         /*reserved*/ {0, 0, 0}
 212 };
 213
 214 static struct scsi_rigid_disk_page rigid_disk_page_default = {
 215         /*page_code*/SMS_RIGID_DISK_PAGE,
 216         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 217         /*cylinders*/ {0, 0, 0},
 218         /*heads*/ CTL_DEFAULT_HEADS,
 219         /*start_write_precomp*/ {0, 0, 0},
 220         /*start_reduced_current*/ {0, 0, 0},
 221         /*step_rate*/ {0, 0},
 222         /*landing_zone_cylinder*/ {0, 0, 0},
 223         /*rpl*/ SRDP_RPL_DISABLED,
 224         /*rotational_offset*/ 0,
 225         /*reserved1*/ 0,
 226         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
 227                            CTL_DEFAULT_ROTATION_RATE & 0xff},
 228         /*reserved2*/ {0, 0}
 229 };
 230
 231 static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
 232         /*page_code*/SMS_RIGID_DISK_PAGE,
 233         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 234         /*cylinders*/ {0, 0, 0},
 235         /*heads*/ 0,
 236         /*start_write_precomp*/ {0, 0, 0},
 237         /*start_reduced_current*/ {0, 0, 0},
 238         /*step_rate*/ {0, 0},
 239         /*landing_zone_cylinder*/ {0, 0, 0},
 240         /*rpl*/ 0,
 241         /*rotational_offset*/ 0,
 242         /*reserved1*/ 0,
 243         /*rotation_rate*/ {0, 0},
 244         /*reserved2*/ {0, 0}
 245 };
 246
 247 static struct scsi_caching_page caching_page_default = {
 248         /*page_code*/SMS_CACHING_PAGE,
 249         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 250         /*flags1*/ SCP_DISC | SCP_WCE,
 251         /*ret_priority*/ 0,
 252         /*disable_pf_transfer_len*/ {0xff, 0xff},
 253         /*min_prefetch*/ {0, 0},
 254         /*max_prefetch*/ {0xff, 0xff},
 255         /*max_pf_ceiling*/ {0xff, 0xff},
 256         /*flags2*/ 0,
 257         /*cache_segments*/ 0,
 258         /*cache_seg_size*/ {0, 0},
 259         /*reserved*/ 0,
 260         /*non_cache_seg_size*/ {0, 0, 0}
 261 };
 262
 263 static struct scsi_caching_page caching_page_changeable = {
 264         /*page_code*/SMS_CACHING_PAGE,
 265         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 266         /*flags1*/ SCP_WCE | SCP_RCD,
 267         /*ret_priority*/ 0,
 268         /*disable_pf_transfer_len*/ {0, 0},
 269         /*min_prefetch*/ {0, 0},
 270         /*max_prefetch*/ {0, 0},
 271         /*max_pf_ceiling*/ {0, 0},
 272         /*flags2*/ 0,
 273         /*cache_segments*/ 0,
 274         /*cache_seg_size*/ {0, 0},
 275         /*reserved*/ 0,
 276         /*non_cache_seg_size*/ {0, 0, 0}
 277 };
 278
 279 static struct scsi_control_page control_page_default = {
 280         /*page_code*/SMS_CONTROL_MODE_PAGE,
 281         /*page_length*/sizeof(struct scsi_control_page) - 2,
 282         /*rlec*/0,
 283         /*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
 284         /*eca_and_aen*/0,
 285         /*flags4*/SCP_TAS,
 286         /*aen_holdoff_period*/{0, 0},
 287         /*busy_timeout_period*/{0, 0},
 288         /*extended_selftest_completion_time*/{0, 0}
 289 };
 290
 291 static struct scsi_control_page control_page_changeable = {
 292         /*page_code*/SMS_CONTROL_MODE_PAGE,
 293         /*page_length*/sizeof(struct scsi_control_page) - 2,
 294         /*rlec*/SCP_DSENSE,
 295         /*queue_flags*/SCP_QUEUE_ALG_MASK,
 296         /*eca_and_aen*/0,
 297         /*flags4*/0,
 298         /*aen_holdoff_period*/{0, 0},
 299         /*busy_timeout_period*/{0, 0},
 300         /*extended_selftest_completion_time*/{0, 0}
 301 };
 302
 303
 304 /*
 305  * XXX KDM move these into the softc.
 306  */
 307 static int rcv_sync_msg;
 308 static int persis_offset;
 309 static uint8_t ctl_pause_rtr;
 310 static int     ctl_is_single = 1;
 311 static int     index_to_aps_page;
 312
 313 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
 314 static int worker_threads = -1;
 315 TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
 316 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 317     &worker_threads, 1, "Number of worker threads");
 318 static int verbose = 0;
 319 TUNABLE_INT("kern.cam.ctl.verbose", &verbose);
 320 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
 321     &verbose, 0, "Show SCSI errors returned to initiator");
 322
 323 /*
 324  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 325  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 326  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
 327  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
 328  */
 329 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
 330
 331 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 332                                   int param);
 333 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 334 static int ctl_init(void);
 335 void ctl_shutdown(void);
 336 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 337 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 338 static void ctl_ioctl_online(void *arg);
 339 static void ctl_ioctl_offline(void *arg);
 340 static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
 341 static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
 342 static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
 343 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 344 static int ctl_ioctl_submit_wait(union ctl_io *io);
 345 static void ctl_ioctl_datamove(union ctl_io *io);
 346 static void ctl_ioctl_done(union ctl_io *io);
 347 static void ctl_ioctl_hard_startstop_callback(void *arg,
 348                                               struct cfi_metatask *metatask);
 349 static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
 350 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 351                               struct ctl_ooa *ooa_hdr,
 352                               struct ctl_ooa_entry *kern_entries);
 353 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 354                      struct thread *td);
 355 static uint32_t ctl_map_lun(int port_num, uint32_t lun);
 356 static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
 357 #ifdef unused
 358 static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
 359                                    uint32_t targ_target, uint32_t targ_lun,
 360                                    int can_wait);
 361 static void ctl_kfree_io(union ctl_io *io);
 362 #endif /* unused */
 363 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 364                          struct ctl_be_lun *be_lun, struct ctl_id target_id);
 365 static int ctl_free_lun(struct ctl_lun *lun);
 366 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 367 /**
 368 static void ctl_failover_change_pages(struct ctl_softc *softc,
 369                                       struct ctl_scsiio *ctsio, int master);
 370 **/
 371
 372 static int ctl_do_mode_select(union ctl_io *io);
 373 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 374                            uint64_t res_key, uint64_t sa_res_key,
 375                            uint8_t type, uint32_t residx,
 376                            struct ctl_scsiio *ctsio,
 377                            struct scsi_per_res_out *cdb,
 378                            struct scsi_per_res_out_parms* param);
 379 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 380                                   union ctl_ha_msg *msg);
 381 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
 382 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 383 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 384 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 385 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 386 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 387 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 388                                          int alloc_len);
 389 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 390                                          int alloc_len);
 391 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 392 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 393 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 394 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 395 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 396 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
 397 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 398     union ctl_io *pending_io, union ctl_io *ooa_io);
 399 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 400                                 union ctl_io *starting_io);
 401 static int ctl_check_blocked(struct ctl_lun *lun);
 402 static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
 403                                 struct ctl_lun *lun,
 404                                 const struct ctl_cmd_entry *entry,
 405                                 struct ctl_scsiio *ctsio);
 406 //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
 407 static void ctl_failover(void);
 408 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 409                                struct ctl_scsiio *ctsio);
 410 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 411
 412 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 413 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
 414                             ctl_ua_type ua_type);
 415 static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
 416                          ctl_ua_type ua_type);
 417 static int ctl_abort_task(union ctl_io *io);
 418 static int ctl_abort_task_set(union ctl_io *io);
 419 static int ctl_i_t_nexus_reset(union ctl_io *io);
 420 static void ctl_run_task(union ctl_io *io);
 421 #ifdef CTL_IO_DELAY
 422 static void ctl_datamove_timer_wakeup(void *arg);
 423 static void ctl_done_timer_wakeup(void *arg);
 424 #endif /* CTL_IO_DELAY */
 425
 426 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 427 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 428 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 429 static void ctl_datamove_remote_write(union ctl_io *io);
 430 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 431 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 432 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 433 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 434                                     ctl_ha_dt_cb callback);
 435 static void ctl_datamove_remote_read(union ctl_io *io);
 436 static void ctl_datamove_remote(union ctl_io *io);
 437 static int ctl_process_done(union ctl_io *io);
 438 static void ctl_lun_thread(void *arg);
 439 static void ctl_work_thread(void *arg);
 440 static void ctl_enqueue_incoming(union ctl_io *io);
 441 static void ctl_enqueue_rtr(union ctl_io *io);
 442 static void ctl_enqueue_done(union ctl_io *io);
 443 static void ctl_enqueue_isc(union ctl_io *io);
 444 static const struct ctl_cmd_entry *
 445     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 446 static const struct ctl_cmd_entry *
 447     ctl_validate_command(struct ctl_scsiio *ctsio);
 448 static int ctl_cmd_applicable(uint8_t lun_type,
 449     const struct ctl_cmd_entry *entry);
 450
 451 /*
 452  * Load the serialization table.  This isn't very pretty, but is probably
 453  * the easiest way to do it.
 454  */
 455 #include "ctl_ser_table.c"
 456
 457 /*
 458  * We only need to define open, close and ioctl routines for this driver.
 459  */
 460 static struct cdevsw ctl_cdevsw = {
 461         .d_version =    D_VERSION,
 462         .d_flags =      0,
 463         .d_open =       ctl_open,
 464         .d_close =      ctl_close,
 465         .d_ioctl =      ctl_ioctl,
 466         .d_name =       "ctl",
 467 };
 468
 469
 470 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 471 MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
 472
 473 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 474
 475 static moduledata_t ctl_moduledata = {
 476         "ctl",
 477         ctl_module_event_handler,
 478         NULL
 479 };
 480
 481 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 482 MODULE_VERSION(ctl, 1);
 483
 484 static struct ctl_frontend ioctl_frontend =
 485 {
 486         .name = "ioctl",
 487 };
 488
 489 static void
 490 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 491                             union ctl_ha_msg *msg_info)
 492 {
 493         struct ctl_scsiio *ctsio;
 494
 495         if (msg_info->hdr.original_sc == NULL) {
 496                 printf("%s: original_sc == NULL!\n", __func__);
 497                 /* XXX KDM now what? */
 498                 return;
 499         }
 500
 501         ctsio = &msg_info->hdr.original_sc->scsiio;
 502         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 503         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 504         ctsio->io_hdr.status = msg_info->hdr.status;
 505         ctsio->scsi_status = msg_info->scsi.scsi_status;
 506         ctsio->sense_len = msg_info->scsi.sense_len;
 507         ctsio->sense_residual = msg_info->scsi.sense_residual;
 508         ctsio->residual = msg_info->scsi.residual;
 509         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 510                sizeof(ctsio->sense_data));
 511         memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
 512                &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
 513         ctl_enqueue_isc((union ctl_io *)ctsio);
 514 }
 515
 516 static void
 517 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 518                                 union ctl_ha_msg *msg_info)
 519 {
 520         struct ctl_scsiio *ctsio;
 521
 522         if (msg_info->hdr.serializing_sc == NULL) {
 523                 printf("%s: serializing_sc == NULL!\n", __func__);
 524                 /* XXX KDM now what? */
 525                 return;
 526         }
 527
 528         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 529 #if 0
 530         /*
 531          * Attempt to catch the situation where an I/O has
 532          * been freed, and we're using it again.
 533          */
 534         if (ctsio->io_hdr.io_type == 0xff) {
 535                 union ctl_io *tmp_io;
 536                 tmp_io = (union ctl_io *)ctsio;
 537                 printf("%s: %p use after free!\n", __func__,
 538                        ctsio);
 539                 printf("%s: type %d msg %d cdb %x iptl: "
 540                        "%d:%d:%d:%d tag 0x%04x "
 541                        "flag %#x status %x\n",
 542                         __func__,
 543                         tmp_io->io_hdr.io_type,
 544                         tmp_io->io_hdr.msg_type,
 545                         tmp_io->scsiio.cdb[0],
 546                         tmp_io->io_hdr.nexus.initid.id,
 547                         tmp_io->io_hdr.nexus.targ_port,
 548                         tmp_io->io_hdr.nexus.targ_target.id,
 549                         tmp_io->io_hdr.nexus.targ_lun,
 550                         (tmp_io->io_hdr.io_type ==
 551                         CTL_IO_TASK) ?
 552                         tmp_io->taskio.tag_num :
 553                         tmp_io->scsiio.tag_num,
 554                         tmp_io->io_hdr.flags,
 555                         tmp_io->io_hdr.status);
 556         }
 557 #endif
 558         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 559         ctl_enqueue_isc((union ctl_io *)ctsio);
 560 }
 561
 562 /*
 563  * ISC (Inter Shelf Communication) event handler.  Events from the HA
 564  * subsystem come in here.
 565  */
 566 static void
 567 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
 568 {
 569         struct ctl_softc *ctl_softc;
 570         union ctl_io *io;
 571         struct ctl_prio *presio;
 572         ctl_ha_status isc_status;
 573
 574         ctl_softc = control_softc;
 575         io = NULL;
 576
 577
 578 #if 0
 579         printf("CTL: Isc Msg event %d\n", event);
 580 #endif
 581         if (event == CTL_HA_EVT_MSG_RECV) {
 582                 union ctl_ha_msg msg_info;
 583
 584                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
 585                                              sizeof(msg_info), /*wait*/ 0);
 586 #if 0
 587                 printf("CTL: msg_type %d\n", msg_info.msg_type);
 588 #endif
 589                 if (isc_status != 0) {
 590                         printf("Error receiving message, status = %d\n",
 591                                isc_status);
 592                         return;
 593                 }
 594
 595                 switch (msg_info.hdr.msg_type) {
 596                 case CTL_MSG_SERIALIZE:
 597 #if 0
 598                         printf("Serialize\n");
 599 #endif
 600                         io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
 601                         if (io == NULL) {
 602                                 printf("ctl_isc_event_handler: can't allocate "
 603                                        "ctl_io!\n");
 604                                 /* Bad Juju */
 605                                 /* Need to set busy and send msg back */
 606                                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
 607                                 msg_info.hdr.status = CTL_SCSI_ERROR;
 608                                 msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
 609                                 msg_info.scsi.sense_len = 0;
 610                                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
 611                                     sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
 612                                 }
 613                                 goto bailout;
 614                         }
 615                         ctl_zero_io(io);
 616                         // populate ctsio from msg_info
 617                         io->io_hdr.io_type = CTL_IO_SCSI;
 618                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
 619                         io->io_hdr.original_sc = msg_info.hdr.original_sc;
 620 #if 0
 621                         printf("pOrig %x\n", (int)msg_info.original_sc);
 622 #endif
 623                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
 624                                             CTL_FLAG_IO_ACTIVE;
 625                         /*
 626                          * If we're in serialization-only mode, we don't
 627                          * want to go through full done processing.  Thus
 628                          * the COPY flag.
 629                          *
 630                          * XXX KDM add another flag that is more specific.
 631                          */
 632                         if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
 633                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
 634                         io->io_hdr.nexus = msg_info.hdr.nexus;
 635 #if 0
 636                         printf("targ %d, port %d, iid %d, lun %d\n",
 637                                io->io_hdr.nexus.targ_target.id,
 638                                io->io_hdr.nexus.targ_port,
 639                                io->io_hdr.nexus.initid.id,
 640                                io->io_hdr.nexus.targ_lun);
 641 #endif
 642                         io->scsiio.tag_num = msg_info.scsi.tag_num;
 643                         io->scsiio.tag_type = msg_info.scsi.tag_type;
 644                         memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
 645                                CTL_MAX_CDBLEN);
 646                         if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
 647                                 const struct ctl_cmd_entry *entry;
 648
 649                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
 650                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
 651                                 io->io_hdr.flags |=
 652                                         entry->flags & CTL_FLAG_DATA_MASK;
 653                         }
 654                         ctl_enqueue_isc(io);
 655                         break;
 656
 657                 /* Performed on the Originating SC, XFER mode only */
 658                 case CTL_MSG_DATAMOVE: {
 659                         struct ctl_sg_entry *sgl;
 660                         int i, j;
 661
 662                         io = msg_info.hdr.original_sc;
 663                         if (io == NULL) {
 664                                 printf("%s: original_sc == NULL!\n", __func__);
 665                                 /* XXX KDM do something here */
 666                                 break;
 667                         }
 668                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
 669                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 670                         /*
 671                          * Keep track of this, we need to send it back over
 672                          * when the datamove is complete.
 673                          */
 674                         io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
 675
 676                         if (msg_info.dt.sg_sequence == 0) {
 677                                 /*
 678                                  * XXX KDM we use the preallocated S/G list
 679                                  * here, but we'll need to change this to
 680                                  * dynamic allocation if we need larger S/G
 681                                  * lists.
 682                                  */
 683                                 if (msg_info.dt.kern_sg_entries >
 684                                     sizeof(io->io_hdr.remote_sglist) /
 685                                     sizeof(io->io_hdr.remote_sglist[0])) {
 686                                         printf("%s: number of S/G entries "
 687                                             "needed %u > allocated num %zd\n",
 688                                             __func__,
 689                                             msg_info.dt.kern_sg_entries,
 690                                             sizeof(io->io_hdr.remote_sglist)/
 691                                             sizeof(io->io_hdr.remote_sglist[0]));
 692
 693                                         /*
 694                                          * XXX KDM send a message back to
 695                                          * the other side to shut down the
 696                                          * DMA.  The error will come back
 697                                          * through via the normal channel.
 698                                          */
 699                                         break;
 700                                 }
 701                                 sgl = io->io_hdr.remote_sglist;
 702                                 memset(sgl, 0,
 703                                        sizeof(io->io_hdr.remote_sglist));
 704
 705                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
 706
 707                                 io->scsiio.kern_sg_entries =
 708                                         msg_info.dt.kern_sg_entries;
 709                                 io->scsiio.rem_sg_entries =
 710                                         msg_info.dt.kern_sg_entries;
 711                                 io->scsiio.kern_data_len =
 712                                         msg_info.dt.kern_data_len;
 713                                 io->scsiio.kern_total_len =
 714                                         msg_info.dt.kern_total_len;
 715                                 io->scsiio.kern_data_resid =
 716                                         msg_info.dt.kern_data_resid;
 717                                 io->scsiio.kern_rel_offset =
 718                                         msg_info.dt.kern_rel_offset;
 719                                 /*
 720                                  * Clear out per-DMA flags.
 721                                  */
 722                                 io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
 723                                 /*
 724                                  * Add per-DMA flags that are set for this
 725                                  * particular DMA request.
 726                                  */
 727                                 io->io_hdr.flags |= msg_info.dt.flags &
 728                                                     CTL_FLAG_RDMA_MASK;
 729                         } else
 730                                 sgl = (struct ctl_sg_entry *)
 731                                         io->scsiio.kern_data_ptr;
 732
 733                         for (i = msg_info.dt.sent_sg_entries, j = 0;
 734                              i < (msg_info.dt.sent_sg_entries +
 735                              msg_info.dt.cur_sg_entries); i++, j++) {
 736                                 sgl[i].addr = msg_info.dt.sg_list[j].addr;
 737                                 sgl[i].len = msg_info.dt.sg_list[j].len;
 738
 739 #if 0
 740                                 printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
 741                                        __func__,
 742                                        msg_info.dt.sg_list[j].addr,
 743                                        msg_info.dt.sg_list[j].len,
 744                                        sgl[i].addr, sgl[i].len, j, i);
 745 #endif
 746                         }
 747 #if 0
 748                         memcpy(&sgl[msg_info.dt.sent_sg_entries],
 749                                msg_info.dt.sg_list,
 750                                sizeof(*sgl) * msg_info.dt.cur_sg_entries);
 751 #endif
 752
 753                         /*
 754                          * If this is the last piece of the I/O, we've got
 755                          * the full S/G list.  Queue processing in the thread.
 756                          * Otherwise wait for the next piece.
 757                          */
 758                         if (msg_info.dt.sg_last != 0)
 759                                 ctl_enqueue_isc(io);
 760                         break;
 761                 }
 762                 /* Performed on the Serializing (primary) SC, XFER mode only */
 763                 case CTL_MSG_DATAMOVE_DONE: {
 764                         if (msg_info.hdr.serializing_sc == NULL) {
 765                                 printf("%s: serializing_sc == NULL!\n",
 766                                        __func__);
 767                                 /* XXX KDM now what? */
 768                                 break;
 769                         }
 770                         /*
 771                          * We grab the sense information here in case
 772                          * there was a failure, so we can return status
 773                          * back to the initiator.
 774                          */
 775                         io = msg_info.hdr.serializing_sc;
 776                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
 777                         io->io_hdr.status = msg_info.hdr.status;
 778                         io->scsiio.scsi_status = msg_info.scsi.scsi_status;
 779                         io->scsiio.sense_len = msg_info.scsi.sense_len;
 780                         io->scsiio.sense_residual =msg_info.scsi.sense_residual;
 781                         io->io_hdr.port_status = msg_info.scsi.fetd_status;
 782                         io->scsiio.residual = msg_info.scsi.residual;
 783                         memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
 784                                sizeof(io->scsiio.sense_data));
 785                         ctl_enqueue_isc(io);
 786                         break;
 787                 }
 788
 789                 /* Preformed on Originating SC, SER_ONLY mode */
 790                 case CTL_MSG_R2R:
 791                         io = msg_info.hdr.original_sc;
 792                         if (io == NULL) {
 793                                 printf("%s: Major Bummer\n", __func__);
 794                                 return;
 795                         } else {
 796 #if 0
 797                                 printf("pOrig %x\n",(int) ctsio);
 798 #endif
 799                         }
 800                         io->io_hdr.msg_type = CTL_MSG_R2R;
 801                         io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
 802                         ctl_enqueue_isc(io);
 803                         break;
 804
 805                 /*
 806                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
 807                  * mode.
 808                  * Performed on the Originating (i.e. secondary) SC in XFER
 809                  * mode
 810                  */
 811                 case CTL_MSG_FINISH_IO:
 812                         if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
 813                                 ctl_isc_handler_finish_xfer(ctl_softc,
 814                                                             &msg_info);
 815                         else
 816                                 ctl_isc_handler_finish_ser_only(ctl_softc,
 817                                                                 &msg_info);
 818                         break;
 819
 820                 /* Preformed on Originating SC */
 821                 case CTL_MSG_BAD_JUJU:
 822                         io = msg_info.hdr.original_sc;
 823                         if (io == NULL) {
 824                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
 825                                        __func__);
 826                                 break;
 827                         }
 828                         ctl_copy_sense_data(&msg_info, io);
 829                         /*
 830                          * IO should have already been cleaned up on other
 831                          * SC so clear this flag so we won't send a message
 832                          * back to finish the IO there.
 833                          */
 834                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
 835                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 836
 837                         /* io = msg_info.hdr.serializing_sc; */
 838                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
 839                         ctl_enqueue_isc(io);
 840                         break;
 841
 842                 /* Handle resets sent from the other side */
 843                 case CTL_MSG_MANAGE_TASKS: {
 844                         struct ctl_taskio *taskio;
 845                         taskio = (struct ctl_taskio *)ctl_alloc_io(
 846                                 (void *)ctl_softc->othersc_pool);
 847                         if (taskio == NULL) {
 848                                 printf("ctl_isc_event_handler: can't allocate "
 849                                        "ctl_io!\n");
 850                                 /* Bad Juju */
 851                                 /* should I just call the proper reset func
 852                                    here??? */
 853                                 goto bailout;
 854                         }
 855                         ctl_zero_io((union ctl_io *)taskio);
 856                         taskio->io_hdr.io_type = CTL_IO_TASK;
 857                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
 858                         taskio->io_hdr.nexus = msg_info.hdr.nexus;
 859                         taskio->task_action = msg_info.task.task_action;
 860                         taskio->tag_num = msg_info.task.tag_num;
 861                         taskio->tag_type = msg_info.task.tag_type;
 862 #ifdef CTL_TIME_IO
 863                         taskio->io_hdr.start_time = time_uptime;
 864                         getbintime(&taskio->io_hdr.start_bt);
 865 #if 0
 866                         cs_prof_gettime(&taskio->io_hdr.start_ticks);
 867 #endif
 868 #endif /* CTL_TIME_IO */
 869                         ctl_run_task((union ctl_io *)taskio);
 870                         break;
 871                 }
 872                 /* Persistent Reserve action which needs attention */
 873                 case CTL_MSG_PERS_ACTION:
 874                         presio = (struct ctl_prio *)ctl_alloc_io(
 875                                 (void *)ctl_softc->othersc_pool);
 876                         if (presio == NULL) {
 877                                 printf("ctl_isc_event_handler: can't allocate "
 878                                        "ctl_io!\n");
 879                                 /* Bad Juju */
 880                                 /* Need to set busy and send msg back */
 881                                 goto bailout;
 882                         }
 883                         ctl_zero_io((union ctl_io *)presio);
 884                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
 885                         presio->pr_msg = msg_info.pr;
 886                         ctl_enqueue_isc((union ctl_io *)presio);
 887                         break;
 888                 case CTL_MSG_SYNC_FE:
 889                         rcv_sync_msg = 1;
 890                         break;
 891                 case CTL_MSG_APS_LOCK: {
 892                         // It's quicker to execute this then to
 893                         // queue it.
 894                         struct ctl_lun *lun;
 895                         struct ctl_page_index *page_index;
 896                         struct copan_aps_subpage *current_sp;
 897                         uint32_t targ_lun;
 898
 899                         targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
 900                         lun = ctl_softc->ctl_luns[targ_lun];
 901                         mtx_lock(&lun->lun_lock);
 902                         page_index = &lun->mode_pages.index[index_to_aps_page];
 903                         current_sp = (struct copan_aps_subpage *)
 904                                      (page_index->page_data +
 905                                      (page_index->page_len * CTL_PAGE_CURRENT));
 906
 907                         current_sp->lock_active = msg_info.aps.lock_flag;
 908                         mtx_unlock(&lun->lun_lock);
 909                         break;
 910                 }
 911                 default:
 912                         printf("How did I get here?\n");
 913                 }
 914         } else if (event == CTL_HA_EVT_MSG_SENT) {
 915                 if (param != CTL_HA_STATUS_SUCCESS) {
 916                         printf("Bad status from ctl_ha_msg_send status %d\n",
 917                                param);
 918                 }
 919                 return;
 920         } else if (event == CTL_HA_EVT_DISCONNECT) {
 921                 printf("CTL: Got a disconnect from Isc\n");
 922                 return;
 923         } else {
 924                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
 925                 return;
 926         }
 927
 928 bailout:
 929         return;
 930 }
 931
 932 static void
 933 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
 934 {
 935         struct scsi_sense_data *sense;
 936
 937         sense = &dest->scsiio.sense_data;
 938         bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
 939         dest->scsiio.scsi_status = src->scsi.scsi_status;
 940         dest->scsiio.sense_len = src->scsi.sense_len;
 941         dest->io_hdr.status = src->hdr.status;
 942 }
 943
 944 static int
 945 ctl_init(void)
 946 {
 947         struct ctl_softc *softc;
 948         struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
 949         struct ctl_port *port;
 950         uint8_t sc_id =0;
 951         int i, error, retval;
 952         //int isc_retval;
 953
 954         retval = 0;
 955         ctl_pause_rtr = 0;
 956         rcv_sync_msg = 0;
 957
 958         control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
 959                                M_WAITOK | M_ZERO);
 960         softc = control_softc;
 961
 962         softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
 963                               "cam/ctl");
 964
 965         softc->dev->si_drv1 = softc;
 966
 967         /*
 968          * By default, return a "bad LUN" peripheral qualifier for unknown
 969          * LUNs.  The user can override this default using the tunable or
 970          * sysctl.  See the comment in ctl_inquiry_std() for more details.
 971          */
 972         softc->inquiry_pq_no_lun = 1;
 973         TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
 974                           &softc->inquiry_pq_no_lun);
 975         sysctl_ctx_init(&softc->sysctl_ctx);
 976         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
 977                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
 978                 CTLFLAG_RD, 0, "CAM Target Layer");
 979
 980         if (softc->sysctl_tree == NULL) {
 981                 printf("%s: unable to allocate sysctl tree\n", __func__);
 982                 destroy_dev(softc->dev);
 983                 free(control_softc, M_DEVBUF);
 984                 control_softc = NULL;
 985                 return (ENOMEM);
 986         }
 987
 988         SYSCTL_ADD_INT(&softc->sysctl_ctx,
 989                        SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
 990                        "inquiry_pq_no_lun", CTLFLAG_RW,
 991                        &softc->inquiry_pq_no_lun, 0,
 992                        "Report no lun possible for invalid LUNs");
 993
 994         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
 995         mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
 996         softc->open_count = 0;
 997
 998         /*
 999          * Default to actually sending a SYNCHRONIZE CACHE command down to
1000          * the drive.
1001          */
1002         softc->flags = CTL_FLAG_REAL_SYNC;
1003
1004         /*
1005          * In Copan's HA scheme, the "master" and "slave" roles are
1006          * figured out through the slot the controller is in.  Although it
1007          * is an active/active system, someone has to be in charge.
1008          */
1009 #ifdef NEEDTOPORT
1010         scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1011 #endif
1012
1013         if (sc_id == 0) {
1014                 softc->flags |= CTL_FLAG_MASTER_SHELF;
1015                 persis_offset = 0;
1016         } else
1017                 persis_offset = CTL_MAX_INITIATORS;
1018
1019         /*
1020          * XXX KDM need to figure out where we want to get our target ID
1021          * and WWID.  Is it different on each port?
1022          */
1023         softc->target.id = 0;
1024         softc->target.wwid[0] = 0x12345678;
1025         softc->target.wwid[1] = 0x87654321;
1026         STAILQ_INIT(&softc->lun_list);
1027         STAILQ_INIT(&softc->pending_lun_queue);
1028         STAILQ_INIT(&softc->fe_list);
1029         STAILQ_INIT(&softc->port_list);
1030         STAILQ_INIT(&softc->be_list);
1031         STAILQ_INIT(&softc->io_pools);
1032         ctl_tpc_init(softc);
1033
1034         if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1035                             &internal_pool)!= 0){
1036                 printf("ctl: can't allocate %d entry internal pool, "
1037                        "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1038                 return (ENOMEM);
1039         }
1040
1041         if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1042                             CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1043                 printf("ctl: can't allocate %d entry emergency pool, "
1044                        "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1045                 ctl_pool_free(internal_pool);
1046                 return (ENOMEM);
1047         }
1048
1049         if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1050                             &other_pool) != 0)
1051         {
1052                 printf("ctl: can't allocate %d entry other SC pool, "
1053                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1054                 ctl_pool_free(internal_pool);
1055                 ctl_pool_free(emergency_pool);
1056                 return (ENOMEM);
1057         }
1058
1059         softc->internal_pool = internal_pool;
1060         softc->emergency_pool = emergency_pool;
1061         softc->othersc_pool = other_pool;
1062
1063         if (worker_threads <= 0)
1064                 worker_threads = max(1, mp_ncpus / 4);
1065         if (worker_threads > CTL_MAX_THREADS)
1066                 worker_threads = CTL_MAX_THREADS;
1067
1068         for (i = 0; i < worker_threads; i++) {
1069                 struct ctl_thread *thr = &softc->threads[i];
1070
1071                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1072                 thr->ctl_softc = softc;
1073                 STAILQ_INIT(&thr->incoming_queue);
1074                 STAILQ_INIT(&thr->rtr_queue);
1075                 STAILQ_INIT(&thr->done_queue);
1076                 STAILQ_INIT(&thr->isc_queue);
1077
1078                 error = kproc_kthread_add(ctl_work_thread, thr,
1079                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1080                 if (error != 0) {
1081                         printf("error creating CTL work thread!\n");
1082                         ctl_pool_free(internal_pool);
1083                         ctl_pool_free(emergency_pool);
1084                         ctl_pool_free(other_pool);
1085                         return (error);
1086                 }
1087         }
1088         error = kproc_kthread_add(ctl_lun_thread, softc,
1089             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1090         if (error != 0) {
1091                 printf("error creating CTL lun thread!\n");
1092                 ctl_pool_free(internal_pool);
1093                 ctl_pool_free(emergency_pool);
1094                 ctl_pool_free(other_pool);
1095                 return (error);
1096         }
1097         if (bootverbose)
1098                 printf("ctl: CAM Target Layer loaded\n");
1099
1100         /*
1101          * Initialize the ioctl front end.
1102          */
1103         ctl_frontend_register(&ioctl_frontend);
1104         port = &softc->ioctl_info.port;
1105         port->frontend = &ioctl_frontend;
1106         sprintf(softc->ioctl_info.port_name, "ioctl");
1107         port->port_type = CTL_PORT_IOCTL;
1108         port->num_requested_ctl_io = 100;
1109         port->port_name = softc->ioctl_info.port_name;
1110         port->port_online = ctl_ioctl_online;
1111         port->port_offline = ctl_ioctl_offline;
1112         port->onoff_arg = &softc->ioctl_info;
1113         port->lun_enable = ctl_ioctl_lun_enable;
1114         port->lun_disable = ctl_ioctl_lun_disable;
1115         port->targ_lun_arg = &softc->ioctl_info;
1116         port->fe_datamove = ctl_ioctl_datamove;
1117         port->fe_done = ctl_ioctl_done;
1118         port->max_targets = 15;
1119         port->max_target_id = 15;
1120
1121         if (ctl_port_register(&softc->ioctl_info.port,
1122                           (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1123                 printf("ctl: ioctl front end registration failed, will "
1124                        "continue anyway\n");
1125         }
1126
1127 #ifdef CTL_IO_DELAY
1128         if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1129                 printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1130                        sizeof(struct callout), CTL_TIMER_BYTES);
1131                 return (EINVAL);
1132         }
1133 #endif /* CTL_IO_DELAY */
1134
1135         return (0);
1136 }
1137
1138 void
1139 ctl_shutdown(void)
1140 {
1141         struct ctl_softc *softc;
1142         struct ctl_lun *lun, *next_lun;
1143         struct ctl_io_pool *pool;
1144
1145         softc = (struct ctl_softc *)control_softc;
1146
1147         if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1148                 printf("ctl: ioctl front end deregistration failed\n");
1149
1150         mtx_lock(&softc->ctl_lock);
1151
1152         /*
1153          * Free up each LUN.
1154          */
1155         for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1156                 next_lun = STAILQ_NEXT(lun, links);
1157                 ctl_free_lun(lun);
1158         }
1159
1160         mtx_unlock(&softc->ctl_lock);
1161
1162         ctl_frontend_deregister(&ioctl_frontend);
1163
1164         /*
1165          * This will rip the rug out from under any FETDs or anyone else
1166          * that has a pool allocated.  Since we increment our module
1167          * refcount any time someone outside the main CTL module allocates
1168          * a pool, we shouldn't have any problems here.  The user won't be
1169          * able to unload the CTL module until client modules have
1170          * successfully unloaded.
1171          */
1172         while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1173                 ctl_pool_free(pool);
1174
1175 #if 0
1176         ctl_shutdown_thread(softc->work_thread);
1177         mtx_destroy(&softc->queue_lock);
1178 #endif
1179
1180         ctl_tpc_shutdown(softc);
1181         mtx_destroy(&softc->pool_lock);
1182         mtx_destroy(&softc->ctl_lock);
1183
1184         destroy_dev(softc->dev);
1185
1186         sysctl_ctx_free(&softc->sysctl_ctx);
1187
1188         free(control_softc, M_DEVBUF);
1189         control_softc = NULL;
1190
1191         if (bootverbose)
1192                 printf("ctl: CAM Target Layer unloaded\n");
1193 }
1194
1195 static int
1196 ctl_module_event_handler(module_t mod, int what, void *arg)
1197 {
1198
1199         switch (what) {
1200         case MOD_LOAD:
1201                 return (ctl_init());
1202         case MOD_UNLOAD:
1203                 return (EBUSY);
1204         default:
1205                 return (EOPNOTSUPP);
1206         }
1207 }
1208
1209 /*
1210  * XXX KDM should we do some access checks here?  Bump a reference count to
1211  * prevent a CTL module from being unloaded while someone has it open?
1212  */
1213 static int
1214 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1215 {
1216         return (0);
1217 }
1218
1219 static int
1220 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1221 {
1222         return (0);
1223 }
1224
1225 int
1226 ctl_port_enable(ctl_port_type port_type)
1227 {
1228         struct ctl_softc *softc;
1229         struct ctl_port *port;
1230
1231         if (ctl_is_single == 0) {
1232                 union ctl_ha_msg msg_info;
1233                 int isc_retval;
1234
1235 #if 0
1236                 printf("%s: HA mode, synchronizing frontend enable\n",
1237                         __func__);
1238 #endif
1239                 msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1240                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1241                         sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1242                         printf("Sync msg send error retval %d\n", isc_retval);
1243                 }
1244                 if (!rcv_sync_msg) {
1245                         isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1246                                 sizeof(msg_info), 1);
1247                 }
1248 #if 0
1249                 printf("CTL:Frontend Enable\n");
1250         } else {
1251                 printf("%s: single mode, skipping frontend synchronization\n",
1252                         __func__);
1253 #endif
1254         }
1255
1256         softc = control_softc;
1257
1258         STAILQ_FOREACH(port, &softc->port_list, links) {
1259                 if (port_type & port->port_type)
1260                 {
1261 #if 0
1262                         printf("port %d\n", port->targ_port);
1263 #endif
1264                         ctl_port_online(port);
1265                 }
1266         }
1267
1268         return (0);
1269 }
1270
1271 int
1272 ctl_port_disable(ctl_port_type port_type)
1273 {
1274         struct ctl_softc *softc;
1275         struct ctl_port *port;
1276
1277         softc = control_softc;
1278
1279         STAILQ_FOREACH(port, &softc->port_list, links) {
1280                 if (port_type & port->port_type)
1281                         ctl_port_offline(port);
1282         }
1283
1284         return (0);
1285 }
1286
1287 /*
1288  * Returns 0 for success, 1 for failure.
1289  * Currently the only failure mode is if there aren't enough entries
1290  * allocated.  So, in case of a failure, look at num_entries_dropped,
1291  * reallocate and try again.
1292  */
1293 int
1294 ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1295               int *num_entries_filled, int *num_entries_dropped,
1296               ctl_port_type port_type, int no_virtual)
1297 {
1298         struct ctl_softc *softc;
1299         struct ctl_port *port;
1300         int entries_dropped, entries_filled;
1301         int retval;
1302         int i;
1303
1304         softc = control_softc;
1305
1306         retval = 0;
1307         entries_filled = 0;
1308         entries_dropped = 0;
1309
1310         i = 0;
1311         mtx_lock(&softc->ctl_lock);
1312         STAILQ_FOREACH(port, &softc->port_list, links) {
1313                 struct ctl_port_entry *entry;
1314
1315                 if ((port->port_type & port_type) == 0)
1316                         continue;
1317
1318                 if ((no_virtual != 0)
1319                  && (port->virtual_port != 0))
1320                         continue;
1321
1322                 if (entries_filled >= num_entries_alloced) {
1323                         entries_dropped++;
1324                         continue;
1325                 }
1326                 entry = &entries[i];
1327
1328                 entry->port_type = port->port_type;
1329                 strlcpy(entry->port_name, port->port_name,
1330                         sizeof(entry->port_name));
1331                 entry->physical_port = port->physical_port;
1332                 entry->virtual_port = port->virtual_port;
1333                 entry->wwnn = port->wwnn;
1334                 entry->wwpn = port->wwpn;
1335
1336                 i++;
1337                 entries_filled++;
1338         }
1339
1340         mtx_unlock(&softc->ctl_lock);
1341
1342         if (entries_dropped > 0)
1343                 retval = 1;
1344
1345         *num_entries_dropped = entries_dropped;
1346         *num_entries_filled = entries_filled;
1347
1348         return (retval);
1349 }
1350
1351 static void
1352 ctl_ioctl_online(void *arg)
1353 {
1354         struct ctl_ioctl_info *ioctl_info;
1355
1356         ioctl_info = (struct ctl_ioctl_info *)arg;
1357
1358         ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1359 }
1360
1361 static void
1362 ctl_ioctl_offline(void *arg)
1363 {
1364         struct ctl_ioctl_info *ioctl_info;
1365
1366         ioctl_info = (struct ctl_ioctl_info *)arg;
1367
1368         ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1369 }
1370
1371 /*
1372  * Remove an initiator by port number and initiator ID.
1373  * Returns 0 for success, -1 for failure.
1374  */
1375 int
1376 ctl_remove_initiator(struct ctl_port *port, int iid)
1377 {
1378         struct ctl_softc *softc = control_softc;
1379
1380         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1381
1382         if (iid > CTL_MAX_INIT_PER_PORT) {
1383                 printf("%s: initiator ID %u > maximun %u!\n",
1384                        __func__, iid, CTL_MAX_INIT_PER_PORT);
1385                 return (-1);
1386         }
1387
1388         mtx_lock(&softc->ctl_lock);
1389         port->wwpn_iid[iid].in_use--;
1390         port->wwpn_iid[iid].last_use = time_uptime;
1391         mtx_unlock(&softc->ctl_lock);
1392
1393         return (0);
1394 }
1395
1396 /*
1397  * Add an initiator to the initiator map.
1398  * Returns iid for success, < 0 for failure.
1399  */
1400 int
1401 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1402 {
1403         struct ctl_softc *softc = control_softc;
1404         time_t best_time;
1405         int i, best;
1406
1407         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1408
1409         if (iid >= CTL_MAX_INIT_PER_PORT) {
1410                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1411                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1412                 free(name, M_CTL);
1413                 return (-1);
1414         }
1415
1416         mtx_lock(&softc->ctl_lock);
1417
1418         if (iid < 0 && (wwpn != 0 || name != NULL)) {
1419                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1420                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1421                                 iid = i;
1422                                 break;
1423                         }
1424                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
1425                             strcmp(name, port->wwpn_iid[i].name) == 0) {
1426                                 iid = i;
1427                                 break;
1428                         }
1429                 }
1430         }
1431
1432         if (iid < 0) {
1433                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1434                         if (port->wwpn_iid[i].in_use == 0 &&
1435                             port->wwpn_iid[i].wwpn == 0 &&
1436                             port->wwpn_iid[i].name == NULL) {
1437                                 iid = i;
1438                                 break;
1439                         }
1440                 }
1441         }
1442
1443         if (iid < 0) {
1444                 best = -1;
1445                 best_time = INT32_MAX;
1446                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1447                         if (port->wwpn_iid[i].in_use == 0) {
1448                                 if (port->wwpn_iid[i].last_use < best_time) {
1449                                         best = i;
1450                                         best_time = port->wwpn_iid[i].last_use;
1451                                 }
1452                         }
1453                 }
1454                 iid = best;
1455         }
1456
1457         if (iid < 0) {
1458                 mtx_unlock(&softc->ctl_lock);
1459                 free(name, M_CTL);
1460                 return (-2);
1461         }
1462
1463         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1464                 /*
1465                  * This is not an error yet.
1466                  */
1467                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1468 #if 0
1469                         printf("%s: port %d iid %u WWPN %#jx arrived"
1470                             " again\n", __func__, port->targ_port,
1471                             iid, (uintmax_t)wwpn);
1472 #endif
1473                         goto take;
1474                 }
1475                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1476                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
1477 #if 0
1478                         printf("%s: port %d iid %u name '%s' arrived"
1479                             " again\n", __func__, port->targ_port,
1480                             iid, name);
1481 #endif
1482                         goto take;
1483                 }
1484
1485                 /*
1486                  * This is an error, but what do we do about it?  The
1487                  * driver is telling us we have a new WWPN for this
1488                  * initiator ID, so we pretty much need to use it.
1489                  */
1490                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1491                     " but WWPN %#jx '%s' is still at that address\n",
1492                     __func__, port->targ_port, iid, wwpn, name,
1493                     (uintmax_t)port->wwpn_iid[iid].wwpn,
1494                     port->wwpn_iid[iid].name);
1495
1496                 /*
1497                  * XXX KDM clear have_ca and ua_pending on each LUN for
1498                  * this initiator.
1499                  */
1500         }
1501 take:
1502         free(port->wwpn_iid[iid].name, M_CTL);
1503         port->wwpn_iid[iid].name = name;
1504         port->wwpn_iid[iid].wwpn = wwpn;
1505         port->wwpn_iid[iid].in_use++;
1506         mtx_unlock(&softc->ctl_lock);
1507
1508         return (iid);
1509 }
1510
1511 static int
1512 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1513 {
1514         int len;
1515
1516         switch (port->port_type) {
1517         case CTL_PORT_FC:
1518         {
1519                 struct scsi_transportid_fcp *id =
1520                     (struct scsi_transportid_fcp *)buf;
1521                 if (port->wwpn_iid[iid].wwpn == 0)
1522                         return (0);
1523                 memset(id, 0, sizeof(*id));
1524                 id->format_protocol = SCSI_PROTO_FC;
1525                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1526                 return (sizeof(*id));
1527         }
1528         case CTL_PORT_ISCSI:
1529         {
1530                 struct scsi_transportid_iscsi_port *id =
1531                     (struct scsi_transportid_iscsi_port *)buf;
1532                 if (port->wwpn_iid[iid].name == NULL)
1533                         return (0);
1534                 memset(id, 0, 256);
1535                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1536                     SCSI_PROTO_ISCSI;
1537                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1538                 len = roundup2(min(len, 252), 4);
1539                 scsi_ulto2b(len, id->additional_length);
1540                 return (sizeof(*id) + len);
1541         }
1542         case CTL_PORT_SAS:
1543         {
1544                 struct scsi_transportid_sas *id =
1545                     (struct scsi_transportid_sas *)buf;
1546                 if (port->wwpn_iid[iid].wwpn == 0)
1547                         return (0);
1548                 memset(id, 0, sizeof(*id));
1549                 id->format_protocol = SCSI_PROTO_SAS;
1550                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1551                 return (sizeof(*id));
1552         }
1553         default:
1554         {
1555                 struct scsi_transportid_spi *id =
1556                     (struct scsi_transportid_spi *)buf;
1557                 memset(id, 0, sizeof(*id));
1558                 id->format_protocol = SCSI_PROTO_SPI;
1559                 scsi_ulto2b(iid, id->scsi_addr);
1560                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1561                 return (sizeof(*id));
1562         }
1563         }
1564 }
1565
1566 static int
1567 ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1568 {
1569         return (0);
1570 }
1571
1572 static int
1573 ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1574 {
1575         return (0);
1576 }
1577
1578 /*
1579  * Data movement routine for the CTL ioctl frontend port.
1580  */
1581 static int
1582 ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1583 {
1584         struct ctl_sg_entry *ext_sglist, *kern_sglist;
1585         struct ctl_sg_entry ext_entry, kern_entry;
1586         int ext_sglen, ext_sg_entries, kern_sg_entries;
1587         int ext_sg_start, ext_offset;
1588         int len_to_copy, len_copied;
1589         int kern_watermark, ext_watermark;
1590         int ext_sglist_malloced;
1591         int i, j;
1592
1593         ext_sglist_malloced = 0;
1594         ext_sg_start = 0;
1595         ext_offset = 0;
1596
1597         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1598
1599         /*
1600          * If this flag is set, fake the data transfer.
1601          */
1602         if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1603                 ctsio->ext_data_filled = ctsio->ext_data_len;
1604                 goto bailout;
1605         }
1606
1607         /*
1608          * To simplify things here, if we have a single buffer, stick it in
1609          * a S/G entry and just make it a single entry S/G list.
1610          */
1611         if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1612                 int len_seen;
1613
1614                 ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1615
1616                 ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1617                                                            M_WAITOK);
1618                 ext_sglist_malloced = 1;
1619                 if (copyin(ctsio->ext_data_ptr, ext_sglist,
1620                                    ext_sglen) != 0) {
1621                         ctl_set_internal_failure(ctsio,
1622                                                  /*sks_valid*/ 0,
1623                                                  /*retry_count*/ 0);
1624                         goto bailout;
1625                 }
1626                 ext_sg_entries = ctsio->ext_sg_entries;
1627                 len_seen = 0;
1628                 for (i = 0; i < ext_sg_entries; i++) {
1629                         if ((len_seen + ext_sglist[i].len) >=
1630                              ctsio->ext_data_filled) {
1631                                 ext_sg_start = i;
1632                                 ext_offset = ctsio->ext_data_filled - len_seen;
1633                                 break;
1634                         }
1635                         len_seen += ext_sglist[i].len;
1636                 }
1637         } else {
1638                 ext_sglist = &ext_entry;
1639                 ext_sglist->addr = ctsio->ext_data_ptr;
1640                 ext_sglist->len = ctsio->ext_data_len;
1641                 ext_sg_entries = 1;
1642                 ext_sg_start = 0;
1643                 ext_offset = ctsio->ext_data_filled;
1644         }
1645
1646         if (ctsio->kern_sg_entries > 0) {
1647                 kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1648                 kern_sg_entries = ctsio->kern_sg_entries;
1649         } else {
1650                 kern_sglist = &kern_entry;
1651                 kern_sglist->addr = ctsio->kern_data_ptr;
1652                 kern_sglist->len = ctsio->kern_data_len;
1653                 kern_sg_entries = 1;
1654         }
1655
1656
1657         kern_watermark = 0;
1658         ext_watermark = ext_offset;
1659         len_copied = 0;
1660         for (i = ext_sg_start, j = 0;
1661              i < ext_sg_entries && j < kern_sg_entries;) {
1662                 uint8_t *ext_ptr, *kern_ptr;
1663
1664                 len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1665                                       kern_sglist[j].len - kern_watermark);
1666
1667                 ext_ptr = (uint8_t *)ext_sglist[i].addr;
1668                 ext_ptr = ext_ptr + ext_watermark;
1669                 if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1670                         /*
1671                          * XXX KDM fix this!
1672                          */
1673                         panic("need to implement bus address support");
1674 #if 0
1675                         kern_ptr = bus_to_virt(kern_sglist[j].addr);
1676 #endif
1677                 } else
1678                         kern_ptr = (uint8_t *)kern_sglist[j].addr;
1679                 kern_ptr = kern_ptr + kern_watermark;
1680
1681                 kern_watermark += len_to_copy;
1682                 ext_watermark += len_to_copy;
1683
1684                 if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1685                      CTL_FLAG_DATA_IN) {
1686                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1687                                          "bytes to user\n", len_to_copy));
1688                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1689                                          "to %p\n", kern_ptr, ext_ptr));
1690                         if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1691                                 ctl_set_internal_failure(ctsio,
1692                                                          /*sks_valid*/ 0,
1693                                                          /*retry_count*/ 0);
1694                                 goto bailout;
1695                         }
1696                 } else {
1697                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1698                                          "bytes from user\n", len_to_copy));
1699                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1700                                          "to %p\n", ext_ptr, kern_ptr));
1701                         if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1702                                 ctl_set_internal_failure(ctsio,
1703                                                          /*sks_valid*/ 0,
1704                                                          /*retry_count*/0);
1705                                 goto bailout;
1706                         }
1707                 }
1708
1709                 len_copied += len_to_copy;
1710
1711                 if (ext_sglist[i].len == ext_watermark) {
1712                         i++;
1713                         ext_watermark = 0;
1714                 }
1715
1716                 if (kern_sglist[j].len == kern_watermark) {
1717                         j++;
1718                         kern_watermark = 0;
1719                 }
1720         }
1721
1722         ctsio->ext_data_filled += len_copied;
1723
1724         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1725                          "kern_sg_entries: %d\n", ext_sg_entries,
1726                          kern_sg_entries));
1727         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1728                          "kern_data_len = %d\n", ctsio->ext_data_len,
1729                          ctsio->kern_data_len));
1730
1731
1732         /* XXX KDM set residual?? */
1733 bailout:
1734
1735         if (ext_sglist_malloced != 0)
1736                 free(ext_sglist, M_CTL);
1737
1738         return (CTL_RETVAL_COMPLETE);
1739 }
1740
1741 /*
1742  * Serialize a command that went down the "wrong" side, and so was sent to
1743  * this controller for execution.  The logic is a little different than the
1744  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1745  * sent back to the other side, but in the success case, we execute the
1746  * command on this side (XFER mode) or tell the other side to execute it
1747  * (SER_ONLY mode).
1748  */
1749 static int
1750 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1751 {
1752         struct ctl_softc *ctl_softc;
1753         union ctl_ha_msg msg_info;
1754         struct ctl_lun *lun;
1755         int retval = 0;
1756         uint32_t targ_lun;
1757
1758         ctl_softc = control_softc;
1759
1760         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1761         lun = ctl_softc->ctl_luns[targ_lun];
1762         if (lun==NULL)
1763         {
1764                 /*
1765                  * Why isn't LUN defined? The other side wouldn't
1766                  * send a cmd if the LUN is undefined.
1767                  */
1768                 printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1769
1770                 /* "Logical unit not supported" */
1771                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1772                                    lun,
1773                                    /*sense_format*/SSD_TYPE_NONE,
1774                                    /*current_error*/ 1,
1775                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1776                                    /*asc*/ 0x25,
1777                                    /*ascq*/ 0x00,
1778                                    SSD_ELEM_NONE);
1779
1780                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1781                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1782                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1783                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1784                 msg_info.hdr.serializing_sc = NULL;
1785                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1786                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1787                                 sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1788                 }
1789                 return(1);
1790
1791         }
1792
1793         mtx_lock(&lun->lun_lock);
1794         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1795
1796         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1797                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1798                  ooa_links))) {
1799         case CTL_ACTION_BLOCK:
1800                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1801                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1802                                   blocked_links);
1803                 break;
1804         case CTL_ACTION_PASS:
1805         case CTL_ACTION_SKIP:
1806                 if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1807                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1808                         ctl_enqueue_rtr((union ctl_io *)ctsio);
1809                 } else {
1810
1811                         /* send msg back to other side */
1812                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1813                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1814                         msg_info.hdr.msg_type = CTL_MSG_R2R;
1815 #if 0
1816                         printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1817 #endif
1818                         if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1819                             sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1820                         }
1821                 }
1822                 break;
1823         case CTL_ACTION_OVERLAP:
1824                 /* OVERLAPPED COMMANDS ATTEMPTED */
1825                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1826                                    lun,
1827                                    /*sense_format*/SSD_TYPE_NONE,
1828                                    /*current_error*/ 1,
1829                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1830                                    /*asc*/ 0x4E,
1831                                    /*ascq*/ 0x00,
1832                                    SSD_ELEM_NONE);
1833
1834                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1835                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1836                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1837                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1838                 msg_info.hdr.serializing_sc = NULL;
1839                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1840 #if 0
1841                 printf("BAD JUJU:Major Bummer Overlap\n");
1842 #endif
1843                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1844                 retval = 1;
1845                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1846                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1847                 }
1848                 break;
1849         case CTL_ACTION_OVERLAP_TAG:
1850                 /* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1851                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1852                                    lun,
1853                                    /*sense_format*/SSD_TYPE_NONE,
1854                                    /*current_error*/ 1,
1855                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1856                                    /*asc*/ 0x4D,
1857                                    /*ascq*/ ctsio->tag_num & 0xff,
1858                                    SSD_ELEM_NONE);
1859
1860                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1861                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1862                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1863                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1864                 msg_info.hdr.serializing_sc = NULL;
1865                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1866 #if 0
1867                 printf("BAD JUJU:Major Bummer Overlap Tag\n");
1868 #endif
1869                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1870                 retval = 1;
1871                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1872                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1873                 }
1874                 break;
1875         case CTL_ACTION_ERROR:
1876         default:
1877                 /* "Internal target failure" */
1878                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1879                                    lun,
1880                                    /*sense_format*/SSD_TYPE_NONE,
1881                                    /*current_error*/ 1,
1882                                    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1883                                    /*asc*/ 0x44,
1884                                    /*ascq*/ 0x00,
1885                                    SSD_ELEM_NONE);
1886
1887                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1888                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1889                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1890                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1891                 msg_info.hdr.serializing_sc = NULL;
1892                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1893 #if 0
1894                 printf("BAD JUJU:Major Bummer HW Error\n");
1895 #endif
1896                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1897                 retval = 1;
1898                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1899                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1900                 }
1901                 break;
1902         }
1903         mtx_unlock(&lun->lun_lock);
1904         return (retval);
1905 }
1906
1907 static int
1908 ctl_ioctl_submit_wait(union ctl_io *io)
1909 {
1910         struct ctl_fe_ioctl_params params;
1911         ctl_fe_ioctl_state last_state;
1912         int done, retval;
1913
1914         retval = 0;
1915
1916         bzero(&params, sizeof(params));
1917
1918         mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1919         cv_init(&params.sem, "ctlioccv");
1920         params.state = CTL_IOCTL_INPROG;
1921         last_state = params.state;
1922
1923         io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1924
1925         CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1926
1927         /* This shouldn't happen */
1928         if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1929                 return (retval);
1930
1931         done = 0;
1932
1933         do {
1934                 mtx_lock(&params.ioctl_mtx);
1935                 /*
1936                  * Check the state here, and don't sleep if the state has
1937                  * already changed (i.e. wakeup has already occured, but we
1938                  * weren't waiting yet).
1939                  */
1940                 if (params.state == last_state) {
1941                         /* XXX KDM cv_wait_sig instead? */
1942                         cv_wait(&params.sem, &params.ioctl_mtx);
1943                 }
1944                 last_state = params.state;
1945
1946                 switch (params.state) {
1947                 case CTL_IOCTL_INPROG:
1948                         /* Why did we wake up? */
1949                         /* XXX KDM error here? */
1950                         mtx_unlock(&params.ioctl_mtx);
1951                         break;
1952                 case CTL_IOCTL_DATAMOVE:
1953                         CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1954
1955                         /*
1956                          * change last_state back to INPROG to avoid
1957                          * deadlock on subsequent data moves.
1958                          */
1959                         params.state = last_state = CTL_IOCTL_INPROG;
1960
1961                         mtx_unlock(&params.ioctl_mtx);
1962                         ctl_ioctl_do_datamove(&io->scsiio);
1963                         /*
1964                          * Note that in some cases, most notably writes,
1965                          * this will queue the I/O and call us back later.
1966                          * In other cases, generally reads, this routine
1967                          * will immediately call back and wake us up,
1968                          * probably using our own context.
1969                          */
1970                         io->scsiio.be_move_done(io);
1971                         break;
1972                 case CTL_IOCTL_DONE:
1973                         mtx_unlock(&params.ioctl_mtx);
1974                         CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1975                         done = 1;
1976                         break;
1977                 default:
1978                         mtx_unlock(&params.ioctl_mtx);
1979                         /* XXX KDM error here? */
1980                         break;
1981                 }
1982         } while (done == 0);
1983
1984         mtx_destroy(&params.ioctl_mtx);
1985         cv_destroy(&params.sem);
1986
1987         return (CTL_RETVAL_COMPLETE);
1988 }
1989
1990 static void
1991 ctl_ioctl_datamove(union ctl_io *io)
1992 {
1993         struct ctl_fe_ioctl_params *params;
1994
1995         params = (struct ctl_fe_ioctl_params *)
1996                 io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1997
1998         mtx_lock(&params->ioctl_mtx);
1999         params->state = CTL_IOCTL_DATAMOVE;
2000         cv_broadcast(&params->sem);
2001         mtx_unlock(&params->ioctl_mtx);
2002 }
2003
2004 static void
2005 ctl_ioctl_done(union ctl_io *io)
2006 {
2007         struct ctl_fe_ioctl_params *params;
2008
2009         params = (struct ctl_fe_ioctl_params *)
2010                 io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2011
2012         mtx_lock(&params->ioctl_mtx);
2013         params->state = CTL_IOCTL_DONE;
2014         cv_broadcast(&params->sem);
2015         mtx_unlock(&params->ioctl_mtx);
2016 }
2017
2018 static void
2019 ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2020 {
2021         struct ctl_fe_ioctl_startstop_info *sd_info;
2022
2023         sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2024
2025         sd_info->hs_info.status = metatask->status;
2026         sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2027         sd_info->hs_info.luns_complete =
2028                 metatask->taskinfo.startstop.luns_complete;
2029         sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2030
2031         cv_broadcast(&sd_info->sem);
2032 }
2033
2034 static void
2035 ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2036 {
2037         struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2038
2039         fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2040
2041         mtx_lock(fe_bbr_info->lock);
2042         fe_bbr_info->bbr_info->status = metatask->status;
2043         fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2044         fe_bbr_info->wakeup_done = 1;
2045         mtx_unlock(fe_bbr_info->lock);
2046
2047         cv_broadcast(&fe_bbr_info->sem);
2048 }
2049
2050 /*
2051  * Returns 0 for success, errno for failure.
2052  */
2053 static int
2054 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2055                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2056 {
2057         union ctl_io *io;
2058         int retval;
2059
2060         retval = 0;
2061
2062         mtx_lock(&lun->lun_lock);
2063         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2064              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2065              ooa_links)) {
2066                 struct ctl_ooa_entry *entry;
2067
2068                 /*
2069                  * If we've got more than we can fit, just count the
2070                  * remaining entries.
2071                  */
2072                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2073                         continue;
2074
2075                 entry = &kern_entries[*cur_fill_num];
2076
2077                 entry->tag_num = io->scsiio.tag_num;
2078                 entry->lun_num = lun->lun;
2079 #ifdef CTL_TIME_IO
2080                 entry->start_bt = io->io_hdr.start_bt;
2081 #endif
2082                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2083                 entry->cdb_len = io->scsiio.cdb_len;
2084                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2085                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2086
2087                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2088                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2089
2090                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2091                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2092
2093                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2094                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2095
2096                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2097                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2098         }
2099         mtx_unlock(&lun->lun_lock);
2100
2101         return (retval);
2102 }
2103
2104 static void *
2105 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2106                  size_t error_str_len)
2107 {
2108         void *kptr;
2109
2110         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2111
2112         if (copyin(user_addr, kptr, len) != 0) {
2113                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2114                          "from user address %p to kernel address %p", len,
2115                          user_addr, kptr);
2116                 free(kptr, M_CTL);
2117                 return (NULL);
2118         }
2119
2120         return (kptr);
2121 }
2122
2123 static void
2124 ctl_free_args(int num_args, struct ctl_be_arg *args)
2125 {
2126         int i;
2127
2128         if (args == NULL)
2129                 return;
2130
2131         for (i = 0; i < num_args; i++) {
2132                 free(args[i].kname, M_CTL);
2133                 free(args[i].kvalue, M_CTL);
2134         }
2135
2136         free(args, M_CTL);
2137 }
2138
2139 static struct ctl_be_arg *
2140 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2141                 char *error_str, size_t error_str_len)
2142 {
2143         struct ctl_be_arg *args;
2144         int i;
2145
2146         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2147                                 error_str, error_str_len);
2148
2149         if (args == NULL)
2150                 goto bailout;
2151
2152         for (i = 0; i < num_args; i++) {
2153                 args[i].kname = NULL;
2154                 args[i].kvalue = NULL;
2155         }
2156
2157         for (i = 0; i < num_args; i++) {
2158                 uint8_t *tmpptr;
2159
2160                 args[i].kname = ctl_copyin_alloc(args[i].name,
2161                         args[i].namelen, error_str, error_str_len);
2162                 if (args[i].kname == NULL)
2163                         goto bailout;
2164
2165                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2166                         snprintf(error_str, error_str_len, "Argument %d "
2167                                  "name is not NUL-terminated", i);
2168                         goto bailout;
2169                 }
2170
2171                 if (args[i].flags & CTL_BEARG_RD) {
2172                         tmpptr = ctl_copyin_alloc(args[i].value,
2173                                 args[i].vallen, error_str, error_str_len);
2174                         if (tmpptr == NULL)
2175                                 goto bailout;
2176                         if ((args[i].flags & CTL_BEARG_ASCII)
2177                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2178                                 snprintf(error_str, error_str_len, "Argument "
2179                                     "%d value is not NUL-terminated", i);
2180                                 goto bailout;
2181                         }
2182                         args[i].kvalue = tmpptr;
2183                 } else {
2184                         args[i].kvalue = malloc(args[i].vallen,
2185                             M_CTL, M_WAITOK | M_ZERO);
2186                 }
2187         }
2188
2189         return (args);
2190 bailout:
2191
2192         ctl_free_args(num_args, args);
2193
2194         return (NULL);
2195 }
2196
2197 static void
2198 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2199 {
2200         int i;
2201
2202         for (i = 0; i < num_args; i++) {
2203                 if (args[i].flags & CTL_BEARG_WR)
2204                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2205         }
2206 }
2207
2208 /*
2209  * Escape characters that are illegal or not recommended in XML.
2210  */
2211 int
2212 ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2213 {
2214         int retval;
2215
2216         retval = 0;
2217
2218         for (; *str; str++) {
2219                 switch (*str) {
2220                 case '&':
2221                         retval = sbuf_printf(sb, "&amp;");
2222                         break;
2223                 case '>':
2224                         retval = sbuf_printf(sb, "&gt;");
2225                         break;
2226                 case '<':
2227                         retval = sbuf_printf(sb, "&lt;");
2228                         break;
2229                 default:
2230                         retval = sbuf_putc(sb, *str);
2231                         break;
2232                 }
2233
2234                 if (retval != 0)
2235                         break;
2236
2237         }
2238
2239         return (retval);
2240 }
2241
2242 static int
2243 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2244           struct thread *td)
2245 {
2246         struct ctl_softc *softc;
2247         int retval;
2248
2249         softc = control_softc;
2250
2251         retval = 0;
2252
2253         switch (cmd) {
2254         case CTL_IO: {
2255                 union ctl_io *io;
2256                 void *pool_tmp;
2257
2258                 /*
2259                  * If we haven't been "enabled", don't allow any SCSI I/O
2260                  * to this FETD.
2261                  */
2262                 if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2263                         retval = EPERM;
2264                         break;
2265                 }
2266
2267                 io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2268                 if (io == NULL) {
2269                         printf("ctl_ioctl: can't allocate ctl_io!\n");
2270                         retval = ENOSPC;
2271                         break;
2272                 }
2273
2274                 /*
2275                  * Need to save the pool reference so it doesn't get
2276                  * spammed by the user's ctl_io.
2277                  */
2278                 pool_tmp = io->io_hdr.pool;
2279
2280                 memcpy(io, (void *)addr, sizeof(*io));
2281
2282                 io->io_hdr.pool = pool_tmp;
2283                 /*
2284                  * No status yet, so make sure the status is set properly.
2285                  */
2286                 io->io_hdr.status = CTL_STATUS_NONE;
2287
2288                 /*
2289                  * The user sets the initiator ID, target and LUN IDs.
2290                  */
2291                 io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2292                 io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2293                 if ((io->io_hdr.io_type == CTL_IO_SCSI)
2294                  && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2295                         io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2296
2297                 retval = ctl_ioctl_submit_wait(io);
2298
2299                 if (retval != 0) {
2300                         ctl_free_io(io);
2301                         break;
2302                 }
2303
2304                 memcpy((void *)addr, io, sizeof(*io));
2305
2306                 /* return this to our pool */
2307                 ctl_free_io(io);
2308
2309                 break;
2310         }
2311         case CTL_ENABLE_PORT:
2312         case CTL_DISABLE_PORT:
2313         case CTL_SET_PORT_WWNS: {
2314                 struct ctl_port *port;
2315                 struct ctl_port_entry *entry;
2316
2317                 entry = (struct ctl_port_entry *)addr;
2318
2319                 mtx_lock(&softc->ctl_lock);
2320                 STAILQ_FOREACH(port, &softc->port_list, links) {
2321                         int action, done;
2322
2323                         action = 0;
2324                         done = 0;
2325
2326                         if ((entry->port_type == CTL_PORT_NONE)
2327                          && (entry->targ_port == port->targ_port)) {
2328                                 /*
2329                                  * If the user only wants to enable or
2330                                  * disable or set WWNs on a specific port,
2331                                  * do the operation and we're done.
2332                                  */
2333                                 action = 1;
2334                                 done = 1;
2335                         } else if (entry->port_type & port->port_type) {
2336                                 /*
2337                                  * Compare the user's type mask with the
2338                                  * particular frontend type to see if we
2339                                  * have a match.
2340                                  */
2341                                 action = 1;
2342                                 done = 0;
2343
2344                                 /*
2345                                  * Make sure the user isn't trying to set
2346                                  * WWNs on multiple ports at the same time.
2347                                  */
2348                                 if (cmd == CTL_SET_PORT_WWNS) {
2349                                         printf("%s: Can't set WWNs on "
2350                                                "multiple ports\n", __func__);
2351                                         retval = EINVAL;
2352                                         break;
2353                                 }
2354                         }
2355                         if (action != 0) {
2356                                 /*
2357                                  * XXX KDM we have to drop the lock here,
2358                                  * because the online/offline operations
2359                                  * can potentially block.  We need to
2360                                  * reference count the frontends so they
2361                                  * can't go away,
2362                                  */
2363                                 mtx_unlock(&softc->ctl_lock);
2364
2365                                 if (cmd == CTL_ENABLE_PORT) {
2366                                         struct ctl_lun *lun;
2367
2368                                         STAILQ_FOREACH(lun, &softc->lun_list,
2369                                                        links) {
2370                                                 port->lun_enable(port->targ_lun_arg,
2371                                                     lun->target,
2372                                                     lun->lun);
2373                                         }
2374
2375                                         ctl_port_online(port);
2376                                 } else if (cmd == CTL_DISABLE_PORT) {
2377                                         struct ctl_lun *lun;
2378
2379                                         ctl_port_offline(port);
2380
2381                                         STAILQ_FOREACH(lun, &softc->lun_list,
2382                                                        links) {
2383                                                 port->lun_disable(
2384                                                     port->targ_lun_arg,
2385                                                     lun->target,
2386                                                     lun->lun);
2387                                         }
2388                                 }
2389
2390                                 mtx_lock(&softc->ctl_lock);
2391
2392                                 if (cmd == CTL_SET_PORT_WWNS)
2393                                         ctl_port_set_wwns(port,
2394                                             (entry->flags & CTL_PORT_WWNN_VALID) ?
2395                                             1 : 0, entry->wwnn,
2396                                             (entry->flags & CTL_PORT_WWPN_VALID) ?
2397                                             1 : 0, entry->wwpn);
2398                         }
2399                         if (done != 0)
2400                                 break;
2401                 }
2402                 mtx_unlock(&softc->ctl_lock);
2403                 break;
2404         }
2405         case CTL_GET_PORT_LIST: {
2406                 struct ctl_port *port;
2407                 struct ctl_port_list *list;
2408                 int i;
2409
2410                 list = (struct ctl_port_list *)addr;
2411
2412                 if (list->alloc_len != (list->alloc_num *
2413                     sizeof(struct ctl_port_entry))) {
2414                         printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2415                                "alloc_num %u * sizeof(struct ctl_port_entry) "
2416                                "%zu\n", __func__, list->alloc_len,
2417                                list->alloc_num, sizeof(struct ctl_port_entry));
2418                         retval = EINVAL;
2419                         break;
2420                 }
2421                 list->fill_len = 0;
2422                 list->fill_num = 0;
2423                 list->dropped_num = 0;
2424                 i = 0;
2425                 mtx_lock(&softc->ctl_lock);
2426                 STAILQ_FOREACH(port, &softc->port_list, links) {
2427                         struct ctl_port_entry entry, *list_entry;
2428
2429                         if (list->fill_num >= list->alloc_num) {
2430                                 list->dropped_num++;
2431                                 continue;
2432                         }
2433
2434                         entry.port_type = port->port_type;
2435                         strlcpy(entry.port_name, port->port_name,
2436                                 sizeof(entry.port_name));
2437                         entry.targ_port = port->targ_port;
2438                         entry.physical_port = port->physical_port;
2439                         entry.virtual_port = port->virtual_port;
2440                         entry.wwnn = port->wwnn;
2441                         entry.wwpn = port->wwpn;
2442                         if (port->status & CTL_PORT_STATUS_ONLINE)
2443                                 entry.online = 1;
2444                         else
2445                                 entry.online = 0;
2446
2447                         list_entry = &list->entries[i];
2448
2449                         retval = copyout(&entry, list_entry, sizeof(entry));
2450                         if (retval != 0) {
2451                                 printf("%s: CTL_GET_PORT_LIST: copyout "
2452                                        "returned %d\n", __func__, retval);
2453                                 break;
2454                         }
2455                         i++;
2456                         list->fill_num++;
2457                         list->fill_len += sizeof(entry);
2458                 }
2459                 mtx_unlock(&softc->ctl_lock);
2460
2461                 /*
2462                  * If this is non-zero, we had a copyout fault, so there's
2463                  * probably no point in attempting to set the status inside
2464                  * the structure.
2465                  */
2466                 if (retval != 0)
2467                         break;
2468
2469                 if (list->dropped_num > 0)
2470                         list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2471                 else
2472                         list->status = CTL_PORT_LIST_OK;
2473                 break;
2474         }
2475         case CTL_DUMP_OOA: {
2476                 struct ctl_lun *lun;
2477                 union ctl_io *io;
2478                 char printbuf[128];
2479                 struct sbuf sb;
2480
2481                 mtx_lock(&softc->ctl_lock);
2482                 printf("Dumping OOA queues:\n");
2483                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2484                         mtx_lock(&lun->lun_lock);
2485                         for (io = (union ctl_io *)TAILQ_FIRST(
2486                              &lun->ooa_queue); io != NULL;
2487                              io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2488                              ooa_links)) {
2489                                 sbuf_new(&sb, printbuf, sizeof(printbuf),
2490                                          SBUF_FIXEDLEN);
2491                                 sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2492                                             (intmax_t)lun->lun,
2493                                             io->scsiio.tag_num,
2494                                             (io->io_hdr.flags &
2495                                             CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2496                                             (io->io_hdr.flags &
2497                                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2498                                             (io->io_hdr.flags &
2499                                             CTL_FLAG_ABORT) ? " ABORT" : "",
2500                                             (io->io_hdr.flags &
2501                                         CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2502                                 ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2503                                 sbuf_finish(&sb);
2504                                 printf("%s\n", sbuf_data(&sb));
2505                         }
2506                         mtx_unlock(&lun->lun_lock);
2507                 }
2508                 printf("OOA queues dump done\n");
2509                 mtx_unlock(&softc->ctl_lock);
2510                 break;
2511         }
2512         case CTL_GET_OOA: {
2513                 struct ctl_lun *lun;
2514                 struct ctl_ooa *ooa_hdr;
2515                 struct ctl_ooa_entry *entries;
2516                 uint32_t cur_fill_num;
2517
2518                 ooa_hdr = (struct ctl_ooa *)addr;
2519
2520                 if ((ooa_hdr->alloc_len == 0)
2521                  || (ooa_hdr->alloc_num == 0)) {
2522                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2523                                "must be non-zero\n", __func__,
2524                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2525                         retval = EINVAL;
2526                         break;
2527                 }
2528
2529                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2530                     sizeof(struct ctl_ooa_entry))) {
2531                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2532                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2533                                __func__, ooa_hdr->alloc_len,
2534                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2535                         retval = EINVAL;
2536                         break;
2537                 }
2538
2539                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2540                 if (entries == NULL) {
2541                         printf("%s: could not allocate %d bytes for OOA "
2542                                "dump\n", __func__, ooa_hdr->alloc_len);
2543                         retval = ENOMEM;
2544                         break;
2545                 }
2546
2547                 mtx_lock(&softc->ctl_lock);
2548                 if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2549                  && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2550                   || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2551                         mtx_unlock(&softc->ctl_lock);
2552                         free(entries, M_CTL);
2553                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2554                                __func__, (uintmax_t)ooa_hdr->lun_num);
2555                         retval = EINVAL;
2556                         break;
2557                 }
2558
2559                 cur_fill_num = 0;
2560
2561                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2562                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2563                                 retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2564                                         ooa_hdr, entries);
2565                                 if (retval != 0)
2566                                         break;
2567                         }
2568                         if (retval != 0) {
2569                                 mtx_unlock(&softc->ctl_lock);
2570                                 free(entries, M_CTL);
2571                                 break;
2572                         }
2573                 } else {
2574                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2575
2576                         retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2577                                                     entries);
2578                 }
2579                 mtx_unlock(&softc->ctl_lock);
2580
2581                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2582                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2583                         sizeof(struct ctl_ooa_entry);
2584                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2585                 if (retval != 0) {
2586                         printf("%s: error copying out %d bytes for OOA dump\n",
2587                                __func__, ooa_hdr->fill_len);
2588                 }
2589
2590                 getbintime(&ooa_hdr->cur_bt);
2591
2592                 if (cur_fill_num > ooa_hdr->alloc_num) {
2593                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2594                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2595                 } else {
2596                         ooa_hdr->dropped_num = 0;
2597                         ooa_hdr->status = CTL_OOA_OK;
2598                 }
2599
2600                 free(entries, M_CTL);
2601                 break;
2602         }
2603         case CTL_CHECK_OOA: {
2604                 union ctl_io *io;
2605                 struct ctl_lun *lun;
2606                 struct ctl_ooa_info *ooa_info;
2607
2608
2609                 ooa_info = (struct ctl_ooa_info *)addr;
2610
2611                 if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2612                         ooa_info->status = CTL_OOA_INVALID_LUN;
2613                         break;
2614                 }
2615                 mtx_lock(&softc->ctl_lock);
2616                 lun = softc->ctl_luns[ooa_info->lun_id];
2617                 if (lun == NULL) {
2618                         mtx_unlock(&softc->ctl_lock);
2619                         ooa_info->status = CTL_OOA_INVALID_LUN;
2620                         break;
2621                 }
2622                 mtx_lock(&lun->lun_lock);
2623                 mtx_unlock(&softc->ctl_lock);
2624                 ooa_info->num_entries = 0;
2625                 for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2626                      io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2627                      &io->io_hdr, ooa_links)) {
2628                         ooa_info->num_entries++;
2629                 }
2630                 mtx_unlock(&lun->lun_lock);
2631
2632                 ooa_info->status = CTL_OOA_SUCCESS;
2633
2634                 break;
2635         }
2636         case CTL_HARD_START:
2637         case CTL_HARD_STOP: {
2638                 struct ctl_fe_ioctl_startstop_info ss_info;
2639                 struct cfi_metatask *metatask;
2640                 struct mtx hs_mtx;
2641
2642                 mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2643
2644                 cv_init(&ss_info.sem, "hard start/stop cv" );
2645
2646                 metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2647                 if (metatask == NULL) {
2648                         retval = ENOMEM;
2649                         mtx_destroy(&hs_mtx);
2650                         break;
2651                 }
2652
2653                 if (cmd == CTL_HARD_START)
2654                         metatask->tasktype = CFI_TASK_STARTUP;
2655                 else
2656                         metatask->tasktype = CFI_TASK_SHUTDOWN;
2657
2658                 metatask->callback = ctl_ioctl_hard_startstop_callback;
2659                 metatask->callback_arg = &ss_info;
2660
2661                 cfi_action(metatask);
2662
2663                 /* Wait for the callback */
2664                 mtx_lock(&hs_mtx);
2665                 cv_wait_sig(&ss_info.sem, &hs_mtx);
2666                 mtx_unlock(&hs_mtx);
2667
2668                 /*
2669                  * All information has been copied from the metatask by the
2670                  * time cv_broadcast() is called, so we free the metatask here.
2671                  */
2672                 cfi_free_metatask(metatask);
2673
2674                 memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2675
2676                 mtx_destroy(&hs_mtx);
2677                 break;
2678         }
2679         case CTL_BBRREAD: {
2680                 struct ctl_bbrread_info *bbr_info;
2681                 struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2682                 struct mtx bbr_mtx;
2683                 struct cfi_metatask *metatask;
2684
2685                 bbr_info = (struct ctl_bbrread_info *)addr;
2686
2687                 bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2688
2689                 bzero(&bbr_mtx, sizeof(bbr_mtx));
2690                 mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2691
2692                 fe_bbr_info.bbr_info = bbr_info;
2693                 fe_bbr_info.lock = &bbr_mtx;
2694
2695                 cv_init(&fe_bbr_info.sem, "BBR read cv");
2696                 metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2697
2698                 if (metatask == NULL) {
2699                         mtx_destroy(&bbr_mtx);
2700                         cv_destroy(&fe_bbr_info.sem);
2701                         retval = ENOMEM;
2702                         break;
2703                 }
2704                 metatask->tasktype = CFI_TASK_BBRREAD;
2705                 metatask->callback = ctl_ioctl_bbrread_callback;
2706                 metatask->callback_arg = &fe_bbr_info;
2707                 metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2708                 metatask->taskinfo.bbrread.lba = bbr_info->lba;
2709                 metatask->taskinfo.bbrread.len = bbr_info->len;
2710
2711                 cfi_action(metatask);
2712
2713                 mtx_lock(&bbr_mtx);
2714                 while (fe_bbr_info.wakeup_done == 0)
2715                         cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2716                 mtx_unlock(&bbr_mtx);
2717
2718                 bbr_info->status = metatask->status;
2719                 bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2720                 bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2721                 memcpy(&bbr_info->sense_data,
2722                        &metatask->taskinfo.bbrread.sense_data,
2723                        ctl_min(sizeof(bbr_info->sense_data),
2724                                sizeof(metatask->taskinfo.bbrread.sense_data)));
2725
2726                 cfi_free_metatask(metatask);
2727
2728                 mtx_destroy(&bbr_mtx);
2729                 cv_destroy(&fe_bbr_info.sem);
2730
2731                 break;
2732         }
2733         case CTL_DELAY_IO: {
2734                 struct ctl_io_delay_info *delay_info;
2735 #ifdef CTL_IO_DELAY
2736                 struct ctl_lun *lun;
2737 #endif /* CTL_IO_DELAY */
2738
2739                 delay_info = (struct ctl_io_delay_info *)addr;
2740
2741 #ifdef CTL_IO_DELAY
2742                 mtx_lock(&softc->ctl_lock);
2743
2744                 if ((delay_info->lun_id >= CTL_MAX_LUNS)
2745                  || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2746                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2747                 } else {
2748                         lun = softc->ctl_luns[delay_info->lun_id];
2749                         mtx_lock(&lun->lun_lock);
2750
2751                         delay_info->status = CTL_DELAY_STATUS_OK;
2752
2753                         switch (delay_info->delay_type) {
2754                         case CTL_DELAY_TYPE_CONT:
2755                                 break;
2756                         case CTL_DELAY_TYPE_ONESHOT:
2757                                 break;
2758                         default:
2759                                 delay_info->status =
2760                                         CTL_DELAY_STATUS_INVALID_TYPE;
2761                                 break;
2762                         }
2763
2764                         switch (delay_info->delay_loc) {
2765                         case CTL_DELAY_LOC_DATAMOVE:
2766                                 lun->delay_info.datamove_type =
2767                                         delay_info->delay_type;
2768                                 lun->delay_info.datamove_delay =
2769                                         delay_info->delay_secs;
2770                                 break;
2771                         case CTL_DELAY_LOC_DONE:
2772                                 lun->delay_info.done_type =
2773                                         delay_info->delay_type;
2774                                 lun->delay_info.done_delay =
2775                                         delay_info->delay_secs;
2776                                 break;
2777                         default:
2778                                 delay_info->status =
2779                                         CTL_DELAY_STATUS_INVALID_LOC;
2780                                 break;
2781                         }
2782                         mtx_unlock(&lun->lun_lock);
2783                 }
2784
2785                 mtx_unlock(&softc->ctl_lock);
2786 #else
2787                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2788 #endif /* CTL_IO_DELAY */
2789                 break;
2790         }
2791         case CTL_REALSYNC_SET: {
2792                 int *syncstate;
2793
2794                 syncstate = (int *)addr;
2795
2796                 mtx_lock(&softc->ctl_lock);
2797                 switch (*syncstate) {
2798                 case 0:
2799                         softc->flags &= ~CTL_FLAG_REAL_SYNC;
2800                         break;
2801                 case 1:
2802                         softc->flags |= CTL_FLAG_REAL_SYNC;
2803                         break;
2804                 default:
2805                         retval = EINVAL;
2806                         break;
2807                 }
2808                 mtx_unlock(&softc->ctl_lock);
2809                 break;
2810         }
2811         case CTL_REALSYNC_GET: {
2812                 int *syncstate;
2813
2814                 syncstate = (int*)addr;
2815
2816                 mtx_lock(&softc->ctl_lock);
2817                 if (softc->flags & CTL_FLAG_REAL_SYNC)
2818                         *syncstate = 1;
2819                 else
2820                         *syncstate = 0;
2821                 mtx_unlock(&softc->ctl_lock);
2822
2823                 break;
2824         }
2825         case CTL_SETSYNC:
2826         case CTL_GETSYNC: {
2827                 struct ctl_sync_info *sync_info;
2828                 struct ctl_lun *lun;
2829
2830                 sync_info = (struct ctl_sync_info *)addr;
2831
2832                 mtx_lock(&softc->ctl_lock);
2833                 lun = softc->ctl_luns[sync_info->lun_id];
2834                 if (lun == NULL) {
2835                         mtx_unlock(&softc->ctl_lock);
2836                         sync_info->status = CTL_GS_SYNC_NO_LUN;
2837                 }
2838                 /*
2839                  * Get or set the sync interval.  We're not bounds checking
2840                  * in the set case, hopefully the user won't do something
2841                  * silly.
2842                  */
2843                 mtx_lock(&lun->lun_lock);
2844                 mtx_unlock(&softc->ctl_lock);
2845                 if (cmd == CTL_GETSYNC)
2846                         sync_info->sync_interval = lun->sync_interval;
2847                 else
2848                         lun->sync_interval = sync_info->sync_interval;
2849                 mtx_unlock(&lun->lun_lock);
2850
2851                 sync_info->status = CTL_GS_SYNC_OK;
2852
2853                 break;
2854         }
2855         case CTL_GETSTATS: {
2856                 struct ctl_stats *stats;
2857                 struct ctl_lun *lun;
2858                 int i;
2859
2860                 stats = (struct ctl_stats *)addr;
2861
2862                 if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2863                      stats->alloc_len) {
2864                         stats->status = CTL_SS_NEED_MORE_SPACE;
2865                         stats->num_luns = softc->num_luns;
2866                         break;
2867                 }
2868                 /*
2869                  * XXX KDM no locking here.  If the LUN list changes,
2870                  * things can blow up.
2871                  */
2872                 for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2873                      i++, lun = STAILQ_NEXT(lun, links)) {
2874                         retval = copyout(&lun->stats, &stats->lun_stats[i],
2875                                          sizeof(lun->stats));
2876                         if (retval != 0)
2877                                 break;
2878                 }
2879                 stats->num_luns = softc->num_luns;
2880                 stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2881                                  softc->num_luns;
2882                 stats->status = CTL_SS_OK;
2883 #ifdef CTL_TIME_IO
2884                 stats->flags = CTL_STATS_FLAG_TIME_VALID;
2885 #else
2886                 stats->flags = CTL_STATS_FLAG_NONE;
2887 #endif
2888                 getnanouptime(&stats->timestamp);
2889                 break;
2890         }
2891         case CTL_ERROR_INJECT: {
2892                 struct ctl_error_desc *err_desc, *new_err_desc;
2893                 struct ctl_lun *lun;
2894
2895                 err_desc = (struct ctl_error_desc *)addr;
2896
2897                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2898                                       M_WAITOK | M_ZERO);
2899                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2900
2901                 mtx_lock(&softc->ctl_lock);
2902                 lun = softc->ctl_luns[err_desc->lun_id];
2903                 if (lun == NULL) {
2904                         mtx_unlock(&softc->ctl_lock);
2905                         free(new_err_desc, M_CTL);
2906                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2907                                __func__, (uintmax_t)err_desc->lun_id);
2908                         retval = EINVAL;
2909                         break;
2910                 }
2911                 mtx_lock(&lun->lun_lock);
2912                 mtx_unlock(&softc->ctl_lock);
2913
2914                 /*
2915                  * We could do some checking here to verify the validity
2916                  * of the request, but given the complexity of error
2917                  * injection requests, the checking logic would be fairly
2918                  * complex.
2919                  *
2920                  * For now, if the request is invalid, it just won't get
2921                  * executed and might get deleted.
2922                  */
2923                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2924
2925                 /*
2926                  * XXX KDM check to make sure the serial number is unique,
2927                  * in case we somehow manage to wrap.  That shouldn't
2928                  * happen for a very long time, but it's the right thing to
2929                  * do.
2930                  */
2931                 new_err_desc->serial = lun->error_serial;
2932                 err_desc->serial = lun->error_serial;
2933                 lun->error_serial++;
2934
2935                 mtx_unlock(&lun->lun_lock);
2936                 break;
2937         }
2938         case CTL_ERROR_INJECT_DELETE: {
2939                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2940                 struct ctl_lun *lun;
2941                 int delete_done;
2942
2943                 delete_desc = (struct ctl_error_desc *)addr;
2944                 delete_done = 0;
2945
2946                 mtx_lock(&softc->ctl_lock);
2947                 lun = softc->ctl_luns[delete_desc->lun_id];
2948                 if (lun == NULL) {
2949                         mtx_unlock(&softc->ctl_lock);
2950                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2951                                __func__, (uintmax_t)delete_desc->lun_id);
2952                         retval = EINVAL;
2953                         break;
2954                 }
2955                 mtx_lock(&lun->lun_lock);
2956                 mtx_unlock(&softc->ctl_lock);
2957                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2958                         if (desc->serial != delete_desc->serial)
2959                                 continue;
2960
2961                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2962                                       links);
2963                         free(desc, M_CTL);
2964                         delete_done = 1;
2965                 }
2966                 mtx_unlock(&lun->lun_lock);
2967                 if (delete_done == 0) {
2968                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2969                                "error serial %ju on LUN %u\n", __func__,
2970                                delete_desc->serial, delete_desc->lun_id);
2971                         retval = EINVAL;
2972                         break;
2973                 }
2974                 break;
2975         }
2976         case CTL_DUMP_STRUCTS: {
2977                 int i, j, k, idx;
2978                 struct ctl_port *port;
2979                 struct ctl_frontend *fe;
2980
2981                 mtx_lock(&softc->ctl_lock);
2982                 printf("CTL Persistent Reservation information start:\n");
2983                 for (i = 0; i < CTL_MAX_LUNS; i++) {
2984                         struct ctl_lun *lun;
2985
2986                         lun = softc->ctl_luns[i];
2987
2988                         if ((lun == NULL)
2989                          || ((lun->flags & CTL_LUN_DISABLED) != 0))
2990                                 continue;
2991
2992                         for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2993                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2994                                         idx = j * CTL_MAX_INIT_PER_PORT + k;
2995                                         if (lun->per_res[idx].registered == 0)
2996                                                 continue;
2997                                         printf("  LUN %d port %d iid %d key "
2998                                                "%#jx\n", i, j, k,
2999                                                (uintmax_t)scsi_8btou64(
3000                                                lun->per_res[idx].res_key.key));
3001                                 }
3002                         }
3003                 }
3004                 printf("CTL Persistent Reservation information end\n");
3005                 printf("CTL Ports:\n");
3006                 STAILQ_FOREACH(port, &softc->port_list, links) {
3007                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3008                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3009                                port->frontend->name, port->port_type,
3010                                port->physical_port, port->virtual_port,
3011                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3012                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3013                                 if (port->wwpn_iid[j].in_use == 0 &&
3014                                     port->wwpn_iid[j].wwpn == 0 &&
3015                                     port->wwpn_iid[j].name == NULL)
3016                                         continue;
3017
3018                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
3019                                     j, port->wwpn_iid[j].in_use,
3020                                     (uintmax_t)port->wwpn_iid[j].wwpn,
3021                                     port->wwpn_iid[j].name);
3022                         }
3023                 }
3024                 printf("CTL Port information end\n");
3025                 mtx_unlock(&softc->ctl_lock);
3026                 /*
3027                  * XXX KDM calling this without a lock.  We'd likely want
3028                  * to drop the lock before calling the frontend's dump
3029                  * routine anyway.
3030                  */
3031                 printf("CTL Frontends:\n");
3032                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
3033                         printf("  Frontend '%s'\n", fe->name);
3034                         if (fe->fe_dump != NULL)
3035                                 fe->fe_dump();
3036                 }
3037                 printf("CTL Frontend information end\n");
3038                 break;
3039         }
3040         case CTL_LUN_REQ: {
3041                 struct ctl_lun_req *lun_req;
3042                 struct ctl_backend_driver *backend;
3043
3044                 lun_req = (struct ctl_lun_req *)addr;
3045
3046                 backend = ctl_backend_find(lun_req->backend);
3047                 if (backend == NULL) {
3048                         lun_req->status = CTL_LUN_ERROR;
3049                         snprintf(lun_req->error_str,
3050                                  sizeof(lun_req->error_str),
3051                                  "Backend \"%s\" not found.",
3052                                  lun_req->backend);
3053                         break;
3054                 }
3055                 if (lun_req->num_be_args > 0) {
3056                         lun_req->kern_be_args = ctl_copyin_args(
3057                                 lun_req->num_be_args,
3058                                 lun_req->be_args,
3059                                 lun_req->error_str,
3060                                 sizeof(lun_req->error_str));
3061                         if (lun_req->kern_be_args == NULL) {
3062                                 lun_req->status = CTL_LUN_ERROR;
3063                                 break;
3064                         }
3065                 }
3066
3067                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3068
3069                 if (lun_req->num_be_args > 0) {
3070                         ctl_copyout_args(lun_req->num_be_args,
3071                                       lun_req->kern_be_args);
3072                         ctl_free_args(lun_req->num_be_args,
3073                                       lun_req->kern_be_args);
3074                 }
3075                 break;
3076         }
3077         case CTL_LUN_LIST: {
3078                 struct sbuf *sb;
3079                 struct ctl_lun *lun;
3080                 struct ctl_lun_list *list;
3081                 struct ctl_option *opt;
3082
3083                 list = (struct ctl_lun_list *)addr;
3084
3085                 /*
3086                  * Allocate a fixed length sbuf here, based on the length
3087                  * of the user's buffer.  We could allocate an auto-extending
3088                  * buffer, and then tell the user how much larger our
3089                  * amount of data is than his buffer, but that presents
3090                  * some problems:
3091                  *
3092                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3093                  *     we can't hold a lock while calling them with an
3094                  *     auto-extending buffer.
3095                  *
3096                  * 2.  There is not currently a LUN reference counting
3097                  *     mechanism, outside of outstanding transactions on
3098                  *     the LUN's OOA queue.  So a LUN could go away on us
3099                  *     while we're getting the LUN number, backend-specific
3100                  *     information, etc.  Thus, given the way things
3101                  *     currently work, we need to hold the CTL lock while
3102                  *     grabbing LUN information.
3103                  *
3104                  * So, from the user's standpoint, the best thing to do is
3105                  * allocate what he thinks is a reasonable buffer length,
3106                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3107                  * double the buffer length and try again.  (And repeat
3108                  * that until he succeeds.)
3109                  */
3110                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3111                 if (sb == NULL) {
3112                         list->status = CTL_LUN_LIST_ERROR;
3113                         snprintf(list->error_str, sizeof(list->error_str),
3114                                  "Unable to allocate %d bytes for LUN list",
3115                                  list->alloc_len);
3116                         break;
3117                 }
3118
3119                 sbuf_printf(sb, "<ctllunlist>\n");
3120
3121                 mtx_lock(&softc->ctl_lock);
3122                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3123                         mtx_lock(&lun->lun_lock);
3124                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3125                                              (uintmax_t)lun->lun);
3126
3127                         /*
3128                          * Bail out as soon as we see that we've overfilled
3129                          * the buffer.
3130                          */
3131                         if (retval != 0)
3132                                 break;
3133
3134                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3135                                              "</backend_type>\n",
3136                                              (lun->backend == NULL) ?  "none" :
3137                                              lun->backend->name);
3138
3139                         if (retval != 0)
3140                                 break;
3141
3142                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3143                                              lun->be_lun->lun_type);
3144
3145                         if (retval != 0)
3146                                 break;
3147
3148                         if (lun->backend == NULL) {
3149                                 retval = sbuf_printf(sb, "</lun>\n");
3150                                 if (retval != 0)
3151                                         break;
3152                                 continue;
3153                         }
3154
3155                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3156                                              (lun->be_lun->maxlba > 0) ?
3157                                              lun->be_lun->maxlba + 1 : 0);
3158
3159                         if (retval != 0)
3160                                 break;
3161
3162                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3163                                              lun->be_lun->blocksize);
3164
3165                         if (retval != 0)
3166                                 break;
3167
3168                         retval = sbuf_printf(sb, "\t<serial_number>");
3169
3170                         if (retval != 0)
3171                                 break;
3172
3173                         retval = ctl_sbuf_printf_esc(sb,
3174                                                      lun->be_lun->serial_num);
3175
3176                         if (retval != 0)
3177                                 break;
3178
3179                         retval = sbuf_printf(sb, "</serial_number>\n");
3180
3181                         if (retval != 0)
3182                                 break;
3183
3184                         retval = sbuf_printf(sb, "\t<device_id>");
3185
3186                         if (retval != 0)
3187                                 break;
3188
3189                         retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3190
3191                         if (retval != 0)
3192                                 break;
3193
3194                         retval = sbuf_printf(sb, "</device_id>\n");
3195
3196                         if (retval != 0)
3197                                 break;
3198
3199                         if (lun->backend->lun_info != NULL) {
3200                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3201                                 if (retval != 0)
3202                                         break;
3203                         }
3204                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3205                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3206                                     opt->name, opt->value, opt->name);
3207                                 if (retval != 0)
3208                                         break;
3209                         }
3210
3211                         retval = sbuf_printf(sb, "</lun>\n");
3212
3213                         if (retval != 0)
3214                                 break;
3215                         mtx_unlock(&lun->lun_lock);
3216                 }
3217                 if (lun != NULL)
3218                         mtx_unlock(&lun->lun_lock);
3219                 mtx_unlock(&softc->ctl_lock);
3220
3221                 if ((retval != 0)
3222                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3223                         retval = 0;
3224                         sbuf_delete(sb);
3225                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3226                         snprintf(list->error_str, sizeof(list->error_str),
3227                                  "Out of space, %d bytes is too small",
3228                                  list->alloc_len);
3229                         break;
3230                 }
3231
3232                 sbuf_finish(sb);
3233
3234                 retval = copyout(sbuf_data(sb), list->lun_xml,
3235                                  sbuf_len(sb) + 1);
3236
3237                 list->fill_len = sbuf_len(sb) + 1;
3238                 list->status = CTL_LUN_LIST_OK;
3239                 sbuf_delete(sb);
3240                 break;
3241         }
3242         case CTL_ISCSI: {
3243                 struct ctl_iscsi *ci;
3244                 struct ctl_frontend *fe;
3245
3246                 ci = (struct ctl_iscsi *)addr;
3247
3248                 fe = ctl_frontend_find("iscsi");
3249                 if (fe == NULL) {
3250                         ci->status = CTL_ISCSI_ERROR;
3251                         snprintf(ci->error_str, sizeof(ci->error_str),
3252                             "Frontend \"iscsi\" not found.");
3253                         break;
3254                 }
3255
3256                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3257                 break;
3258         }
3259         case CTL_PORT_REQ: {
3260                 struct ctl_req *req;
3261                 struct ctl_frontend *fe;
3262
3263                 req = (struct ctl_req *)addr;
3264
3265                 fe = ctl_frontend_find(req->driver);
3266                 if (fe == NULL) {
3267                         req->status = CTL_LUN_ERROR;
3268                         snprintf(req->error_str, sizeof(req->error_str),
3269                             "Frontend \"%s\" not found.", req->driver);
3270                         break;
3271                 }
3272                 if (req->num_args > 0) {
3273                         req->kern_args = ctl_copyin_args(req->num_args,
3274                             req->args, req->error_str, sizeof(req->error_str));
3275                         if (req->kern_args == NULL) {
3276                                 req->status = CTL_LUN_ERROR;
3277                                 break;
3278                         }
3279                 }
3280
3281                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3282
3283                 if (req->num_args > 0) {
3284                         ctl_copyout_args(req->num_args, req->kern_args);
3285                         ctl_free_args(req->num_args, req->kern_args);
3286                 }
3287                 break;
3288         }
3289         case CTL_PORT_LIST: {
3290                 struct sbuf *sb;
3291                 struct ctl_port *port;
3292                 struct ctl_lun_list *list;
3293                 struct ctl_option *opt;
3294
3295                 list = (struct ctl_lun_list *)addr;
3296
3297                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3298                 if (sb == NULL) {
3299                         list->status = CTL_LUN_LIST_ERROR;
3300                         snprintf(list->error_str, sizeof(list->error_str),
3301                                  "Unable to allocate %d bytes for LUN list",
3302                                  list->alloc_len);
3303                         break;
3304                 }
3305
3306                 sbuf_printf(sb, "<ctlportlist>\n");
3307
3308                 mtx_lock(&softc->ctl_lock);
3309                 STAILQ_FOREACH(port, &softc->port_list, links) {
3310                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3311                                              (uintmax_t)port->targ_port);
3312
3313                         /*
3314                          * Bail out as soon as we see that we've overfilled
3315                          * the buffer.
3316                          */
3317                         if (retval != 0)
3318                                 break;
3319
3320                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3321                             "</frontend_type>\n", port->frontend->name);
3322                         if (retval != 0)
3323                                 break;
3324
3325                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3326                                              port->port_type);
3327                         if (retval != 0)
3328                                 break;
3329
3330                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3331                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3332                         if (retval != 0)
3333                                 break;
3334
3335                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3336                             port->port_name);
3337                         if (retval != 0)
3338                                 break;
3339
3340                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3341                             port->physical_port);
3342                         if (retval != 0)
3343                                 break;
3344
3345                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3346                             port->virtual_port);
3347                         if (retval != 0)
3348                                 break;
3349
3350                         retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3351                             (uintmax_t)port->wwnn);
3352                         if (retval != 0)
3353                                 break;
3354
3355                         retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3356                             (uintmax_t)port->wwpn);
3357                         if (retval != 0)
3358                                 break;
3359
3360                         if (port->port_info != NULL) {
3361                                 retval = port->port_info(port->onoff_arg, sb);
3362                                 if (retval != 0)
3363                                         break;
3364                         }
3365                         STAILQ_FOREACH(opt, &port->options, links) {
3366                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3367                                     opt->name, opt->value, opt->name);
3368                                 if (retval != 0)
3369                                         break;
3370                         }
3371
3372                         retval = sbuf_printf(sb, "</targ_port>\n");
3373                         if (retval != 0)
3374                                 break;
3375                 }
3376                 mtx_unlock(&softc->ctl_lock);
3377
3378                 if ((retval != 0)
3379                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3380                         retval = 0;
3381                         sbuf_delete(sb);
3382                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3383                         snprintf(list->error_str, sizeof(list->error_str),
3384                                  "Out of space, %d bytes is too small",
3385                                  list->alloc_len);
3386                         break;
3387                 }
3388
3389                 sbuf_finish(sb);
3390
3391                 retval = copyout(sbuf_data(sb), list->lun_xml,
3392                                  sbuf_len(sb) + 1);
3393
3394                 list->fill_len = sbuf_len(sb) + 1;
3395                 list->status = CTL_LUN_LIST_OK;
3396                 sbuf_delete(sb);
3397                 break;
3398         }
3399         default: {
3400                 /* XXX KDM should we fix this? */
3401 #if 0
3402                 struct ctl_backend_driver *backend;
3403                 unsigned int type;
3404                 int found;
3405
3406                 found = 0;
3407
3408                 /*
3409                  * We encode the backend type as the ioctl type for backend
3410                  * ioctls.  So parse it out here, and then search for a
3411                  * backend of this type.
3412                  */
3413                 type = _IOC_TYPE(cmd);
3414
3415                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3416                         if (backend->type == type) {
3417                                 found = 1;
3418                                 break;
3419                         }
3420                 }
3421                 if (found == 0) {
3422                         printf("ctl: unknown ioctl command %#lx or backend "
3423                                "%d\n", cmd, type);
3424                         retval = EINVAL;
3425                         break;
3426                 }
3427                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3428 #endif
3429                 retval = ENOTTY;
3430                 break;
3431         }
3432         }
3433         return (retval);
3434 }
3435
3436 uint32_t
3437 ctl_get_initindex(struct ctl_nexus *nexus)
3438 {
3439         if (nexus->targ_port < CTL_MAX_PORTS)
3440                 return (nexus->initid.id +
3441                         (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3442         else
3443                 return (nexus->initid.id +
3444                        ((nexus->targ_port - CTL_MAX_PORTS) *
3445                         CTL_MAX_INIT_PER_PORT));
3446 }
3447
3448 uint32_t
3449 ctl_get_resindex(struct ctl_nexus *nexus)
3450 {
3451         return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3452 }
3453
3454 uint32_t
3455 ctl_port_idx(int port_num)
3456 {
3457         if (port_num < CTL_MAX_PORTS)
3458                 return(port_num);
3459         else
3460                 return(port_num - CTL_MAX_PORTS);
3461 }
3462
3463 static uint32_t
3464 ctl_map_lun(int port_num, uint32_t lun_id)
3465 {
3466         struct ctl_port *port;
3467
3468         port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3469         if (port == NULL)
3470                 return (UINT32_MAX);
3471         if (port->lun_map == NULL)
3472                 return (lun_id);
3473         return (port->lun_map(port->targ_lun_arg, lun_id));
3474 }
3475
3476 static uint32_t
3477 ctl_map_lun_back(int port_num, uint32_t lun_id)
3478 {
3479         struct ctl_port *port;
3480         uint32_t i;
3481
3482         port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3483         if (port->lun_map == NULL)
3484                 return (lun_id);
3485         for (i = 0; i < CTL_MAX_LUNS; i++) {
3486                 if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3487                         return (i);
3488         }
3489         return (UINT32_MAX);
3490 }
3491
3492 /*
3493  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3494  * that are a power of 2.
3495  */
3496 int
3497 ctl_ffz(uint32_t *mask, uint32_t size)
3498 {
3499         uint32_t num_chunks, num_pieces;
3500         int i, j;
3501
3502         num_chunks = (size >> 5);
3503         if (num_chunks == 0)
3504                 num_chunks++;
3505         num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3506
3507         for (i = 0; i < num_chunks; i++) {
3508                 for (j = 0; j < num_pieces; j++) {
3509                         if ((mask[i] & (1 << j)) == 0)
3510                                 return ((i << 5) + j);
3511                 }
3512         }
3513
3514         return (-1);
3515 }
3516
3517 int
3518 ctl_set_mask(uint32_t *mask, uint32_t bit)
3519 {
3520         uint32_t chunk, piece;
3521
3522         chunk = bit >> 5;
3523         piece = bit % (sizeof(uint32_t) * 8);
3524
3525         if ((mask[chunk] & (1 << piece)) != 0)
3526                 return (-1);
3527         else
3528                 mask[chunk] |= (1 << piece);
3529
3530         return (0);
3531 }
3532
3533 int
3534 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3535 {
3536         uint32_t chunk, piece;
3537
3538         chunk = bit >> 5;
3539         piece = bit % (sizeof(uint32_t) * 8);
3540
3541         if ((mask[chunk] & (1 << piece)) == 0)
3542                 return (-1);
3543         else
3544                 mask[chunk] &= ~(1 << piece);
3545
3546         return (0);
3547 }
3548
3549 int
3550 ctl_is_set(uint32_t *mask, uint32_t bit)
3551 {
3552         uint32_t chunk, piece;
3553
3554         chunk = bit >> 5;
3555         piece = bit % (sizeof(uint32_t) * 8);
3556
3557         if ((mask[chunk] & (1 << piece)) == 0)
3558                 return (0);
3559         else
3560                 return (1);
3561 }
3562
3563 #ifdef unused
3564 /*
3565  * The bus, target and lun are optional, they can be filled in later.
3566  * can_wait is used to determine whether we can wait on the malloc or not.
3567  */
3568 union ctl_io*
3569 ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3570               uint32_t targ_lun, int can_wait)
3571 {
3572         union ctl_io *io;
3573
3574         if (can_wait)
3575                 io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3576         else
3577                 io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3578
3579         if (io != NULL) {
3580                 io->io_hdr.io_type = io_type;
3581                 io->io_hdr.targ_port = targ_port;
3582                 /*
3583                  * XXX KDM this needs to change/go away.  We need to move
3584                  * to a preallocated pool of ctl_scsiio structures.
3585                  */
3586                 io->io_hdr.nexus.targ_target.id = targ_target;
3587                 io->io_hdr.nexus.targ_lun = targ_lun;
3588         }
3589
3590         return (io);
3591 }
3592
3593 void
3594 ctl_kfree_io(union ctl_io *io)
3595 {
3596         free(io, M_CTL);
3597 }
3598 #endif /* unused */
3599
3600 /*
3601  * ctl_softc, pool_type, total_ctl_io are passed in.
3602  * npool is passed out.
3603  */
3604 int
3605 ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3606                 uint32_t total_ctl_io, struct ctl_io_pool **npool)
3607 {
3608         uint32_t i;
3609         union ctl_io *cur_io, *next_io;
3610         struct ctl_io_pool *pool;
3611         int retval;
3612
3613         retval = 0;
3614
3615         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3616                                             M_NOWAIT | M_ZERO);
3617         if (pool == NULL) {
3618                 retval = ENOMEM;
3619                 goto bailout;
3620         }
3621
3622         pool->type = pool_type;
3623         pool->ctl_softc = ctl_softc;
3624
3625         mtx_lock(&ctl_softc->pool_lock);
3626         pool->id = ctl_softc->cur_pool_id++;
3627         mtx_unlock(&ctl_softc->pool_lock);
3628
3629         pool->flags = CTL_POOL_FLAG_NONE;
3630         pool->refcount = 1;             /* Reference for validity. */
3631         STAILQ_INIT(&pool->free_queue);
3632
3633         /*
3634          * XXX KDM other options here:
3635          * - allocate a page at a time
3636          * - allocate one big chunk of memory.
3637          * Page allocation might work well, but would take a little more
3638          * tracking.
3639          */
3640         for (i = 0; i < total_ctl_io; i++) {
3641                 cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3642                                                 M_NOWAIT);
3643                 if (cur_io == NULL) {
3644                         retval = ENOMEM;
3645                         break;
3646                 }
3647                 cur_io->io_hdr.pool = pool;
3648                 STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3649                 pool->total_ctl_io++;
3650                 pool->free_ctl_io++;
3651         }
3652
3653         if (retval != 0) {
3654                 for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3655                      cur_io != NULL; cur_io = next_io) {
3656                         next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3657                                                               links);
3658                         STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3659                                       ctl_io_hdr, links);
3660                         free(cur_io, M_CTLIO);
3661                 }
3662
3663                 free(pool, M_CTL);
3664                 goto bailout;
3665         }
3666         mtx_lock(&ctl_softc->pool_lock);
3667         ctl_softc->num_pools++;
3668         STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3669         /*
3670          * Increment our usage count if this is an external consumer, so we
3671          * can't get unloaded until the external consumer (most likely a
3672          * FETD) unloads and frees his pool.
3673          *
3674          * XXX KDM will this increment the caller's module use count, or
3675          * mine?
3676          */
3677 #if 0
3678         if ((pool_type != CTL_POOL_EMERGENCY)
3679          && (pool_type != CTL_POOL_INTERNAL)
3680          && (pool_type != CTL_POOL_4OTHERSC))
3681                 MOD_INC_USE_COUNT;
3682 #endif
3683
3684         mtx_unlock(&ctl_softc->pool_lock);
3685
3686         *npool = pool;
3687
3688 bailout:
3689
3690         return (retval);
3691 }
3692
3693 static int
3694 ctl_pool_acquire(struct ctl_io_pool *pool)
3695 {
3696
3697         mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3698
3699         if (pool->flags & CTL_POOL_FLAG_INVALID)
3700                 return (EINVAL);
3701
3702         pool->refcount++;
3703
3704         return (0);
3705 }
3706
3707 static void
3708 ctl_pool_release(struct ctl_io_pool *pool)
3709 {
3710         struct ctl_softc *ctl_softc = pool->ctl_softc;
3711         union ctl_io *io;
3712
3713         mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3714
3715         if (--pool->refcount != 0)
3716                 return;
3717
3718         while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3719                 STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3720                               links);
3721                 free(io, M_CTLIO);
3722         }
3723
3724         STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3725         ctl_softc->num_pools--;
3726
3727         /*
3728          * XXX KDM will this decrement the caller's usage count or mine?
3729          */
3730 #if 0
3731         if ((pool->type != CTL_POOL_EMERGENCY)
3732          && (pool->type != CTL_POOL_INTERNAL)
3733          && (pool->type != CTL_POOL_4OTHERSC))
3734                 MOD_DEC_USE_COUNT;
3735 #endif
3736
3737         free(pool, M_CTL);
3738 }
3739
3740 void
3741 ctl_pool_free(struct ctl_io_pool *pool)
3742 {
3743         struct ctl_softc *ctl_softc;
3744
3745         if (pool == NULL)
3746                 return;
3747
3748         ctl_softc = pool->ctl_softc;
3749         mtx_lock(&ctl_softc->pool_lock);
3750         pool->flags |= CTL_POOL_FLAG_INVALID;
3751         ctl_pool_release(pool);
3752         mtx_unlock(&ctl_softc->pool_lock);
3753 }
3754
3755 /*
3756  * This routine does not block (except for spinlocks of course).
3757  * It tries to allocate a ctl_io union from the caller's pool as quickly as
3758  * possible.
3759  */
3760 union ctl_io *
3761 ctl_alloc_io(void *pool_ref)
3762 {
3763         union ctl_io *io;
3764         struct ctl_softc *ctl_softc;
3765         struct ctl_io_pool *pool, *npool;
3766         struct ctl_io_pool *emergency_pool;
3767
3768         pool = (struct ctl_io_pool *)pool_ref;
3769
3770         if (pool == NULL) {
3771                 printf("%s: pool is NULL\n", __func__);
3772                 return (NULL);
3773         }
3774
3775         emergency_pool = NULL;
3776
3777         ctl_softc = pool->ctl_softc;
3778
3779         mtx_lock(&ctl_softc->pool_lock);
3780         /*
3781          * First, try to get the io structure from the user's pool.
3782          */
3783         if (ctl_pool_acquire(pool) == 0) {
3784                 io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3785                 if (io != NULL) {
3786                         STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3787                         pool->total_allocated++;
3788                         pool->free_ctl_io--;
3789                         mtx_unlock(&ctl_softc->pool_lock);
3790                         return (io);
3791                 } else
3792                         ctl_pool_release(pool);
3793         }
3794         /*
3795          * If he doesn't have any io structures left, search for an
3796          * emergency pool and grab one from there.
3797          */
3798         STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3799                 if (npool->type != CTL_POOL_EMERGENCY)
3800                         continue;
3801
3802                 if (ctl_pool_acquire(npool) != 0)
3803                         continue;
3804
3805                 emergency_pool = npool;
3806
3807                 io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3808                 if (io != NULL) {
3809                         STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3810                         npool->total_allocated++;
3811                         npool->free_ctl_io--;
3812                         mtx_unlock(&ctl_softc->pool_lock);
3813                         return (io);
3814                 } else
3815                         ctl_pool_release(npool);
3816         }
3817
3818         /* Drop the spinlock before we malloc */
3819         mtx_unlock(&ctl_softc->pool_lock);
3820
3821         /*
3822          * The emergency pool (if it exists) didn't have one, so try an
3823          * atomic (i.e. nonblocking) malloc and see if we get lucky.
3824          */
3825         io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3826         if (io != NULL) {
3827                 /*
3828                  * If the emergency pool exists but is empty, add this
3829                  * ctl_io to its list when it gets freed.
3830                  */
3831                 if (emergency_pool != NULL) {
3832                         mtx_lock(&ctl_softc->pool_lock);
3833                         if (ctl_pool_acquire(emergency_pool) == 0) {
3834                                 io->io_hdr.pool = emergency_pool;
3835                                 emergency_pool->total_ctl_io++;
3836                                 /*
3837                                  * Need to bump this, otherwise
3838                                  * total_allocated and total_freed won't
3839                                  * match when we no longer have anything
3840                                  * outstanding.
3841                                  */
3842                                 emergency_pool->total_allocated++;
3843                         }
3844                         mtx_unlock(&ctl_softc->pool_lock);
3845                 } else
3846                         io->io_hdr.pool = NULL;
3847         }
3848
3849         return (io);
3850 }
3851
3852 void
3853 ctl_free_io(union ctl_io *io)
3854 {
3855         if (io == NULL)
3856                 return;
3857
3858         /*
3859          * If this ctl_io has a pool, return it to that pool.
3860          */
3861         if (io->io_hdr.pool != NULL) {
3862                 struct ctl_io_pool *pool;
3863
3864                 pool = (struct ctl_io_pool *)io->io_hdr.pool;
3865                 mtx_lock(&pool->ctl_softc->pool_lock);
3866                 io->io_hdr.io_type = 0xff;
3867                 STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3868                 pool->total_freed++;
3869                 pool->free_ctl_io++;
3870                 ctl_pool_release(pool);
3871                 mtx_unlock(&pool->ctl_softc->pool_lock);
3872         } else {
3873                 /*
3874                  * Otherwise, just free it.  We probably malloced it and
3875                  * the emergency pool wasn't available.
3876                  */
3877                 free(io, M_CTLIO);
3878         }
3879
3880 }
3881
3882 void
3883 ctl_zero_io(union ctl_io *io)
3884 {
3885         void *pool_ref;
3886
3887         if (io == NULL)
3888                 return;
3889
3890         /*
3891          * May need to preserve linked list pointers at some point too.
3892          */
3893         pool_ref = io->io_hdr.pool;
3894
3895         memset(io, 0, sizeof(*io));
3896
3897         io->io_hdr.pool = pool_ref;
3898 }
3899
3900 /*
3901  * This routine is currently used for internal copies of ctl_ios that need
3902  * to persist for some reason after we've already returned status to the
3903  * FETD.  (Thus the flag set.)
3904  *
3905  * XXX XXX
3906  * Note that this makes a blind copy of all fields in the ctl_io, except
3907  * for the pool reference.  This includes any memory that has been
3908  * allocated!  That memory will no longer be valid after done has been
3909  * called, so this would be VERY DANGEROUS for command that actually does
3910  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3911  * start and stop commands, which don't transfer any data, so this is not a
3912  * problem.  If it is used for anything else, the caller would also need to
3913  * allocate data buffer space and this routine would need to be modified to
3914  * copy the data buffer(s) as well.
3915  */
3916 void
3917 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3918 {
3919         void *pool_ref;
3920
3921         if ((src == NULL)
3922          || (dest == NULL))
3923                 return;
3924
3925         /*
3926          * May need to preserve linked list pointers at some point too.
3927          */
3928         pool_ref = dest->io_hdr.pool;
3929
3930         memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3931
3932         dest->io_hdr.pool = pool_ref;
3933         /*
3934          * We need to know that this is an internal copy, and doesn't need
3935          * to get passed back to the FETD that allocated it.
3936          */
3937         dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3938 }
3939
3940 #ifdef NEEDTOPORT
3941 static void
3942 ctl_update_power_subpage(struct copan_power_subpage *page)
3943 {
3944         int num_luns, num_partitions, config_type;
3945         struct ctl_softc *softc;
3946         cs_BOOL_t aor_present, shelf_50pct_power;
3947         cs_raidset_personality_t rs_type;
3948         int max_active_luns;
3949
3950         softc = control_softc;
3951
3952         /* subtract out the processor LUN */
3953         num_luns = softc->num_luns - 1;
3954         /*
3955          * Default to 7 LUNs active, which was the only number we allowed
3956          * in the past.
3957          */
3958         max_active_luns = 7;
3959
3960         num_partitions = config_GetRsPartitionInfo();
3961         config_type = config_GetConfigType();
3962         shelf_50pct_power = config_GetShelfPowerMode();
3963         aor_present = config_IsAorRsPresent();
3964
3965         rs_type = ddb_GetRsRaidType(1);
3966         if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3967          && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3968                 EPRINT(0, "Unsupported RS type %d!", rs_type);
3969         }
3970
3971
3972         page->total_luns = num_luns;
3973
3974         switch (config_type) {
3975         case 40:
3976                 /*
3977                  * In a 40 drive configuration, it doesn't matter what DC
3978                  * cards we have, whether we have AOR enabled or not,
3979                  * partitioning or not, or what type of RAIDset we have.
3980                  * In that scenario, we can power up every LUN we present
3981                  * to the user.
3982                  */
3983                 max_active_luns = num_luns;
3984
3985                 break;
3986         case 64:
3987                 if (shelf_50pct_power == CS_FALSE) {
3988                         /* 25% power */
3989                         if (aor_present == CS_TRUE) {
3990                                 if (rs_type ==
3991                                      CS_RAIDSET_PERSONALITY_RAID5) {
3992                                         max_active_luns = 7;
3993                                 } else if (rs_type ==
3994                                          CS_RAIDSET_PERSONALITY_RAID1){
3995                                         max_active_luns = 14;
3996                                 } else {
3997                                         /* XXX KDM now what?? */
3998                                 }
3999                         } else {
4000                                 if (rs_type ==
4001                                      CS_RAIDSET_PERSONALITY_RAID5) {
4002                                         max_active_luns = 8;
4003                                 } else if (rs_type ==
4004                                          CS_RAIDSET_PERSONALITY_RAID1){
4005                                         max_active_luns = 16;
4006                                 } else {
4007                                         /* XXX KDM now what?? */
4008                                 }
4009                         }
4010                 } else {
4011                         /* 50% power */
4012                         /*
4013                          * With 50% power in a 64 drive configuration, we
4014                          * can power all LUNs we present.
4015                          */
4016                         max_active_luns = num_luns;
4017                 }
4018                 break;
4019         case 112:
4020                 if (shelf_50pct_power == CS_FALSE) {
4021                         /* 25% power */
4022                         if (aor_present == CS_TRUE) {
4023                                 if (rs_type ==
4024                                      CS_RAIDSET_PERSONALITY_RAID5) {
4025                                         max_active_luns = 7;
4026                                 } else if (rs_type ==
4027                                          CS_RAIDSET_PERSONALITY_RAID1){
4028                                         max_active_luns = 14;
4029                                 } else {
4030                                         /* XXX KDM now what?? */
4031                                 }
4032                         } else {
4033                                 if (rs_type ==
4034                                      CS_RAIDSET_PERSONALITY_RAID5) {
4035                                         max_active_luns = 8;
4036                                 } else if (rs_type ==
4037                                          CS_RAIDSET_PERSONALITY_RAID1){
4038                                         max_active_luns = 16;
4039                                 } else {
4040                                         /* XXX KDM now what?? */
4041                                 }
4042                         }
4043                 } else {
4044                         /* 50% power */
4045                         if (aor_present == CS_TRUE) {
4046                                 if (rs_type ==
4047                                      CS_RAIDSET_PERSONALITY_RAID5) {
4048                                         max_active_luns = 14;
4049                                 } else if (rs_type ==
4050                                          CS_RAIDSET_PERSONALITY_RAID1){
4051                                         /*
4052                                          * We're assuming here that disk
4053                                          * caching is enabled, and so we're
4054                                          * able to power up half of each
4055                                          * LUN, and cache all writes.
4056                                          */
4057                                         max_active_luns = num_luns;
4058                                 } else {
4059                                         /* XXX KDM now what?? */
4060                                 }
4061                         } else {
4062                                 if (rs_type ==
4063                                      CS_RAIDSET_PERSONALITY_RAID5) {
4064                                         max_active_luns = 15;
4065                                 } else if (rs_type ==
4066                                          CS_RAIDSET_PERSONALITY_RAID1){
4067                                         max_active_luns = 30;
4068                                 } else {
4069                                         /* XXX KDM now what?? */
4070                                 }
4071                         }
4072                 }
4073                 break;
4074         default:
4075                 /*
4076                  * In this case, we have an unknown configuration, so we
4077                  * just use the default from above.
4078                  */
4079                 break;
4080         }
4081
4082         page->max_active_luns = max_active_luns;
4083 #if 0
4084         printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4085                page->total_luns, page->max_active_luns);
4086 #endif
4087 }
4088 #endif /* NEEDTOPORT */
4089
4090 /*
4091  * This routine could be used in the future to load default and/or saved
4092  * mode page parameters for a particuar lun.
4093  */
4094 static int
4095 ctl_init_page_index(struct ctl_lun *lun)
4096 {
4097         int i;
4098         struct ctl_page_index *page_index;
4099         struct ctl_softc *softc;
4100         const char *value;
4101
4102         memcpy(&lun->mode_pages.index, page_index_template,
4103                sizeof(page_index_template));
4104
4105         softc = lun->ctl_softc;
4106
4107         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4108
4109                 page_index = &lun->mode_pages.index[i];
4110                 /*
4111                  * If this is a disk-only mode page, there's no point in
4112                  * setting it up.  For some pages, we have to have some
4113                  * basic information about the disk in order to calculate the
4114                  * mode page data.
4115                  */
4116                 if ((lun->be_lun->lun_type != T_DIRECT)
4117                  && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4118                         continue;
4119
4120                 switch (page_index->page_code & SMPH_PC_MASK) {
4121                 case SMS_FORMAT_DEVICE_PAGE: {
4122                         struct scsi_format_page *format_page;
4123
4124                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4125                                 panic("subpage is incorrect!");
4126
4127                         /*
4128                          * Sectors per track are set above.  Bytes per
4129                          * sector need to be set here on a per-LUN basis.
4130                          */
4131                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4132                                &format_page_default,
4133                                sizeof(format_page_default));
4134                         memcpy(&lun->mode_pages.format_page[
4135                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4136                                sizeof(format_page_changeable));
4137                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4138                                &format_page_default,
4139                                sizeof(format_page_default));
4140                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4141                                &format_page_default,
4142                                sizeof(format_page_default));
4143
4144                         format_page = &lun->mode_pages.format_page[
4145                                 CTL_PAGE_CURRENT];
4146                         scsi_ulto2b(lun->be_lun->blocksize,
4147                                     format_page->bytes_per_sector);
4148
4149                         format_page = &lun->mode_pages.format_page[
4150                                 CTL_PAGE_DEFAULT];
4151                         scsi_ulto2b(lun->be_lun->blocksize,
4152                                     format_page->bytes_per_sector);
4153
4154                         format_page = &lun->mode_pages.format_page[
4155                                 CTL_PAGE_SAVED];
4156                         scsi_ulto2b(lun->be_lun->blocksize,
4157                                     format_page->bytes_per_sector);
4158
4159                         page_index->page_data =
4160                                 (uint8_t *)lun->mode_pages.format_page;
4161                         break;
4162                 }
4163                 case SMS_RIGID_DISK_PAGE: {
4164                         struct scsi_rigid_disk_page *rigid_disk_page;
4165                         uint32_t sectors_per_cylinder;
4166                         uint64_t cylinders;
4167 #ifndef __XSCALE__
4168                         int shift;
4169 #endif /* !__XSCALE__ */
4170
4171                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4172                                 panic("invalid subpage value %d",
4173                                       page_index->subpage);
4174
4175                         /*
4176                          * Rotation rate and sectors per track are set
4177                          * above.  We calculate the cylinders here based on
4178                          * capacity.  Due to the number of heads and
4179                          * sectors per track we're using, smaller arrays
4180                          * may turn out to have 0 cylinders.  Linux and
4181                          * FreeBSD don't pay attention to these mode pages
4182                          * to figure out capacity, but Solaris does.  It
4183                          * seems to deal with 0 cylinders just fine, and
4184                          * works out a fake geometry based on the capacity.
4185                          */
4186                         memcpy(&lun->mode_pages.rigid_disk_page[
4187                                CTL_PAGE_CURRENT], &rigid_disk_page_default,
4188                                sizeof(rigid_disk_page_default));
4189                         memcpy(&lun->mode_pages.rigid_disk_page[
4190                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4191                                sizeof(rigid_disk_page_changeable));
4192                         memcpy(&lun->mode_pages.rigid_disk_page[
4193                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4194                                sizeof(rigid_disk_page_default));
4195                         memcpy(&lun->mode_pages.rigid_disk_page[
4196                                CTL_PAGE_SAVED], &rigid_disk_page_default,
4197                                sizeof(rigid_disk_page_default));
4198
4199                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4200                                 CTL_DEFAULT_HEADS;
4201
4202                         /*
4203                          * The divide method here will be more accurate,
4204                          * probably, but results in floating point being
4205                          * used in the kernel on i386 (__udivdi3()).  On the
4206                          * XScale, though, __udivdi3() is implemented in
4207                          * software.
4208                          *
4209                          * The shift method for cylinder calculation is
4210                          * accurate if sectors_per_cylinder is a power of
4211                          * 2.  Otherwise it might be slightly off -- you
4212                          * might have a bit of a truncation problem.
4213                          */
4214 #ifdef  __XSCALE__
4215                         cylinders = (lun->be_lun->maxlba + 1) /
4216                                 sectors_per_cylinder;
4217 #else
4218                         for (shift = 31; shift > 0; shift--) {
4219                                 if (sectors_per_cylinder & (1 << shift))
4220                                         break;
4221                         }
4222                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4223 #endif
4224
4225                         /*
4226                          * We've basically got 3 bytes, or 24 bits for the
4227                          * cylinder size in the mode page.  If we're over,
4228                          * just round down to 2^24.
4229                          */
4230                         if (cylinders > 0xffffff)
4231                                 cylinders = 0xffffff;
4232
4233                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4234                                 CTL_PAGE_CURRENT];
4235                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4236
4237                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4238                                 CTL_PAGE_DEFAULT];
4239                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4240
4241                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4242                                 CTL_PAGE_SAVED];
4243                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4244
4245                         page_index->page_data =
4246                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4247                         break;
4248                 }
4249                 case SMS_CACHING_PAGE: {
4250                         struct scsi_caching_page *caching_page;
4251
4252                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4253                                 panic("invalid subpage value %d",
4254                                       page_index->subpage);
4255                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4256                                &caching_page_default,
4257                                sizeof(caching_page_default));
4258                         memcpy(&lun->mode_pages.caching_page[
4259                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4260                                sizeof(caching_page_changeable));
4261                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4262                                &caching_page_default,
4263                                sizeof(caching_page_default));
4264                         caching_page = &lun->mode_pages.caching_page[
4265                             CTL_PAGE_SAVED];
4266                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
4267                         if (value != NULL && strcmp(value, "off") == 0)
4268                                 caching_page->flags1 &= ~SCP_WCE;
4269                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4270                         if (value != NULL && strcmp(value, "off") == 0)
4271                                 caching_page->flags1 |= SCP_RCD;
4272                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4273                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4274                                sizeof(caching_page_default));
4275                         page_index->page_data =
4276                                 (uint8_t *)lun->mode_pages.caching_page;
4277                         break;
4278                 }
4279                 case SMS_CONTROL_MODE_PAGE: {
4280                         struct scsi_control_page *control_page;
4281
4282                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4283                                 panic("invalid subpage value %d",
4284                                       page_index->subpage);
4285
4286                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4287                                &control_page_default,
4288                                sizeof(control_page_default));
4289                         memcpy(&lun->mode_pages.control_page[
4290                                CTL_PAGE_CHANGEABLE], &control_page_changeable,
4291                                sizeof(control_page_changeable));
4292                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4293                                &control_page_default,
4294                                sizeof(control_page_default));
4295                         control_page = &lun->mode_pages.control_page[
4296                             CTL_PAGE_SAVED];
4297                         value = ctl_get_opt(&lun->be_lun->options, "reordering");
4298                         if (value != NULL && strcmp(value, "unrestricted") == 0) {
4299                                 control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4300                                 control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4301                         }
4302                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4303                                &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4304                                sizeof(control_page_default));
4305                         page_index->page_data =
4306                                 (uint8_t *)lun->mode_pages.control_page;
4307                         break;
4308
4309                 }
4310                 case SMS_VENDOR_SPECIFIC_PAGE:{
4311                         switch (page_index->subpage) {
4312                         case PWR_SUBPAGE_CODE: {
4313                                 struct copan_power_subpage *current_page,
4314                                                            *saved_page;
4315
4316                                 memcpy(&lun->mode_pages.power_subpage[
4317                                        CTL_PAGE_CURRENT],
4318                                        &power_page_default,
4319                                        sizeof(power_page_default));
4320                                 memcpy(&lun->mode_pages.power_subpage[
4321                                        CTL_PAGE_CHANGEABLE],
4322                                        &power_page_changeable,
4323                                        sizeof(power_page_changeable));
4324                                 memcpy(&lun->mode_pages.power_subpage[
4325                                        CTL_PAGE_DEFAULT],
4326                                        &power_page_default,
4327                                        sizeof(power_page_default));
4328                                 memcpy(&lun->mode_pages.power_subpage[
4329                                        CTL_PAGE_SAVED],
4330                                        &power_page_default,
4331                                        sizeof(power_page_default));
4332                                 page_index->page_data =
4333                                     (uint8_t *)lun->mode_pages.power_subpage;
4334
4335                                 current_page = (struct copan_power_subpage *)
4336                                         (page_index->page_data +
4337                                          (page_index->page_len *
4338                                           CTL_PAGE_CURRENT));
4339                                 saved_page = (struct copan_power_subpage *)
4340                                         (page_index->page_data +
4341                                          (page_index->page_len *
4342                                           CTL_PAGE_SAVED));
4343                                 break;
4344                         }
4345                         case APS_SUBPAGE_CODE: {
4346                                 struct copan_aps_subpage *current_page,
4347                                                          *saved_page;
4348
4349                                 // This gets set multiple times but
4350                                 // it should always be the same. It's
4351                                 // only done during init so who cares.
4352                                 index_to_aps_page = i;
4353
4354                                 memcpy(&lun->mode_pages.aps_subpage[
4355                                        CTL_PAGE_CURRENT],
4356                                        &aps_page_default,
4357                                        sizeof(aps_page_default));
4358                                 memcpy(&lun->mode_pages.aps_subpage[
4359                                        CTL_PAGE_CHANGEABLE],
4360                                        &aps_page_changeable,
4361                                        sizeof(aps_page_changeable));
4362                                 memcpy(&lun->mode_pages.aps_subpage[
4363                                        CTL_PAGE_DEFAULT],
4364                                        &aps_page_default,
4365                                        sizeof(aps_page_default));
4366                                 memcpy(&lun->mode_pages.aps_subpage[
4367                                        CTL_PAGE_SAVED],
4368                                        &aps_page_default,
4369                                        sizeof(aps_page_default));
4370                                 page_index->page_data =
4371                                         (uint8_t *)lun->mode_pages.aps_subpage;
4372
4373                                 current_page = (struct copan_aps_subpage *)
4374                                         (page_index->page_data +
4375                                          (page_index->page_len *
4376                                           CTL_PAGE_CURRENT));
4377                                 saved_page = (struct copan_aps_subpage *)
4378                                         (page_index->page_data +
4379                                          (page_index->page_len *
4380                                           CTL_PAGE_SAVED));
4381                                 break;
4382                         }
4383                         case DBGCNF_SUBPAGE_CODE: {
4384                                 struct copan_debugconf_subpage *current_page,
4385                                                                *saved_page;
4386
4387                                 memcpy(&lun->mode_pages.debugconf_subpage[
4388                                        CTL_PAGE_CURRENT],
4389                                        &debugconf_page_default,
4390                                        sizeof(debugconf_page_default));
4391                                 memcpy(&lun->mode_pages.debugconf_subpage[
4392                                        CTL_PAGE_CHANGEABLE],
4393                                        &debugconf_page_changeable,
4394                                        sizeof(debugconf_page_changeable));
4395                                 memcpy(&lun->mode_pages.debugconf_subpage[
4396                                        CTL_PAGE_DEFAULT],
4397                                        &debugconf_page_default,
4398                                        sizeof(debugconf_page_default));
4399                                 memcpy(&lun->mode_pages.debugconf_subpage[
4400                                        CTL_PAGE_SAVED],
4401                                        &debugconf_page_default,
4402                                        sizeof(debugconf_page_default));
4403                                 page_index->page_data =
4404                                         (uint8_t *)lun->mode_pages.debugconf_subpage;
4405
4406                                 current_page = (struct copan_debugconf_subpage *)
4407                                         (page_index->page_data +
4408                                          (page_index->page_len *
4409                                           CTL_PAGE_CURRENT));
4410                                 saved_page = (struct copan_debugconf_subpage *)
4411                                         (page_index->page_data +
4412                                          (page_index->page_len *
4413                                           CTL_PAGE_SAVED));
4414                                 break;
4415                         }
4416                         default:
4417                                 panic("invalid subpage value %d",
4418                                       page_index->subpage);
4419                                 break;
4420                         }
4421                         break;
4422                 }
4423                 default:
4424                         panic("invalid page value %d",
4425                               page_index->page_code & SMPH_PC_MASK);
4426                         break;
4427         }
4428         }
4429
4430         return (CTL_RETVAL_COMPLETE);
4431 }
4432
4433 /*
4434  * LUN allocation.
4435  *
4436  * Requirements:
4437  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4438  *   wants us to allocate the LUN and he can block.
4439  * - ctl_softc is always set
4440  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4441  *
4442  * Returns 0 for success, non-zero (errno) for failure.
4443  */
4444 static int
4445 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4446               struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4447 {
4448         struct ctl_lun *nlun, *lun;
4449         struct ctl_port *port;
4450         struct scsi_vpd_id_descriptor *desc;
4451         struct scsi_vpd_id_t10 *t10id;
4452         const char *eui, *naa, *scsiname, *vendor;
4453         int lun_number, i, lun_malloced;
4454         int devidlen, idlen1, idlen2 = 0, len;
4455
4456         if (be_lun == NULL)
4457                 return (EINVAL);
4458
4459         /*
4460          * We currently only support Direct Access or Processor LUN types.
4461          */
4462         switch (be_lun->lun_type) {
4463         case T_DIRECT:
4464                 break;
4465         case T_PROCESSOR:
4466                 break;
4467         case T_SEQUENTIAL:
4468         case T_CHANGER:
4469         default:
4470                 be_lun->lun_config_status(be_lun->be_lun,
4471                                           CTL_LUN_CONFIG_FAILURE);
4472                 break;
4473         }
4474         if (ctl_lun == NULL) {
4475                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4476                 lun_malloced = 1;
4477         } else {
4478                 lun_malloced = 0;
4479                 lun = ctl_lun;
4480         }
4481
4482         memset(lun, 0, sizeof(*lun));
4483         if (lun_malloced)
4484                 lun->flags = CTL_LUN_MALLOCED;
4485
4486         /* Generate LUN ID. */
4487         devidlen = max(CTL_DEVID_MIN_LEN,
4488             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4489         idlen1 = sizeof(*t10id) + devidlen;
4490         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4491         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4492         if (scsiname != NULL) {
4493                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4494                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4495         }
4496         eui = ctl_get_opt(&be_lun->options, "eui");
4497         if (eui != NULL) {
4498                 len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4499         }
4500         naa = ctl_get_opt(&be_lun->options, "naa");
4501         if (naa != NULL) {
4502                 len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4503         }
4504         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4505             M_CTL, M_WAITOK | M_ZERO);
4506         lun->lun_devid->len = len;
4507         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4508         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4509         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4510         desc->length = idlen1;
4511         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4512         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4513         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4514                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4515         } else {
4516                 strncpy(t10id->vendor, vendor,
4517                     min(sizeof(t10id->vendor), strlen(vendor)));
4518         }
4519         strncpy((char *)t10id->vendor_spec_id,
4520             (char *)be_lun->device_id, devidlen);
4521         if (scsiname != NULL) {
4522                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4523                     desc->length);
4524                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4525                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4526                     SVPD_ID_TYPE_SCSI_NAME;
4527                 desc->length = idlen2;
4528                 strlcpy(desc->identifier, scsiname, idlen2);
4529         }
4530         if (eui != NULL) {
4531                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4532                     desc->length);
4533                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4534                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4535                     SVPD_ID_TYPE_EUI64;
4536                 desc->length = 8;
4537                 scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4538         }
4539         if (naa != NULL) {
4540                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4541                     desc->length);
4542                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4543                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4544                     SVPD_ID_TYPE_NAA;
4545                 desc->length = 8;
4546                 scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4547         }
4548
4549         mtx_lock(&ctl_softc->ctl_lock);
4550         /*
4551          * See if the caller requested a particular LUN number.  If so, see
4552          * if it is available.  Otherwise, allocate the first available LUN.
4553          */
4554         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4555                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4556                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4557                         mtx_unlock(&ctl_softc->ctl_lock);
4558                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4559                                 printf("ctl: requested LUN ID %d is higher "
4560                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4561                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4562                         } else {
4563                                 /*
4564                                  * XXX KDM return an error, or just assign
4565                                  * another LUN ID in this case??
4566                                  */
4567                                 printf("ctl: requested LUN ID %d is already "
4568                                        "in use\n", be_lun->req_lun_id);
4569                         }
4570                         if (lun->flags & CTL_LUN_MALLOCED)
4571                                 free(lun, M_CTL);
4572                         be_lun->lun_config_status(be_lun->be_lun,
4573                                                   CTL_LUN_CONFIG_FAILURE);
4574                         return (ENOSPC);
4575                 }
4576                 lun_number = be_lun->req_lun_id;
4577         } else {
4578                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4579                 if (lun_number == -1) {
4580                         mtx_unlock(&ctl_softc->ctl_lock);
4581                         printf("ctl: can't allocate LUN on target %ju, out of "
4582                                "LUNs\n", (uintmax_t)target_id.id);
4583                         if (lun->flags & CTL_LUN_MALLOCED)
4584                                 free(lun, M_CTL);
4585                         be_lun->lun_config_status(be_lun->be_lun,
4586                                                   CTL_LUN_CONFIG_FAILURE);
4587                         return (ENOSPC);
4588                 }
4589         }
4590         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4591
4592         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4593         lun->target = target_id;
4594         lun->lun = lun_number;
4595         lun->be_lun = be_lun;
4596         /*
4597          * The processor LUN is always enabled.  Disk LUNs come on line
4598          * disabled, and must be enabled by the backend.
4599          */
4600         lun->flags |= CTL_LUN_DISABLED;
4601         lun->backend = be_lun->be;
4602         be_lun->ctl_lun = lun;
4603         be_lun->lun_id = lun_number;
4604         atomic_add_int(&be_lun->be->num_luns, 1);
4605         if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4606                 lun->flags |= CTL_LUN_STOPPED;
4607
4608         if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4609                 lun->flags |= CTL_LUN_INOPERABLE;
4610
4611         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4612                 lun->flags |= CTL_LUN_PRIMARY_SC;
4613
4614         lun->ctl_softc = ctl_softc;
4615         TAILQ_INIT(&lun->ooa_queue);
4616         TAILQ_INIT(&lun->blocked_queue);
4617         STAILQ_INIT(&lun->error_list);
4618         ctl_tpc_lun_init(lun);
4619
4620         /*
4621          * Initialize the mode page index.
4622          */
4623         ctl_init_page_index(lun);
4624
4625         /*
4626          * Set the poweron UA for all initiators on this LUN only.
4627          */
4628         for (i = 0; i < CTL_MAX_INITIATORS; i++)
4629                 lun->pending_ua[i] = CTL_UA_POWERON;
4630
4631         /*
4632          * Now, before we insert this lun on the lun list, set the lun
4633          * inventory changed UA for all other luns.
4634          */
4635         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4636                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4637                         nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4638                 }
4639         }
4640
4641         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4642
4643         ctl_softc->ctl_luns[lun_number] = lun;
4644
4645         ctl_softc->num_luns++;
4646
4647         /* Setup statistics gathering */
4648         lun->stats.device_type = be_lun->lun_type;
4649         lun->stats.lun_number = lun_number;
4650         if (lun->stats.device_type == T_DIRECT)
4651                 lun->stats.blocksize = be_lun->blocksize;
4652         else
4653                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4654         for (i = 0;i < CTL_MAX_PORTS;i++)
4655                 lun->stats.ports[i].targ_port = i;
4656
4657         mtx_unlock(&ctl_softc->ctl_lock);
4658
4659         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4660
4661         /*
4662          * Run through each registered FETD and bring it online if it isn't
4663          * already.  Enable the target ID if it hasn't been enabled, and
4664          * enable this particular LUN.
4665          */
4666         STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4667                 int retval;
4668
4669                 retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4670                 if (retval != 0) {
4671                         printf("ctl_alloc_lun: FETD %s port %d returned error "
4672                                "%d for lun_enable on target %ju lun %d\n",
4673                                port->port_name, port->targ_port, retval,
4674                                (uintmax_t)target_id.id, lun_number);
4675                 } else
4676                         port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4677         }
4678         return (0);
4679 }
4680
4681 /*
4682  * Delete a LUN.
4683  * Assumptions:
4684  * - LUN has already been marked invalid and any pending I/O has been taken
4685  *   care of.
4686  */
4687 static int
4688 ctl_free_lun(struct ctl_lun *lun)
4689 {
4690         struct ctl_softc *softc;
4691 #if 0
4692         struct ctl_port *port;
4693 #endif
4694         struct ctl_lun *nlun;
4695         int i;
4696
4697         softc = lun->ctl_softc;
4698
4699         mtx_assert(&softc->ctl_lock, MA_OWNED);
4700
4701         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4702
4703         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4704
4705         softc->ctl_luns[lun->lun] = NULL;
4706
4707         if (!TAILQ_EMPTY(&lun->ooa_queue))
4708                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4709
4710         softc->num_luns--;
4711
4712         /*
4713          * XXX KDM this scheme only works for a single target/multiple LUN
4714          * setup.  It needs to be revamped for a multiple target scheme.
4715          *
4716          * XXX KDM this results in port->lun_disable() getting called twice,
4717          * once when ctl_disable_lun() is called, and a second time here.
4718          * We really need to re-think the LUN disable semantics.  There
4719          * should probably be several steps/levels to LUN removal:
4720          *  - disable
4721          *  - invalidate
4722          *  - free
4723          *
4724          * Right now we only have a disable method when communicating to
4725          * the front end ports, at least for individual LUNs.
4726          */
4727 #if 0
4728         STAILQ_FOREACH(port, &softc->port_list, links) {
4729                 int retval;
4730
4731                 retval = port->lun_disable(port->targ_lun_arg, lun->target,
4732                                          lun->lun);
4733                 if (retval != 0) {
4734                         printf("ctl_free_lun: FETD %s port %d returned error "
4735                                "%d for lun_disable on target %ju lun %jd\n",
4736                                port->port_name, port->targ_port, retval,
4737                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4738                 }
4739
4740                 if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4741                         port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4742
4743                         retval = port->targ_disable(port->targ_lun_arg,lun->target);
4744                         if (retval != 0) {
4745                                 printf("ctl_free_lun: FETD %s port %d "
4746                                        "returned error %d for targ_disable on "
4747                                        "target %ju\n", port->port_name,
4748                                        port->targ_port, retval,
4749                                        (uintmax_t)lun->target.id);
4750                         } else
4751                                 port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4752
4753                         if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4754                                 continue;
4755
4756 #if 0
4757                         port->port_offline(port->onoff_arg);
4758                         port->status &= ~CTL_PORT_STATUS_ONLINE;
4759 #endif
4760                 }
4761         }
4762 #endif
4763
4764         /*
4765          * Tell the backend to free resources, if this LUN has a backend.
4766          */
4767         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4768         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4769
4770         ctl_tpc_lun_shutdown(lun);
4771         mtx_destroy(&lun->lun_lock);
4772         free(lun->lun_devid, M_CTL);
4773         if (lun->flags & CTL_LUN_MALLOCED)
4774                 free(lun, M_CTL);
4775
4776         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4777                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4778                         nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4779                 }
4780         }
4781
4782         return (0);
4783 }
4784
4785 static void
4786 ctl_create_lun(struct ctl_be_lun *be_lun)
4787 {
4788         struct ctl_softc *ctl_softc;
4789
4790         ctl_softc = control_softc;
4791
4792         /*
4793          * ctl_alloc_lun() should handle all potential failure cases.
4794          */
4795         ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4796 }
4797
4798 int
4799 ctl_add_lun(struct ctl_be_lun *be_lun)
4800 {
4801         struct ctl_softc *ctl_softc = control_softc;
4802
4803         mtx_lock(&ctl_softc->ctl_lock);
4804         STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4805         mtx_unlock(&ctl_softc->ctl_lock);
4806         wakeup(&ctl_softc->pending_lun_queue);
4807
4808         return (0);
4809 }
4810
4811 int
4812 ctl_enable_lun(struct ctl_be_lun *be_lun)
4813 {
4814         struct ctl_softc *ctl_softc;
4815         struct ctl_port *port, *nport;
4816         struct ctl_lun *lun;
4817         int retval;
4818
4819         ctl_softc = control_softc;
4820
4821         lun = (struct ctl_lun *)be_lun->ctl_lun;
4822
4823         mtx_lock(&ctl_softc->ctl_lock);
4824         mtx_lock(&lun->lun_lock);
4825         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4826                 /*
4827                  * eh?  Why did we get called if the LUN is already
4828                  * enabled?
4829                  */
4830                 mtx_unlock(&lun->lun_lock);
4831                 mtx_unlock(&ctl_softc->ctl_lock);
4832                 return (0);
4833         }
4834         lun->flags &= ~CTL_LUN_DISABLED;
4835         mtx_unlock(&lun->lun_lock);
4836
4837         for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4838                 nport = STAILQ_NEXT(port, links);
4839
4840                 /*
4841                  * Drop the lock while we call the FETD's enable routine.
4842                  * This can lead to a callback into CTL (at least in the
4843                  * case of the internal initiator frontend.
4844                  */
4845                 mtx_unlock(&ctl_softc->ctl_lock);
4846                 retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4847                 mtx_lock(&ctl_softc->ctl_lock);
4848                 if (retval != 0) {
4849                         printf("%s: FETD %s port %d returned error "
4850                                "%d for lun_enable on target %ju lun %jd\n",
4851                                __func__, port->port_name, port->targ_port, retval,
4852                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4853                 }
4854 #if 0
4855                  else {
4856             /* NOTE:  TODO:  why does lun enable affect port status? */
4857                         port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4858                 }
4859 #endif
4860         }
4861
4862         mtx_unlock(&ctl_softc->ctl_lock);
4863
4864         return (0);
4865 }
4866
4867 int
4868 ctl_disable_lun(struct ctl_be_lun *be_lun)
4869 {
4870         struct ctl_softc *ctl_softc;
4871         struct ctl_port *port;
4872         struct ctl_lun *lun;
4873         int retval;
4874
4875         ctl_softc = control_softc;
4876
4877         lun = (struct ctl_lun *)be_lun->ctl_lun;
4878
4879         mtx_lock(&ctl_softc->ctl_lock);
4880         mtx_lock(&lun->lun_lock);
4881         if (lun->flags & CTL_LUN_DISABLED) {
4882                 mtx_unlock(&lun->lun_lock);
4883                 mtx_unlock(&ctl_softc->ctl_lock);
4884                 return (0);
4885         }
4886         lun->flags |= CTL_LUN_DISABLED;
4887         mtx_unlock(&lun->lun_lock);
4888
4889         STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4890                 mtx_unlock(&ctl_softc->ctl_lock);
4891                 /*
4892                  * Drop the lock before we call the frontend's disable
4893                  * routine, to avoid lock order reversals.
4894                  *
4895                  * XXX KDM what happens if the frontend list changes while
4896                  * we're traversing it?  It's unlikely, but should be handled.
4897                  */
4898                 retval = port->lun_disable(port->targ_lun_arg, lun->target,
4899                                          lun->lun);
4900                 mtx_lock(&ctl_softc->ctl_lock);
4901                 if (retval != 0) {
4902                         printf("ctl_alloc_lun: FETD %s port %d returned error "
4903                                "%d for lun_disable on target %ju lun %jd\n",
4904                                port->port_name, port->targ_port, retval,
4905                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4906                 }
4907         }
4908
4909         mtx_unlock(&ctl_softc->ctl_lock);
4910
4911         return (0);
4912 }
4913
4914 int
4915 ctl_start_lun(struct ctl_be_lun *be_lun)
4916 {
4917         struct ctl_softc *ctl_softc;
4918         struct ctl_lun *lun;
4919
4920         ctl_softc = control_softc;
4921
4922         lun = (struct ctl_lun *)be_lun->ctl_lun;
4923
4924         mtx_lock(&lun->lun_lock);
4925         lun->flags &= ~CTL_LUN_STOPPED;
4926         mtx_unlock(&lun->lun_lock);
4927
4928         return (0);
4929 }
4930
4931 int
4932 ctl_stop_lun(struct ctl_be_lun *be_lun)
4933 {
4934         struct ctl_softc *ctl_softc;
4935         struct ctl_lun *lun;
4936
4937         ctl_softc = control_softc;
4938
4939         lun = (struct ctl_lun *)be_lun->ctl_lun;
4940
4941         mtx_lock(&lun->lun_lock);
4942         lun->flags |= CTL_LUN_STOPPED;
4943         mtx_unlock(&lun->lun_lock);
4944
4945         return (0);
4946 }
4947
4948 int
4949 ctl_lun_offline(struct ctl_be_lun *be_lun)
4950 {
4951         struct ctl_softc *ctl_softc;
4952         struct ctl_lun *lun;
4953
4954         ctl_softc = control_softc;
4955
4956         lun = (struct ctl_lun *)be_lun->ctl_lun;
4957
4958         mtx_lock(&lun->lun_lock);
4959         lun->flags |= CTL_LUN_OFFLINE;
4960         mtx_unlock(&lun->lun_lock);
4961
4962         return (0);
4963 }
4964
4965 int
4966 ctl_lun_online(struct ctl_be_lun *be_lun)
4967 {
4968         struct ctl_softc *ctl_softc;
4969         struct ctl_lun *lun;
4970
4971         ctl_softc = control_softc;
4972
4973         lun = (struct ctl_lun *)be_lun->ctl_lun;
4974
4975         mtx_lock(&lun->lun_lock);
4976         lun->flags &= ~CTL_LUN_OFFLINE;
4977         mtx_unlock(&lun->lun_lock);
4978
4979         return (0);
4980 }
4981
4982 int
4983 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4984 {
4985         struct ctl_softc *ctl_softc;
4986         struct ctl_lun *lun;
4987
4988         ctl_softc = control_softc;
4989
4990         lun = (struct ctl_lun *)be_lun->ctl_lun;
4991
4992         mtx_lock(&lun->lun_lock);
4993
4994         /*
4995          * The LUN needs to be disabled before it can be marked invalid.
4996          */
4997         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4998                 mtx_unlock(&lun->lun_lock);
4999                 return (-1);
5000         }
5001         /*
5002          * Mark the LUN invalid.
5003          */
5004         lun->flags |= CTL_LUN_INVALID;
5005
5006         /*
5007          * If there is nothing in the OOA queue, go ahead and free the LUN.
5008          * If we have something in the OOA queue, we'll free it when the
5009          * last I/O completes.
5010          */
5011         if (TAILQ_EMPTY(&lun->ooa_queue)) {
5012                 mtx_unlock(&lun->lun_lock);
5013                 mtx_lock(&ctl_softc->ctl_lock);
5014                 ctl_free_lun(lun);
5015                 mtx_unlock(&ctl_softc->ctl_lock);
5016         } else
5017                 mtx_unlock(&lun->lun_lock);
5018
5019         return (0);
5020 }
5021
5022 int
5023 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5024 {
5025         struct ctl_softc *ctl_softc;
5026         struct ctl_lun *lun;
5027
5028         ctl_softc = control_softc;
5029         lun = (struct ctl_lun *)be_lun->ctl_lun;
5030
5031         mtx_lock(&lun->lun_lock);
5032         lun->flags |= CTL_LUN_INOPERABLE;
5033         mtx_unlock(&lun->lun_lock);
5034
5035         return (0);
5036 }
5037
5038 int
5039 ctl_lun_operable(struct ctl_be_lun *be_lun)
5040 {
5041         struct ctl_softc *ctl_softc;
5042         struct ctl_lun *lun;
5043
5044         ctl_softc = control_softc;
5045         lun = (struct ctl_lun *)be_lun->ctl_lun;
5046
5047         mtx_lock(&lun->lun_lock);
5048         lun->flags &= ~CTL_LUN_INOPERABLE;
5049         mtx_unlock(&lun->lun_lock);
5050
5051         return (0);
5052 }
5053
5054 int
5055 ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5056                    int lock)
5057 {
5058         struct ctl_softc *softc;
5059         struct ctl_lun *lun;
5060         struct copan_aps_subpage *current_sp;
5061         struct ctl_page_index *page_index;
5062         int i;
5063
5064         softc = control_softc;
5065
5066         mtx_lock(&softc->ctl_lock);
5067
5068         lun = (struct ctl_lun *)be_lun->ctl_lun;
5069         mtx_lock(&lun->lun_lock);
5070
5071         page_index = NULL;
5072         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5073                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5074                      APS_PAGE_CODE)
5075                         continue;
5076
5077                 if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5078                         continue;
5079                 page_index = &lun->mode_pages.index[i];
5080         }
5081
5082         if (page_index == NULL) {
5083                 mtx_unlock(&lun->lun_lock);
5084                 mtx_unlock(&softc->ctl_lock);
5085                 printf("%s: APS subpage not found for lun %ju!\n", __func__,
5086                        (uintmax_t)lun->lun);
5087                 return (1);
5088         }
5089 #if 0
5090         if ((softc->aps_locked_lun != 0)
5091          && (softc->aps_locked_lun != lun->lun)) {
5092                 printf("%s: attempt to lock LUN %llu when %llu is already "
5093                        "locked\n");
5094                 mtx_unlock(&lun->lun_lock);
5095                 mtx_unlock(&softc->ctl_lock);
5096                 return (1);
5097         }
5098 #endif
5099
5100         current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5101                 (page_index->page_len * CTL_PAGE_CURRENT));
5102
5103         if (lock != 0) {
5104                 current_sp->lock_active = APS_LOCK_ACTIVE;
5105                 softc->aps_locked_lun = lun->lun;
5106         } else {
5107                 current_sp->lock_active = 0;
5108                 softc->aps_locked_lun = 0;
5109         }
5110
5111
5112         /*
5113          * If we're in HA mode, try to send the lock message to the other
5114          * side.
5115          */
5116         if (ctl_is_single == 0) {
5117                 int isc_retval;
5118                 union ctl_ha_msg lock_msg;
5119
5120                 lock_msg.hdr.nexus = *nexus;
5121                 lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5122                 if (lock != 0)
5123                         lock_msg.aps.lock_flag = 1;
5124                 else
5125                         lock_msg.aps.lock_flag = 0;
5126                 isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5127                                          sizeof(lock_msg), 0);
5128                 if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5129                         printf("%s: APS (lock=%d) error returned from "
5130                                "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5131                         mtx_unlock(&lun->lun_lock);
5132                         mtx_unlock(&softc->ctl_lock);
5133                         return (1);
5134                 }
5135         }
5136
5137         mtx_unlock(&lun->lun_lock);
5138         mtx_unlock(&softc->ctl_lock);
5139
5140         return (0);
5141 }
5142
5143 void
5144 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5145 {
5146         struct ctl_lun *lun;
5147         struct ctl_softc *softc;
5148         int i;
5149
5150         softc = control_softc;
5151
5152         lun = (struct ctl_lun *)be_lun->ctl_lun;
5153
5154         mtx_lock(&lun->lun_lock);
5155
5156         for (i = 0; i < CTL_MAX_INITIATORS; i++)
5157                 lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5158
5159         mtx_unlock(&lun->lun_lock);
5160 }
5161
5162 /*
5163  * Backend "memory move is complete" callback for requests that never
5164  * make it down to say RAIDCore's configuration code.
5165  */
5166 int
5167 ctl_config_move_done(union ctl_io *io)
5168 {
5169         int retval;
5170
5171         retval = CTL_RETVAL_COMPLETE;
5172
5173
5174         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5175         /*
5176          * XXX KDM this shouldn't happen, but what if it does?
5177          */
5178         if (io->io_hdr.io_type != CTL_IO_SCSI)
5179                 panic("I/O type isn't CTL_IO_SCSI!");
5180
5181         if ((io->io_hdr.port_status == 0)
5182          && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5183          && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5184                 io->io_hdr.status = CTL_SUCCESS;
5185         else if ((io->io_hdr.port_status != 0)
5186               && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5187               && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5188                 /*
5189                  * For hardware error sense keys, the sense key
5190                  * specific value is defined to be a retry count,
5191                  * but we use it to pass back an internal FETD
5192                  * error code.  XXX KDM  Hopefully the FETD is only
5193                  * using 16 bits for an error code, since that's
5194                  * all the space we have in the sks field.
5195                  */
5196                 ctl_set_internal_failure(&io->scsiio,
5197                                          /*sks_valid*/ 1,
5198                                          /*retry_count*/
5199                                          io->io_hdr.port_status);
5200                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5201                         free(io->scsiio.kern_data_ptr, M_CTL);
5202                 ctl_done(io);
5203                 goto bailout;
5204         }
5205
5206         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5207          || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5208          || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5209                 /*
5210                  * XXX KDM just assuming a single pointer here, and not a
5211                  * S/G list.  If we start using S/G lists for config data,
5212                  * we'll need to know how to clean them up here as well.
5213                  */
5214                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5215                         free(io->scsiio.kern_data_ptr, M_CTL);
5216                 /* Hopefully the user has already set the status... */
5217                 ctl_done(io);
5218         } else {
5219                 /*
5220                  * XXX KDM now we need to continue data movement.  Some
5221                  * options:
5222                  * - call ctl_scsiio() again?  We don't do this for data
5223                  *   writes, because for those at least we know ahead of
5224                  *   time where the write will go and how long it is.  For
5225                  *   config writes, though, that information is largely
5226                  *   contained within the write itself, thus we need to
5227                  *   parse out the data again.
5228                  *
5229                  * - Call some other function once the data is in?
5230                  */
5231
5232                 /*
5233                  * XXX KDM call ctl_scsiio() again for now, and check flag
5234                  * bits to see whether we're allocated or not.
5235                  */
5236                 retval = ctl_scsiio(&io->scsiio);
5237         }
5238 bailout:
5239         return (retval);
5240 }
5241
5242 /*
5243  * This gets called by a backend driver when it is done with a
5244  * data_submit method.
5245  */
5246 void
5247 ctl_data_submit_done(union ctl_io *io)
5248 {
5249         /*
5250          * If the IO_CONT flag is set, we need to call the supplied
5251          * function to continue processing the I/O, instead of completing
5252          * the I/O just yet.
5253          *
5254          * If there is an error, though, we don't want to keep processing.
5255          * Instead, just send status back to the initiator.
5256          */
5257         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5258             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5259             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5260              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5261                 io->scsiio.io_cont(io);
5262                 return;
5263         }
5264         ctl_done(io);
5265 }
5266
5267 /*
5268  * This gets called by a backend driver when it is done with a
5269  * configuration write.
5270  */
5271 void
5272 ctl_config_write_done(union ctl_io *io)
5273 {
5274         uint8_t *buf;
5275
5276         /*
5277          * If the IO_CONT flag is set, we need to call the supplied
5278          * function to continue processing the I/O, instead of completing
5279          * the I/O just yet.
5280          *
5281          * If there is an error, though, we don't want to keep processing.
5282          * Instead, just send status back to the initiator.
5283          */
5284         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5285             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5286             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5287              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5288                 io->scsiio.io_cont(io);
5289                 return;
5290         }
5291         /*
5292          * Since a configuration write can be done for commands that actually
5293          * have data allocated, like write buffer, and commands that have
5294          * no data, like start/stop unit, we need to check here.
5295          */
5296         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5297                 buf = io->scsiio.kern_data_ptr;
5298         else
5299                 buf = NULL;
5300         ctl_done(io);
5301         if (buf)
5302                 free(buf, M_CTL);
5303 }
5304
5305 /*
5306  * SCSI release command.
5307  */
5308 int
5309 ctl_scsi_release(struct ctl_scsiio *ctsio)
5310 {
5311         int length, longid, thirdparty_id, resv_id;
5312         struct ctl_softc *ctl_softc;
5313         struct ctl_lun *lun;
5314         uint32_t residx;
5315
5316         length = 0;
5317         resv_id = 0;
5318
5319         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5320
5321         residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5322         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5323         ctl_softc = control_softc;
5324
5325         switch (ctsio->cdb[0]) {
5326         case RELEASE_10: {
5327                 struct scsi_release_10 *cdb;
5328
5329                 cdb = (struct scsi_release_10 *)ctsio->cdb;
5330
5331                 if (cdb->byte2 & SR10_LONGID)
5332                         longid = 1;
5333                 else
5334                         thirdparty_id = cdb->thirdparty_id;
5335
5336                 resv_id = cdb->resv_id;
5337                 length = scsi_2btoul(cdb->length);
5338                 break;
5339         }
5340         }
5341
5342
5343         /*
5344          * XXX KDM right now, we only support LUN reservation.  We don't
5345          * support 3rd party reservations, or extent reservations, which
5346          * might actually need the parameter list.  If we've gotten this
5347          * far, we've got a LUN reservation.  Anything else got kicked out
5348          * above.  So, according to SPC, ignore the length.
5349          */
5350         length = 0;
5351
5352         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5353          && (length > 0)) {
5354                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5355                 ctsio->kern_data_len = length;
5356                 ctsio->kern_total_len = length;
5357                 ctsio->kern_data_resid = 0;
5358                 ctsio->kern_rel_offset = 0;
5359                 ctsio->kern_sg_entries = 0;
5360                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5361                 ctsio->be_move_done = ctl_config_move_done;
5362                 ctl_datamove((union ctl_io *)ctsio);
5363
5364                 return (CTL_RETVAL_COMPLETE);
5365         }
5366
5367         if (length > 0)
5368                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5369
5370         mtx_lock(&lun->lun_lock);
5371
5372         /*
5373          * According to SPC, it is not an error for an intiator to attempt
5374          * to release a reservation on a LUN that isn't reserved, or that
5375          * is reserved by another initiator.  The reservation can only be
5376          * released, though, by the initiator who made it or by one of
5377          * several reset type events.
5378          */
5379         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5380                         lun->flags &= ~CTL_LUN_RESERVED;
5381
5382         mtx_unlock(&lun->lun_lock);
5383
5384         ctsio->scsi_status = SCSI_STATUS_OK;
5385         ctsio->io_hdr.status = CTL_SUCCESS;
5386
5387         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5388                 free(ctsio->kern_data_ptr, M_CTL);
5389                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5390         }
5391
5392         ctl_done((union ctl_io *)ctsio);
5393         return (CTL_RETVAL_COMPLETE);
5394 }
5395
5396 int
5397 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5398 {
5399         int extent, thirdparty, longid;
5400         int resv_id, length;
5401         uint64_t thirdparty_id;
5402         struct ctl_softc *ctl_softc;
5403         struct ctl_lun *lun;
5404         uint32_t residx;
5405
5406         extent = 0;
5407         thirdparty = 0;
5408         longid = 0;
5409         resv_id = 0;
5410         length = 0;
5411         thirdparty_id = 0;
5412
5413         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5414
5415         residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5416         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5417         ctl_softc = control_softc;
5418
5419         switch (ctsio->cdb[0]) {
5420         case RESERVE_10: {
5421                 struct scsi_reserve_10 *cdb;
5422
5423                 cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5424
5425                 if (cdb->byte2 & SR10_LONGID)
5426                         longid = 1;
5427                 else
5428                         thirdparty_id = cdb->thirdparty_id;
5429
5430                 resv_id = cdb->resv_id;
5431                 length = scsi_2btoul(cdb->length);
5432                 break;
5433         }
5434         }
5435
5436         /*
5437          * XXX KDM right now, we only support LUN reservation.  We don't
5438          * support 3rd party reservations, or extent reservations, which
5439          * might actually need the parameter list.  If we've gotten this
5440          * far, we've got a LUN reservation.  Anything else got kicked out
5441          * above.  So, according to SPC, ignore the length.
5442          */
5443         length = 0;
5444
5445         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5446          && (length > 0)) {
5447                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5448                 ctsio->kern_data_len = length;
5449                 ctsio->kern_total_len = length;
5450                 ctsio->kern_data_resid = 0;
5451                 ctsio->kern_rel_offset = 0;
5452                 ctsio->kern_sg_entries = 0;
5453                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5454                 ctsio->be_move_done = ctl_config_move_done;
5455                 ctl_datamove((union ctl_io *)ctsio);
5456
5457                 return (CTL_RETVAL_COMPLETE);
5458         }
5459
5460         if (length > 0)
5461                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5462
5463         mtx_lock(&lun->lun_lock);
5464         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5465                 ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5466                 ctsio->io_hdr.status = CTL_SCSI_ERROR;
5467                 goto bailout;
5468         }
5469
5470         lun->flags |= CTL_LUN_RESERVED;
5471         lun->res_idx = residx;
5472
5473         ctsio->scsi_status = SCSI_STATUS_OK;
5474         ctsio->io_hdr.status = CTL_SUCCESS;
5475
5476 bailout:
5477         mtx_unlock(&lun->lun_lock);
5478
5479         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5480                 free(ctsio->kern_data_ptr, M_CTL);
5481                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5482         }
5483
5484         ctl_done((union ctl_io *)ctsio);
5485         return (CTL_RETVAL_COMPLETE);
5486 }
5487
5488 int
5489 ctl_start_stop(struct ctl_scsiio *ctsio)
5490 {
5491         struct scsi_start_stop_unit *cdb;
5492         struct ctl_lun *lun;
5493         struct ctl_softc *ctl_softc;
5494         int retval;
5495
5496         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5497
5498         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5499         ctl_softc = control_softc;
5500         retval = 0;
5501
5502         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5503
5504         /*
5505          * XXX KDM
5506          * We don't support the immediate bit on a stop unit.  In order to
5507          * do that, we would need to code up a way to know that a stop is
5508          * pending, and hold off any new commands until it completes, one
5509          * way or another.  Then we could accept or reject those commands
5510          * depending on its status.  We would almost need to do the reverse
5511          * of what we do below for an immediate start -- return the copy of
5512          * the ctl_io to the FETD with status to send to the host (and to
5513          * free the copy!) and then free the original I/O once the stop
5514          * actually completes.  That way, the OOA queue mechanism can work
5515          * to block commands that shouldn't proceed.  Another alternative
5516          * would be to put the copy in the queue in place of the original,
5517          * and return the original back to the caller.  That could be
5518          * slightly safer..
5519          */
5520         if ((cdb->byte2 & SSS_IMMED)
5521          && ((cdb->how & SSS_START) == 0)) {
5522                 ctl_set_invalid_field(ctsio,
5523                                       /*sks_valid*/ 1,
5524                                       /*command*/ 1,
5525                                       /*field*/ 1,
5526                                       /*bit_valid*/ 1,
5527                                       /*bit*/ 0);
5528                 ctl_done((union ctl_io *)ctsio);
5529                 return (CTL_RETVAL_COMPLETE);
5530         }
5531
5532         if ((lun->flags & CTL_LUN_PR_RESERVED)
5533          && ((cdb->how & SSS_START)==0)) {
5534                 uint32_t residx;
5535
5536                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5537                 if (!lun->per_res[residx].registered
5538                  || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5539
5540                         ctl_set_reservation_conflict(ctsio);
5541                         ctl_done((union ctl_io *)ctsio);
5542                         return (CTL_RETVAL_COMPLETE);
5543                 }
5544         }
5545
5546         /*
5547          * If there is no backend on this device, we can't start or stop
5548          * it.  In theory we shouldn't get any start/stop commands in the
5549          * first place at this level if the LUN doesn't have a backend.
5550          * That should get stopped by the command decode code.
5551          */
5552         if (lun->backend == NULL) {
5553                 ctl_set_invalid_opcode(ctsio);
5554                 ctl_done((union ctl_io *)ctsio);
5555                 return (CTL_RETVAL_COMPLETE);
5556         }
5557
5558         /*
5559          * XXX KDM Copan-specific offline behavior.
5560          * Figure out a reasonable way to port this?
5561          */
5562 #ifdef NEEDTOPORT
5563         mtx_lock(&lun->lun_lock);
5564
5565         if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5566          && (lun->flags & CTL_LUN_OFFLINE)) {
5567                 /*
5568                  * If the LUN is offline, and the on/offline bit isn't set,
5569                  * reject the start or stop.  Otherwise, let it through.
5570                  */
5571                 mtx_unlock(&lun->lun_lock);
5572                 ctl_set_lun_not_ready(ctsio);
5573                 ctl_done((union ctl_io *)ctsio);
5574         } else {
5575                 mtx_unlock(&lun->lun_lock);
5576 #endif /* NEEDTOPORT */
5577                 /*
5578                  * This could be a start or a stop when we're online,
5579                  * or a stop/offline or start/online.  A start or stop when
5580                  * we're offline is covered in the case above.
5581                  */
5582                 /*
5583                  * In the non-immediate case, we send the request to
5584                  * the backend and return status to the user when
5585                  * it is done.
5586                  *
5587                  * In the immediate case, we allocate a new ctl_io
5588                  * to hold a copy of the request, and send that to
5589                  * the backend.  We then set good status on the
5590                  * user's request and return it immediately.
5591                  */
5592                 if (cdb->byte2 & SSS_IMMED) {
5593                         union ctl_io *new_io;
5594
5595                         new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5596                         if (new_io == NULL) {
5597                                 ctl_set_busy(ctsio);
5598                                 ctl_done((union ctl_io *)ctsio);
5599                         } else {
5600                                 ctl_copy_io((union ctl_io *)ctsio,
5601                                             new_io);
5602                                 retval = lun->backend->config_write(new_io);
5603                                 ctl_set_success(ctsio);
5604                                 ctl_done((union ctl_io *)ctsio);
5605                         }
5606                 } else {
5607                         retval = lun->backend->config_write(
5608                                 (union ctl_io *)ctsio);
5609                 }
5610 #ifdef NEEDTOPORT
5611         }
5612 #endif
5613         return (retval);
5614 }
5615
5616 /*
5617  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5618  * we don't really do anything with the LBA and length fields if the user
5619  * passes them in.  Instead we'll just flush out the cache for the entire
5620  * LUN.
5621  */
5622 int
5623 ctl_sync_cache(struct ctl_scsiio *ctsio)
5624 {
5625         struct ctl_lun *lun;
5626         struct ctl_softc *ctl_softc;
5627         uint64_t starting_lba;
5628         uint32_t block_count;
5629         int retval;
5630
5631         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5632
5633         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5634         ctl_softc = control_softc;
5635         retval = 0;
5636
5637         switch (ctsio->cdb[0]) {
5638         case SYNCHRONIZE_CACHE: {
5639                 struct scsi_sync_cache *cdb;
5640                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5641
5642                 starting_lba = scsi_4btoul(cdb->begin_lba);
5643                 block_count = scsi_2btoul(cdb->lb_count);
5644                 break;
5645         }
5646         case SYNCHRONIZE_CACHE_16: {
5647                 struct scsi_sync_cache_16 *cdb;
5648                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5649
5650                 starting_lba = scsi_8btou64(cdb->begin_lba);
5651                 block_count = scsi_4btoul(cdb->lb_count);
5652                 break;
5653         }
5654         default:
5655                 ctl_set_invalid_opcode(ctsio);
5656                 ctl_done((union ctl_io *)ctsio);
5657                 goto bailout;
5658                 break; /* NOTREACHED */
5659         }
5660
5661         /*
5662          * We check the LBA and length, but don't do anything with them.
5663          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5664          * get flushed.  This check will just help satisfy anyone who wants
5665          * to see an error for an out of range LBA.
5666          */
5667         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5668                 ctl_set_lba_out_of_range(ctsio);
5669                 ctl_done((union ctl_io *)ctsio);
5670                 goto bailout;
5671         }
5672
5673         /*
5674          * If this LUN has no backend, we can't flush the cache anyway.
5675          */
5676         if (lun->backend == NULL) {
5677                 ctl_set_invalid_opcode(ctsio);
5678                 ctl_done((union ctl_io *)ctsio);
5679                 goto bailout;
5680         }
5681
5682         /*
5683          * Check to see whether we're configured to send the SYNCHRONIZE
5684          * CACHE command directly to the back end.
5685          */
5686         mtx_lock(&lun->lun_lock);
5687         if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5688          && (++(lun->sync_count) >= lun->sync_interval)) {
5689                 lun->sync_count = 0;
5690                 mtx_unlock(&lun->lun_lock);
5691                 retval = lun->backend->config_write((union ctl_io *)ctsio);
5692         } else {
5693                 mtx_unlock(&lun->lun_lock);
5694                 ctl_set_success(ctsio);
5695                 ctl_done((union ctl_io *)ctsio);
5696         }
5697
5698 bailout:
5699
5700         return (retval);
5701 }
5702
5703 int
5704 ctl_format(struct ctl_scsiio *ctsio)
5705 {
5706         struct scsi_format *cdb;
5707         struct ctl_lun *lun;
5708         struct ctl_softc *ctl_softc;
5709         int length, defect_list_len;
5710
5711         CTL_DEBUG_PRINT(("ctl_format\n"));
5712
5713         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5714         ctl_softc = control_softc;
5715
5716         cdb = (struct scsi_format *)ctsio->cdb;
5717
5718         length = 0;
5719         if (cdb->byte2 & SF_FMTDATA) {
5720                 if (cdb->byte2 & SF_LONGLIST)
5721                         length = sizeof(struct scsi_format_header_long);
5722                 else
5723                         length = sizeof(struct scsi_format_header_short);
5724         }
5725
5726         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5727          && (length > 0)) {
5728                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5729                 ctsio->kern_data_len = length;
5730                 ctsio->kern_total_len = length;
5731                 ctsio->kern_data_resid = 0;
5732                 ctsio->kern_rel_offset = 0;
5733                 ctsio->kern_sg_entries = 0;
5734                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5735                 ctsio->be_move_done = ctl_config_move_done;
5736                 ctl_datamove((union ctl_io *)ctsio);
5737
5738                 return (CTL_RETVAL_COMPLETE);
5739         }
5740
5741         defect_list_len = 0;
5742
5743         if (cdb->byte2 & SF_FMTDATA) {
5744                 if (cdb->byte2 & SF_LONGLIST) {
5745                         struct scsi_format_header_long *header;
5746
5747                         header = (struct scsi_format_header_long *)
5748                                 ctsio->kern_data_ptr;
5749
5750                         defect_list_len = scsi_4btoul(header->defect_list_len);
5751                         if (defect_list_len != 0) {
5752                                 ctl_set_invalid_field(ctsio,
5753                                                       /*sks_valid*/ 1,
5754                                                       /*command*/ 0,
5755                                                       /*field*/ 2,
5756                                                       /*bit_valid*/ 0,
5757                                                       /*bit*/ 0);
5758                                 goto bailout;
5759                         }
5760                 } else {
5761                         struct scsi_format_header_short *header;
5762
5763                         header = (struct scsi_format_header_short *)
5764                                 ctsio->kern_data_ptr;
5765
5766                         defect_list_len = scsi_2btoul(header->defect_list_len);
5767                         if (defect_list_len != 0) {
5768                                 ctl_set_invalid_field(ctsio,
5769                                                       /*sks_valid*/ 1,
5770                                                       /*command*/ 0,
5771                                                       /*field*/ 2,
5772                                                       /*bit_valid*/ 0,
5773                                                       /*bit*/ 0);
5774                                 goto bailout;
5775                         }
5776                 }
5777         }
5778
5779         /*
5780          * The format command will clear out the "Medium format corrupted"
5781          * status if set by the configuration code.  That status is really
5782          * just a way to notify the host that we have lost the media, and
5783          * get them to issue a command that will basically make them think
5784          * they're blowing away the media.
5785          */
5786         mtx_lock(&lun->lun_lock);
5787         lun->flags &= ~CTL_LUN_INOPERABLE;
5788         mtx_unlock(&lun->lun_lock);
5789
5790         ctsio->scsi_status = SCSI_STATUS_OK;
5791         ctsio->io_hdr.status = CTL_SUCCESS;
5792 bailout:
5793
5794         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5795                 free(ctsio->kern_data_ptr, M_CTL);
5796                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5797         }
5798
5799         ctl_done((union ctl_io *)ctsio);
5800         return (CTL_RETVAL_COMPLETE);
5801 }
5802
5803 int
5804 ctl_read_buffer(struct ctl_scsiio *ctsio)
5805 {
5806         struct scsi_read_buffer *cdb;
5807         struct ctl_lun *lun;
5808         int buffer_offset, len;
5809         static uint8_t descr[4];
5810         static uint8_t echo_descr[4] = { 0 };
5811
5812         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5813
5814         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5815         cdb = (struct scsi_read_buffer *)ctsio->cdb;
5816
5817         if (lun->flags & CTL_LUN_PR_RESERVED) {
5818                 uint32_t residx;
5819
5820                 /*
5821                  * XXX KDM need a lock here.
5822                  */
5823                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5824                 if ((lun->res_type == SPR_TYPE_EX_AC
5825                   && residx != lun->pr_res_idx)
5826                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
5827                    || lun->res_type == SPR_TYPE_EX_AC_AR)
5828                   && !lun->per_res[residx].registered)) {
5829                         ctl_set_reservation_conflict(ctsio);
5830                         ctl_done((union ctl_io *)ctsio);
5831                         return (CTL_RETVAL_COMPLETE);
5832                 }
5833         }
5834
5835         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5836             (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5837             (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5838                 ctl_set_invalid_field(ctsio,
5839                                       /*sks_valid*/ 1,
5840                                       /*command*/ 1,
5841                                       /*field*/ 1,
5842                                       /*bit_valid*/ 1,
5843                                       /*bit*/ 4);
5844                 ctl_done((union ctl_io *)ctsio);
5845                 return (CTL_RETVAL_COMPLETE);
5846         }
5847
5848         len = scsi_3btoul(cdb->length);
5849         buffer_offset = scsi_3btoul(cdb->offset);
5850
5851         if (buffer_offset + len > sizeof(lun->write_buffer)) {
5852                 ctl_set_invalid_field(ctsio,
5853                                       /*sks_valid*/ 1,
5854                                       /*command*/ 1,
5855                                       /*field*/ 6,
5856                                       /*bit_valid*/ 0,
5857                                       /*bit*/ 0);
5858                 ctl_done((union ctl_io *)ctsio);
5859                 return (CTL_RETVAL_COMPLETE);
5860         }
5861
5862         if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5863                 descr[0] = 0;
5864                 scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5865                 ctsio->kern_data_ptr = descr;
5866                 len = min(len, sizeof(descr));
5867         } else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5868                 ctsio->kern_data_ptr = echo_descr;
5869                 len = min(len, sizeof(echo_descr));
5870         } else
5871                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5872         ctsio->kern_data_len = len;
5873         ctsio->kern_total_len = len;
5874         ctsio->kern_data_resid = 0;
5875         ctsio->kern_rel_offset = 0;
5876         ctsio->kern_sg_entries = 0;
5877         ctsio->be_move_done = ctl_config_move_done;
5878         ctl_datamove((union ctl_io *)ctsio);
5879
5880         return (CTL_RETVAL_COMPLETE);
5881 }
5882
5883 int
5884 ctl_write_buffer(struct ctl_scsiio *ctsio)
5885 {
5886         struct scsi_write_buffer *cdb;
5887         struct ctl_lun *lun;
5888         int buffer_offset, len;
5889
5890         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5891
5892         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5893         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5894
5895         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5896                 ctl_set_invalid_field(ctsio,
5897                                       /*sks_valid*/ 1,
5898                                       /*command*/ 1,
5899                                       /*field*/ 1,
5900                                       /*bit_valid*/ 1,
5901                                       /*bit*/ 4);
5902                 ctl_done((union ctl_io *)ctsio);
5903                 return (CTL_RETVAL_COMPLETE);
5904         }
5905
5906         len = scsi_3btoul(cdb->length);
5907         buffer_offset = scsi_3btoul(cdb->offset);
5908
5909         if (buffer_offset + len > sizeof(lun->write_buffer)) {
5910                 ctl_set_invalid_field(ctsio,
5911                                       /*sks_valid*/ 1,
5912                                       /*command*/ 1,
5913                                       /*field*/ 6,
5914                                       /*bit_valid*/ 0,
5915                                       /*bit*/ 0);
5916                 ctl_done((union ctl_io *)ctsio);
5917                 return (CTL_RETVAL_COMPLETE);
5918         }
5919
5920         /*
5921          * If we've got a kernel request that hasn't been malloced yet,
5922          * malloc it and tell the caller the data buffer is here.
5923          */
5924         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5925                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5926                 ctsio->kern_data_len = len;
5927                 ctsio->kern_total_len = len;
5928                 ctsio->kern_data_resid = 0;
5929                 ctsio->kern_rel_offset = 0;
5930                 ctsio->kern_sg_entries = 0;
5931                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5932                 ctsio->be_move_done = ctl_config_move_done;
5933                 ctl_datamove((union ctl_io *)ctsio);
5934
5935                 return (CTL_RETVAL_COMPLETE);
5936         }
5937
5938         ctl_done((union ctl_io *)ctsio);
5939
5940         return (CTL_RETVAL_COMPLETE);
5941 }
5942
5943 int
5944 ctl_write_same(struct ctl_scsiio *ctsio)
5945 {
5946         struct ctl_lun *lun;
5947         struct ctl_lba_len_flags *lbalen;
5948         uint64_t lba;
5949         uint32_t num_blocks;
5950         int len, retval;
5951         uint8_t byte2;
5952
5953         retval = CTL_RETVAL_COMPLETE;
5954
5955         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5956
5957         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5958
5959         switch (ctsio->cdb[0]) {
5960         case WRITE_SAME_10: {
5961                 struct scsi_write_same_10 *cdb;
5962
5963                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5964
5965                 lba = scsi_4btoul(cdb->addr);
5966                 num_blocks = scsi_2btoul(cdb->length);
5967                 byte2 = cdb->byte2;
5968                 break;
5969         }
5970         case WRITE_SAME_16: {
5971                 struct scsi_write_same_16 *cdb;
5972
5973                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5974
5975                 lba = scsi_8btou64(cdb->addr);
5976                 num_blocks = scsi_4btoul(cdb->length);
5977                 byte2 = cdb->byte2;
5978                 break;
5979         }
5980         default:
5981                 /*
5982                  * We got a command we don't support.  This shouldn't
5983                  * happen, commands should be filtered out above us.
5984                  */
5985                 ctl_set_invalid_opcode(ctsio);
5986                 ctl_done((union ctl_io *)ctsio);
5987
5988                 return (CTL_RETVAL_COMPLETE);
5989                 break; /* NOTREACHED */
5990         }
5991
5992         /* NDOB flag can be used only together with UNMAP */
5993         if ((byte2 & (SWS_NDOB | SWS_UNMAP)) == SWS_NDOB) {
5994                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5995                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5996                 ctl_done((union ctl_io *)ctsio);
5997                 return (CTL_RETVAL_COMPLETE);
5998         }
5999
6000         /*
6001          * The first check is to make sure we're in bounds, the second
6002          * check is to catch wrap-around problems.  If the lba + num blocks
6003          * is less than the lba, then we've wrapped around and the block
6004          * range is invalid anyway.
6005          */
6006         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6007          || ((lba + num_blocks) < lba)) {
6008                 ctl_set_lba_out_of_range(ctsio);
6009                 ctl_done((union ctl_io *)ctsio);
6010                 return (CTL_RETVAL_COMPLETE);
6011         }
6012
6013         /* Zero number of blocks means "to the last logical block" */
6014         if (num_blocks == 0) {
6015                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
6016                         ctl_set_invalid_field(ctsio,
6017                                               /*sks_valid*/ 0,
6018                                               /*command*/ 1,
6019                                               /*field*/ 0,
6020                                               /*bit_valid*/ 0,
6021                                               /*bit*/ 0);
6022                         ctl_done((union ctl_io *)ctsio);
6023                         return (CTL_RETVAL_COMPLETE);
6024                 }
6025                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
6026         }
6027
6028         len = lun->be_lun->blocksize;
6029
6030         /*
6031          * If we've got a kernel request that hasn't been malloced yet,
6032          * malloc it and tell the caller the data buffer is here.
6033          */
6034         if ((byte2 & SWS_NDOB) == 0 &&
6035             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6036                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6037                 ctsio->kern_data_len = len;
6038                 ctsio->kern_total_len = len;
6039                 ctsio->kern_data_resid = 0;
6040                 ctsio->kern_rel_offset = 0;
6041                 ctsio->kern_sg_entries = 0;
6042                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6043                 ctsio->be_move_done = ctl_config_move_done;
6044                 ctl_datamove((union ctl_io *)ctsio);
6045
6046                 return (CTL_RETVAL_COMPLETE);
6047         }
6048
6049         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6050         lbalen->lba = lba;
6051         lbalen->len = num_blocks;
6052         lbalen->flags = byte2;
6053         retval = lun->backend->config_write((union ctl_io *)ctsio);
6054
6055         return (retval);
6056 }
6057
6058 int
6059 ctl_unmap(struct ctl_scsiio *ctsio)
6060 {
6061         struct ctl_lun *lun;
6062         struct scsi_unmap *cdb;
6063         struct ctl_ptr_len_flags *ptrlen;
6064         struct scsi_unmap_header *hdr;
6065         struct scsi_unmap_desc *buf, *end, *endnz, *range;
6066         uint64_t lba;
6067         uint32_t num_blocks;
6068         int len, retval;
6069         uint8_t byte2;
6070
6071         retval = CTL_RETVAL_COMPLETE;
6072
6073         CTL_DEBUG_PRINT(("ctl_unmap\n"));
6074
6075         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6076         cdb = (struct scsi_unmap *)ctsio->cdb;
6077
6078         len = scsi_2btoul(cdb->length);
6079         byte2 = cdb->byte2;
6080
6081         /*
6082          * If we've got a kernel request that hasn't been malloced yet,
6083          * malloc it and tell the caller the data buffer is here.
6084          */
6085         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6086                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6087                 ctsio->kern_data_len = len;
6088                 ctsio->kern_total_len = len;
6089                 ctsio->kern_data_resid = 0;
6090                 ctsio->kern_rel_offset = 0;
6091                 ctsio->kern_sg_entries = 0;
6092                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6093                 ctsio->be_move_done = ctl_config_move_done;
6094                 ctl_datamove((union ctl_io *)ctsio);
6095
6096                 return (CTL_RETVAL_COMPLETE);
6097         }
6098
6099         len = ctsio->kern_total_len - ctsio->kern_data_resid;
6100         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6101         if (len < sizeof (*hdr) ||
6102             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6103             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6104             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6105                 ctl_set_invalid_field(ctsio,
6106                                       /*sks_valid*/ 0,
6107                                       /*command*/ 0,
6108                                       /*field*/ 0,
6109                                       /*bit_valid*/ 0,
6110                                       /*bit*/ 0);
6111                 ctl_done((union ctl_io *)ctsio);
6112                 return (CTL_RETVAL_COMPLETE);
6113         }
6114         len = scsi_2btoul(hdr->desc_length);
6115         buf = (struct scsi_unmap_desc *)(hdr + 1);
6116         end = buf + len / sizeof(*buf);
6117
6118         endnz = buf;
6119         for (range = buf; range < end; range++) {
6120                 lba = scsi_8btou64(range->lba);
6121                 num_blocks = scsi_4btoul(range->length);
6122                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6123                  || ((lba + num_blocks) < lba)) {
6124                         ctl_set_lba_out_of_range(ctsio);
6125                         ctl_done((union ctl_io *)ctsio);
6126                         return (CTL_RETVAL_COMPLETE);
6127                 }
6128                 if (num_blocks != 0)
6129                         endnz = range + 1;
6130         }
6131
6132         /*
6133          * Block backend can not handle zero last range.
6134          * Filter it out and return if there is nothing left.
6135          */
6136         len = (uint8_t *)endnz - (uint8_t *)buf;
6137         if (len == 0) {
6138                 ctl_set_success(ctsio);
6139                 ctl_done((union ctl_io *)ctsio);
6140                 return (CTL_RETVAL_COMPLETE);
6141         }
6142
6143         mtx_lock(&lun->lun_lock);
6144         ptrlen = (struct ctl_ptr_len_flags *)
6145             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6146         ptrlen->ptr = (void *)buf;
6147         ptrlen->len = len;
6148         ptrlen->flags = byte2;
6149         ctl_check_blocked(lun);
6150         mtx_unlock(&lun->lun_lock);
6151
6152         retval = lun->backend->config_write((union ctl_io *)ctsio);
6153         return (retval);
6154 }
6155
6156 /*
6157  * Note that this function currently doesn't actually do anything inside
6158  * CTL to enforce things if the DQue bit is turned on.
6159  *
6160  * Also note that this function can't be used in the default case, because
6161  * the DQue bit isn't set in the changeable mask for the control mode page
6162  * anyway.  This is just here as an example for how to implement a page
6163  * handler, and a placeholder in case we want to allow the user to turn
6164  * tagged queueing on and off.
6165  *
6166  * The D_SENSE bit handling is functional, however, and will turn
6167  * descriptor sense on and off for a given LUN.
6168  */
6169 int
6170 ctl_control_page_handler(struct ctl_scsiio *ctsio,
6171                          struct ctl_page_index *page_index, uint8_t *page_ptr)
6172 {
6173         struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6174         struct ctl_lun *lun;
6175         struct ctl_softc *softc;
6176         int set_ua;
6177         uint32_t initidx;
6178
6179         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6180         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6181         set_ua = 0;
6182
6183         user_cp = (struct scsi_control_page *)page_ptr;
6184         current_cp = (struct scsi_control_page *)
6185                 (page_index->page_data + (page_index->page_len *
6186                 CTL_PAGE_CURRENT));
6187         saved_cp = (struct scsi_control_page *)
6188                 (page_index->page_data + (page_index->page_len *
6189                 CTL_PAGE_SAVED));
6190
6191         softc = control_softc;
6192
6193         mtx_lock(&lun->lun_lock);
6194         if (((current_cp->rlec & SCP_DSENSE) == 0)
6195          && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6196                 /*
6197                  * Descriptor sense is currently turned off and the user
6198                  * wants to turn it on.
6199                  */
6200                 current_cp->rlec |= SCP_DSENSE;
6201                 saved_cp->rlec |= SCP_DSENSE;
6202                 lun->flags |= CTL_LUN_SENSE_DESC;
6203                 set_ua = 1;
6204         } else if (((current_cp->rlec & SCP_DSENSE) != 0)
6205                 && ((user_cp->rlec & SCP_DSENSE) == 0)) {
6206                 /*
6207                  * Descriptor sense is currently turned on, and the user
6208                  * wants to turn it off.
6209                  */
6210                 current_cp->rlec &= ~SCP_DSENSE;
6211                 saved_cp->rlec &= ~SCP_DSENSE;
6212                 lun->flags &= ~CTL_LUN_SENSE_DESC;
6213                 set_ua = 1;
6214         }
6215         if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6216             (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6217                 current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6218                 current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6219                 saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6220                 saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6221                 set_ua = 1;
6222         }
6223         if (set_ua != 0) {
6224                 int i;
6225                 /*
6226                  * Let other initiators know that the mode
6227                  * parameters for this LUN have changed.
6228                  */
6229                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6230                         if (i == initidx)
6231                                 continue;
6232
6233                         lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6234                 }
6235         }
6236         mtx_unlock(&lun->lun_lock);
6237
6238         return (0);
6239 }
6240
6241 int
6242 ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6243                      struct ctl_page_index *page_index, uint8_t *page_ptr)
6244 {
6245         struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6246         struct ctl_lun *lun;
6247         int set_ua;
6248         uint32_t initidx;
6249
6250         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6251         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6252         set_ua = 0;
6253
6254         user_cp = (struct scsi_caching_page *)page_ptr;
6255         current_cp = (struct scsi_caching_page *)
6256                 (page_index->page_data + (page_index->page_len *
6257                 CTL_PAGE_CURRENT));
6258         saved_cp = (struct scsi_caching_page *)
6259                 (page_index->page_data + (page_index->page_len *
6260                 CTL_PAGE_SAVED));
6261
6262         mtx_lock(&lun->lun_lock);
6263         if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6264             (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6265                 current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6266                 current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6267                 saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6268                 saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6269                 set_ua = 1;
6270         }
6271         if (set_ua != 0) {
6272                 int i;
6273                 /*
6274                  * Let other initiators know that the mode
6275                  * parameters for this LUN have changed.
6276                  */
6277                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6278                         if (i == initidx)
6279                                 continue;
6280
6281                         lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6282                 }
6283         }
6284         mtx_unlock(&lun->lun_lock);
6285
6286         return (0);
6287 }
6288
6289 int
6290 ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6291                      struct ctl_page_index *page_index, uint8_t *page_ptr)
6292 {
6293         return (0);
6294 }
6295
6296 int
6297 ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6298                            struct ctl_page_index *page_index, int pc)
6299 {
6300         struct copan_power_subpage *page;
6301
6302         page = (struct copan_power_subpage *)page_index->page_data +
6303                 (page_index->page_len * pc);
6304
6305         switch (pc) {
6306         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6307                 /*
6308                  * We don't update the changable bits for this page.
6309                  */
6310                 break;
6311         case SMS_PAGE_CTRL_CURRENT >> 6:
6312         case SMS_PAGE_CTRL_DEFAULT >> 6:
6313         case SMS_PAGE_CTRL_SAVED >> 6:
6314 #ifdef NEEDTOPORT
6315                 ctl_update_power_subpage(page);
6316 #endif
6317                 break;
6318         default:
6319 #ifdef NEEDTOPORT
6320                 EPRINT(0, "Invalid PC %d!!", pc);
6321 #endif
6322                 break;
6323         }
6324         return (0);
6325 }
6326
6327
6328 int
6329 ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6330                    struct ctl_page_index *page_index, uint8_t *page_ptr)
6331 {
6332         struct copan_aps_subpage *user_sp;
6333         struct copan_aps_subpage *current_sp;
6334         union ctl_modepage_info *modepage_info;
6335         struct ctl_softc *softc;
6336         struct ctl_lun *lun;
6337         int retval;
6338
6339         retval = CTL_RETVAL_COMPLETE;
6340         current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6341                      (page_index->page_len * CTL_PAGE_CURRENT));
6342         softc = control_softc;
6343         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6344
6345         user_sp = (struct copan_aps_subpage *)page_ptr;
6346
6347         modepage_info = (union ctl_modepage_info *)
6348                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6349
6350         modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6351         modepage_info->header.subpage = page_index->subpage;
6352         modepage_info->aps.lock_active = user_sp->lock_active;
6353
6354         mtx_lock(&softc->ctl_lock);
6355
6356         /*
6357          * If there is a request to lock the LUN and another LUN is locked
6358          * this is an error. If the requested LUN is already locked ignore
6359          * the request. If no LUN is locked attempt to lock it.
6360          * if there is a request to unlock the LUN and the LUN is currently
6361          * locked attempt to unlock it. Otherwise ignore the request. i.e.
6362          * if another LUN is locked or no LUN is locked.
6363          */
6364         if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6365                 if (softc->aps_locked_lun == lun->lun) {
6366                         /*
6367                          * This LUN is already locked, so we're done.
6368                          */
6369                         retval = CTL_RETVAL_COMPLETE;
6370                 } else if (softc->aps_locked_lun == 0) {
6371                         /*
6372                          * No one has the lock, pass the request to the
6373                          * backend.
6374                          */
6375                         retval = lun->backend->config_write(
6376                                 (union ctl_io *)ctsio);
6377                 } else {
6378                         /*
6379                          * Someone else has the lock, throw out the request.
6380                          */
6381                         ctl_set_already_locked(ctsio);
6382                         free(ctsio->kern_data_ptr, M_CTL);
6383                         ctl_done((union ctl_io *)ctsio);
6384
6385                         /*
6386                          * Set the return value so that ctl_do_mode_select()
6387                          * won't try to complete the command.  We already
6388                          * completed it here.
6389                          */
6390                         retval = CTL_RETVAL_ERROR;
6391                 }
6392         } else if (softc->aps_locked_lun == lun->lun) {
6393                 /*
6394                  * This LUN is locked, so pass the unlock request to the
6395                  * backend.
6396                  */
6397                 retval = lun->backend->config_write((union ctl_io *)ctsio);
6398         }
6399         mtx_unlock(&softc->ctl_lock);
6400
6401         return (retval);
6402 }
6403
6404 int
6405 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6406                                 struct ctl_page_index *page_index,
6407                                 uint8_t *page_ptr)
6408 {
6409         uint8_t *c;
6410         int i;
6411
6412         c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6413         ctl_time_io_secs =
6414                 (c[0] << 8) |
6415                 (c[1] << 0) |
6416                 0;
6417         CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6418         printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6419         printf("page data:");
6420         for (i=0; i<8; i++)
6421                 printf(" %.2x",page_ptr[i]);
6422         printf("\n");
6423         return (0);
6424 }
6425
6426 int
6427 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6428                                struct ctl_page_index *page_index,
6429                                int pc)
6430 {
6431         struct copan_debugconf_subpage *page;
6432
6433         page = (struct copan_debugconf_subpage *)page_index->page_data +
6434                 (page_index->page_len * pc);
6435
6436         switch (pc) {
6437         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6438         case SMS_PAGE_CTRL_DEFAULT >> 6:
6439         case SMS_PAGE_CTRL_SAVED >> 6:
6440                 /*
6441                  * We don't update the changable or default bits for this page.
6442                  */
6443                 break;
6444         case SMS_PAGE_CTRL_CURRENT >> 6:
6445                 page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6446                 page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6447                 break;
6448         default:
6449 #ifdef NEEDTOPORT
6450                 EPRINT(0, "Invalid PC %d!!", pc);
6451 #endif /* NEEDTOPORT */
6452                 break;
6453         }
6454         return (0);
6455 }
6456
6457
6458 static int
6459 ctl_do_mode_select(union ctl_io *io)
6460 {
6461         struct scsi_mode_page_header *page_header;
6462         struct ctl_page_index *page_index;
6463         struct ctl_scsiio *ctsio;
6464         int control_dev, page_len;
6465         int page_len_offset, page_len_size;
6466         union ctl_modepage_info *modepage_info;
6467         struct ctl_lun *lun;
6468         int *len_left, *len_used;
6469         int retval, i;
6470
6471         ctsio = &io->scsiio;
6472         page_index = NULL;
6473         page_len = 0;
6474         retval = CTL_RETVAL_COMPLETE;
6475
6476         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6477
6478         if (lun->be_lun->lun_type != T_DIRECT)
6479                 control_dev = 1;
6480         else
6481                 control_dev = 0;
6482
6483         modepage_info = (union ctl_modepage_info *)
6484                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6485         len_left = &modepage_info->header.len_left;
6486         len_used = &modepage_info->header.len_used;
6487
6488 do_next_page:
6489
6490         page_header = (struct scsi_mode_page_header *)
6491                 (ctsio->kern_data_ptr + *len_used);
6492
6493         if (*len_left == 0) {
6494                 free(ctsio->kern_data_ptr, M_CTL);
6495                 ctl_set_success(ctsio);
6496                 ctl_done((union ctl_io *)ctsio);
6497                 return (CTL_RETVAL_COMPLETE);
6498         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6499
6500                 free(ctsio->kern_data_ptr, M_CTL);
6501                 ctl_set_param_len_error(ctsio);
6502                 ctl_done((union ctl_io *)ctsio);
6503                 return (CTL_RETVAL_COMPLETE);
6504
6505         } else if ((page_header->page_code & SMPH_SPF)
6506                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6507
6508                 free(ctsio->kern_data_ptr, M_CTL);
6509                 ctl_set_param_len_error(ctsio);
6510                 ctl_done((union ctl_io *)ctsio);
6511                 return (CTL_RETVAL_COMPLETE);
6512         }
6513
6514
6515         /*
6516          * XXX KDM should we do something with the block descriptor?
6517          */
6518         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6519
6520                 if ((control_dev != 0)
6521                  && (lun->mode_pages.index[i].page_flags &
6522                      CTL_PAGE_FLAG_DISK_ONLY))
6523                         continue;
6524
6525                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6526                     (page_header->page_code & SMPH_PC_MASK))
6527                         continue;
6528
6529                 /*
6530                  * If neither page has a subpage code, then we've got a
6531                  * match.
6532                  */
6533                 if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6534                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6535                         page_index = &lun->mode_pages.index[i];
6536                         page_len = page_header->page_length;
6537                         break;
6538                 }
6539
6540                 /*
6541                  * If both pages have subpages, then the subpage numbers
6542                  * have to match.
6543                  */
6544                 if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6545                   && (page_header->page_code & SMPH_SPF)) {
6546                         struct scsi_mode_page_header_sp *sph;
6547
6548                         sph = (struct scsi_mode_page_header_sp *)page_header;
6549
6550                         if (lun->mode_pages.index[i].subpage ==
6551                             sph->subpage) {
6552                                 page_index = &lun->mode_pages.index[i];
6553                                 page_len = scsi_2btoul(sph->page_length);
6554                                 break;
6555                         }
6556                 }
6557         }
6558
6559         /*
6560          * If we couldn't find the page, or if we don't have a mode select
6561          * handler for it, send back an error to the user.
6562          */
6563         if ((page_index == NULL)
6564          || (page_index->select_handler == NULL)) {
6565                 ctl_set_invalid_field(ctsio,
6566                                       /*sks_valid*/ 1,
6567                                       /*command*/ 0,
6568                                       /*field*/ *len_used,
6569                                       /*bit_valid*/ 0,
6570                                       /*bit*/ 0);
6571                 free(ctsio->kern_data_ptr, M_CTL);
6572                 ctl_done((union ctl_io *)ctsio);
6573                 return (CTL_RETVAL_COMPLETE);
6574         }
6575
6576         if (page_index->page_code & SMPH_SPF) {
6577                 page_len_offset = 2;
6578                 page_len_size = 2;
6579         } else {
6580                 page_len_size = 1;
6581                 page_len_offset = 1;
6582         }
6583
6584         /*
6585          * If the length the initiator gives us isn't the one we specify in
6586          * the mode page header, or if they didn't specify enough data in
6587          * the CDB to avoid truncating this page, kick out the request.
6588          */
6589         if ((page_len != (page_index->page_len - page_len_offset -
6590                           page_len_size))
6591          || (*len_left < page_index->page_len)) {
6592
6593
6594                 ctl_set_invalid_field(ctsio,
6595                                       /*sks_valid*/ 1,
6596                                       /*command*/ 0,
6597                                       /*field*/ *len_used + page_len_offset,
6598                                       /*bit_valid*/ 0,
6599                                       /*bit*/ 0);
6600                 free(ctsio->kern_data_ptr, M_CTL);
6601                 ctl_done((union ctl_io *)ctsio);
6602                 return (CTL_RETVAL_COMPLETE);
6603         }
6604
6605         /*
6606          * Run through the mode page, checking to make sure that the bits
6607          * the user changed are actually legal for him to change.
6608          */
6609         for (i = 0; i < page_index->page_len; i++) {
6610                 uint8_t *user_byte, *change_mask, *current_byte;
6611                 int bad_bit;
6612                 int j;
6613
6614                 user_byte = (uint8_t *)page_header + i;
6615                 change_mask = page_index->page_data +
6616                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6617                 current_byte = page_index->page_data +
6618                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6619
6620                 /*
6621                  * Check to see whether the user set any bits in this byte
6622                  * that he is not allowed to set.
6623                  */
6624                 if ((*user_byte & ~(*change_mask)) ==
6625                     (*current_byte & ~(*change_mask)))
6626                         continue;
6627
6628                 /*
6629                  * Go through bit by bit to determine which one is illegal.
6630                  */
6631                 bad_bit = 0;
6632                 for (j = 7; j >= 0; j--) {
6633                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6634                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6635                                 bad_bit = i;
6636                                 break;
6637                         }
6638                 }
6639                 ctl_set_invalid_field(ctsio,
6640                                       /*sks_valid*/ 1,
6641                                       /*command*/ 0,
6642                                       /*field*/ *len_used + i,
6643                                       /*bit_valid*/ 1,
6644                                       /*bit*/ bad_bit);
6645                 free(ctsio->kern_data_ptr, M_CTL);
6646                 ctl_done((union ctl_io *)ctsio);
6647                 return (CTL_RETVAL_COMPLETE);
6648         }
6649
6650         /*
6651          * Decrement these before we call the page handler, since we may
6652          * end up getting called back one way or another before the handler
6653          * returns to this context.
6654          */
6655         *len_left -= page_index->page_len;
6656         *len_used += page_index->page_len;
6657
6658         retval = page_index->select_handler(ctsio, page_index,
6659                                             (uint8_t *)page_header);
6660
6661         /*
6662          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6663          * wait until this queued command completes to finish processing
6664          * the mode page.  If it returns anything other than
6665          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6666          * already set the sense information, freed the data pointer, and
6667          * completed the io for us.
6668          */
6669         if (retval != CTL_RETVAL_COMPLETE)
6670                 goto bailout_no_done;
6671
6672         /*
6673          * If the initiator sent us more than one page, parse the next one.
6674          */
6675         if (*len_left > 0)
6676                 goto do_next_page;
6677
6678         ctl_set_success(ctsio);
6679         free(ctsio->kern_data_ptr, M_CTL);
6680         ctl_done((union ctl_io *)ctsio);
6681
6682 bailout_no_done:
6683
6684         return (CTL_RETVAL_COMPLETE);
6685
6686 }
6687
6688 int
6689 ctl_mode_select(struct ctl_scsiio *ctsio)
6690 {
6691         int param_len, pf, sp;
6692         int header_size, bd_len;
6693         int len_left, len_used;
6694         struct ctl_page_index *page_index;
6695         struct ctl_lun *lun;
6696         int control_dev, page_len;
6697         union ctl_modepage_info *modepage_info;
6698         int retval;
6699
6700         pf = 0;
6701         sp = 0;
6702         page_len = 0;
6703         len_used = 0;
6704         len_left = 0;
6705         retval = 0;
6706         bd_len = 0;
6707         page_index = NULL;
6708
6709         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6710
6711         if (lun->be_lun->lun_type != T_DIRECT)
6712                 control_dev = 1;
6713         else
6714                 control_dev = 0;
6715
6716         switch (ctsio->cdb[0]) {
6717         case MODE_SELECT_6: {
6718                 struct scsi_mode_select_6 *cdb;
6719
6720                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6721
6722                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6723                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6724
6725                 param_len = cdb->length;
6726                 header_size = sizeof(struct scsi_mode_header_6);
6727                 break;
6728         }
6729         case MODE_SELECT_10: {
6730                 struct scsi_mode_select_10 *cdb;
6731
6732                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6733
6734                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6735                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6736
6737                 param_len = scsi_2btoul(cdb->length);
6738                 header_size = sizeof(struct scsi_mode_header_10);
6739                 break;
6740         }
6741         default:
6742                 ctl_set_invalid_opcode(ctsio);
6743                 ctl_done((union ctl_io *)ctsio);
6744                 return (CTL_RETVAL_COMPLETE);
6745                 break; /* NOTREACHED */
6746         }
6747
6748         /*
6749          * From SPC-3:
6750          * "A parameter list length of zero indicates that the Data-Out Buffer
6751          * shall be empty. This condition shall not be considered as an error."
6752          */
6753         if (param_len == 0) {
6754                 ctl_set_success(ctsio);
6755                 ctl_done((union ctl_io *)ctsio);
6756                 return (CTL_RETVAL_COMPLETE);
6757         }
6758
6759         /*
6760          * Since we'll hit this the first time through, prior to
6761          * allocation, we don't need to free a data buffer here.
6762          */
6763         if (param_len < header_size) {
6764                 ctl_set_param_len_error(ctsio);
6765                 ctl_done((union ctl_io *)ctsio);
6766                 return (CTL_RETVAL_COMPLETE);
6767         }
6768
6769         /*
6770          * Allocate the data buffer and grab the user's data.  In theory,
6771          * we shouldn't have to sanity check the parameter list length here
6772          * because the maximum size is 64K.  We should be able to malloc
6773          * that much without too many problems.
6774          */
6775         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6776                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6777                 ctsio->kern_data_len = param_len;
6778                 ctsio->kern_total_len = param_len;
6779                 ctsio->kern_data_resid = 0;
6780                 ctsio->kern_rel_offset = 0;
6781                 ctsio->kern_sg_entries = 0;
6782                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6783                 ctsio->be_move_done = ctl_config_move_done;
6784                 ctl_datamove((union ctl_io *)ctsio);
6785
6786                 return (CTL_RETVAL_COMPLETE);
6787         }
6788
6789         switch (ctsio->cdb[0]) {
6790         case MODE_SELECT_6: {
6791                 struct scsi_mode_header_6 *mh6;
6792
6793                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6794                 bd_len = mh6->blk_desc_len;
6795                 break;
6796         }
6797         case MODE_SELECT_10: {
6798                 struct scsi_mode_header_10 *mh10;
6799
6800                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6801                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6802                 break;
6803         }
6804         default:
6805                 panic("Invalid CDB type %#x", ctsio->cdb[0]);
6806                 break;
6807         }
6808
6809         if (param_len < (header_size + bd_len)) {
6810                 free(ctsio->kern_data_ptr, M_CTL);
6811                 ctl_set_param_len_error(ctsio);
6812                 ctl_done((union ctl_io *)ctsio);
6813                 return (CTL_RETVAL_COMPLETE);
6814         }
6815
6816         /*
6817          * Set the IO_CONT flag, so that if this I/O gets passed to
6818          * ctl_config_write_done(), it'll get passed back to
6819          * ctl_do_mode_select() for further processing, or completion if
6820          * we're all done.
6821          */
6822         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6823         ctsio->io_cont = ctl_do_mode_select;
6824
6825         modepage_info = (union ctl_modepage_info *)
6826                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6827
6828         memset(modepage_info, 0, sizeof(*modepage_info));
6829
6830         len_left = param_len - header_size - bd_len;
6831         len_used = header_size + bd_len;
6832
6833         modepage_info->header.len_left = len_left;
6834         modepage_info->header.len_used = len_used;
6835
6836         return (ctl_do_mode_select((union ctl_io *)ctsio));
6837 }
6838
6839 int
6840 ctl_mode_sense(struct ctl_scsiio *ctsio)
6841 {
6842         struct ctl_lun *lun;
6843         int pc, page_code, dbd, llba, subpage;
6844         int alloc_len, page_len, header_len, total_len;
6845         struct scsi_mode_block_descr *block_desc;
6846         struct ctl_page_index *page_index;
6847         int control_dev;
6848
6849         dbd = 0;
6850         llba = 0;
6851         block_desc = NULL;
6852         page_index = NULL;
6853
6854         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6855
6856         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6857
6858         if (lun->be_lun->lun_type != T_DIRECT)
6859                 control_dev = 1;
6860         else
6861                 control_dev = 0;
6862
6863         if (lun->flags & CTL_LUN_PR_RESERVED) {
6864                 uint32_t residx;
6865
6866                 /*
6867                  * XXX KDM need a lock here.
6868                  */
6869                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6870                 if ((lun->res_type == SPR_TYPE_EX_AC
6871                   && residx != lun->pr_res_idx)
6872                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
6873                    || lun->res_type == SPR_TYPE_EX_AC_AR)
6874                   && !lun->per_res[residx].registered)) {
6875                         ctl_set_reservation_conflict(ctsio);
6876                         ctl_done((union ctl_io *)ctsio);
6877                         return (CTL_RETVAL_COMPLETE);
6878                 }
6879         }
6880
6881         switch (ctsio->cdb[0]) {
6882         case MODE_SENSE_6: {
6883                 struct scsi_mode_sense_6 *cdb;
6884
6885                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6886
6887                 header_len = sizeof(struct scsi_mode_hdr_6);
6888                 if (cdb->byte2 & SMS_DBD)
6889                         dbd = 1;
6890                 else
6891                         header_len += sizeof(struct scsi_mode_block_descr);
6892
6893                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6894                 page_code = cdb->page & SMS_PAGE_CODE;
6895                 subpage = cdb->subpage;
6896                 alloc_len = cdb->length;
6897                 break;
6898         }
6899         case MODE_SENSE_10: {
6900                 struct scsi_mode_sense_10 *cdb;
6901
6902                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6903
6904                 header_len = sizeof(struct scsi_mode_hdr_10);
6905
6906                 if (cdb->byte2 & SMS_DBD)
6907                         dbd = 1;
6908                 else
6909                         header_len += sizeof(struct scsi_mode_block_descr);
6910                 if (cdb->byte2 & SMS10_LLBAA)
6911                         llba = 1;
6912                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6913                 page_code = cdb->page & SMS_PAGE_CODE;
6914                 subpage = cdb->subpage;
6915                 alloc_len = scsi_2btoul(cdb->length);
6916                 break;
6917         }
6918         default:
6919                 ctl_set_invalid_opcode(ctsio);
6920                 ctl_done((union ctl_io *)ctsio);
6921                 return (CTL_RETVAL_COMPLETE);
6922                 break; /* NOTREACHED */
6923         }
6924
6925         /*
6926          * We have to make a first pass through to calculate the size of
6927          * the pages that match the user's query.  Then we allocate enough
6928          * memory to hold it, and actually copy the data into the buffer.
6929          */
6930         switch (page_code) {
6931         case SMS_ALL_PAGES_PAGE: {
6932                 int i;
6933
6934                 page_len = 0;
6935
6936                 /*
6937                  * At the moment, values other than 0 and 0xff here are
6938                  * reserved according to SPC-3.
6939                  */
6940                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6941                  && (subpage != SMS_SUBPAGE_ALL)) {
6942                         ctl_set_invalid_field(ctsio,
6943                                               /*sks_valid*/ 1,
6944                                               /*command*/ 1,
6945                                               /*field*/ 3,
6946                                               /*bit_valid*/ 0,
6947                                               /*bit*/ 0);
6948                         ctl_done((union ctl_io *)ctsio);
6949                         return (CTL_RETVAL_COMPLETE);
6950                 }
6951
6952                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6953                         if ((control_dev != 0)
6954                          && (lun->mode_pages.index[i].page_flags &
6955                              CTL_PAGE_FLAG_DISK_ONLY))
6956                                 continue;
6957
6958                         /*
6959                          * We don't use this subpage if the user didn't
6960                          * request all subpages.
6961                          */
6962                         if ((lun->mode_pages.index[i].subpage != 0)
6963                          && (subpage == SMS_SUBPAGE_PAGE_0))
6964                                 continue;
6965
6966 #if 0
6967                         printf("found page %#x len %d\n",
6968                                lun->mode_pages.index[i].page_code &
6969                                SMPH_PC_MASK,
6970                                lun->mode_pages.index[i].page_len);
6971 #endif
6972                         page_len += lun->mode_pages.index[i].page_len;
6973                 }
6974                 break;
6975         }
6976         default: {
6977                 int i;
6978
6979                 page_len = 0;
6980
6981                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6982                         /* Look for the right page code */
6983                         if ((lun->mode_pages.index[i].page_code &
6984                              SMPH_PC_MASK) != page_code)
6985                                 continue;
6986
6987                         /* Look for the right subpage or the subpage wildcard*/
6988                         if ((lun->mode_pages.index[i].subpage != subpage)
6989                          && (subpage != SMS_SUBPAGE_ALL))
6990                                 continue;
6991
6992                         /* Make sure the page is supported for this dev type */
6993                         if ((control_dev != 0)
6994                          && (lun->mode_pages.index[i].page_flags &
6995                              CTL_PAGE_FLAG_DISK_ONLY))
6996                                 continue;
6997
6998 #if 0
6999                         printf("found page %#x len %d\n",
7000                                lun->mode_pages.index[i].page_code &
7001                                SMPH_PC_MASK,
7002                                lun->mode_pages.index[i].page_len);
7003 #endif
7004
7005                         page_len += lun->mode_pages.index[i].page_len;
7006                 }
7007
7008                 if (page_len == 0) {
7009                         ctl_set_invalid_field(ctsio,
7010                                               /*sks_valid*/ 1,
7011                                               /*command*/ 1,
7012                                               /*field*/ 2,
7013                                               /*bit_valid*/ 1,
7014                                               /*bit*/ 5);
7015                         ctl_done((union ctl_io *)ctsio);
7016                         return (CTL_RETVAL_COMPLETE);
7017                 }
7018                 break;
7019         }
7020         }
7021
7022         total_len = header_len + page_len;
7023 #if 0
7024         printf("header_len = %d, page_len = %d, total_len = %d\n",
7025                header_len, page_len, total_len);
7026 #endif
7027
7028         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7029         ctsio->kern_sg_entries = 0;
7030         ctsio->kern_data_resid = 0;
7031         ctsio->kern_rel_offset = 0;
7032         if (total_len < alloc_len) {
7033                 ctsio->residual = alloc_len - total_len;
7034                 ctsio->kern_data_len = total_len;
7035                 ctsio->kern_total_len = total_len;
7036         } else {
7037                 ctsio->residual = 0;
7038                 ctsio->kern_data_len = alloc_len;
7039                 ctsio->kern_total_len = alloc_len;
7040         }
7041
7042         switch (ctsio->cdb[0]) {
7043         case MODE_SENSE_6: {
7044                 struct scsi_mode_hdr_6 *header;
7045
7046                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7047
7048                 header->datalen = ctl_min(total_len - 1, 254);
7049                 if (control_dev == 0)
7050                         header->dev_specific = 0x10; /* DPOFUA */
7051                 if (dbd)
7052                         header->block_descr_len = 0;
7053                 else
7054                         header->block_descr_len =
7055                                 sizeof(struct scsi_mode_block_descr);
7056                 block_desc = (struct scsi_mode_block_descr *)&header[1];
7057                 break;
7058         }
7059         case MODE_SENSE_10: {
7060                 struct scsi_mode_hdr_10 *header;
7061                 int datalen;
7062
7063                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7064
7065                 datalen = ctl_min(total_len - 2, 65533);
7066                 scsi_ulto2b(datalen, header->datalen);
7067                 if (control_dev == 0)
7068                         header->dev_specific = 0x10; /* DPOFUA */
7069                 if (dbd)
7070                         scsi_ulto2b(0, header->block_descr_len);
7071                 else
7072                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7073                                     header->block_descr_len);
7074                 block_desc = (struct scsi_mode_block_descr *)&header[1];
7075                 break;
7076         }
7077         default:
7078                 panic("invalid CDB type %#x", ctsio->cdb[0]);
7079                 break; /* NOTREACHED */
7080         }
7081
7082         /*
7083          * If we've got a disk, use its blocksize in the block
7084          * descriptor.  Otherwise, just set it to 0.
7085          */
7086         if (dbd == 0) {
7087                 if (control_dev == 0)
7088                         scsi_ulto3b(lun->be_lun->blocksize,
7089                                     block_desc->block_len);
7090                 else
7091                         scsi_ulto3b(0, block_desc->block_len);
7092         }
7093
7094         switch (page_code) {
7095         case SMS_ALL_PAGES_PAGE: {
7096                 int i, data_used;
7097
7098                 data_used = header_len;
7099                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7100                         struct ctl_page_index *page_index;
7101
7102                         page_index = &lun->mode_pages.index[i];
7103
7104                         if ((control_dev != 0)
7105                          && (page_index->page_flags &
7106                             CTL_PAGE_FLAG_DISK_ONLY))
7107                                 continue;
7108
7109                         /*
7110                          * We don't use this subpage if the user didn't
7111                          * request all subpages.  We already checked (above)
7112                          * to make sure the user only specified a subpage
7113                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7114                          */
7115                         if ((page_index->subpage != 0)
7116                          && (subpage == SMS_SUBPAGE_PAGE_0))
7117                                 continue;
7118
7119                         /*
7120                          * Call the handler, if it exists, to update the
7121                          * page to the latest values.
7122                          */
7123                         if (page_index->sense_handler != NULL)
7124                                 page_index->sense_handler(ctsio, page_index,pc);
7125
7126                         memcpy(ctsio->kern_data_ptr + data_used,
7127                                page_index->page_data +
7128                                (page_index->page_len * pc),
7129                                page_index->page_len);
7130                         data_used += page_index->page_len;
7131                 }
7132                 break;
7133         }
7134         default: {
7135                 int i, data_used;
7136
7137                 data_used = header_len;
7138
7139                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7140                         struct ctl_page_index *page_index;
7141
7142                         page_index = &lun->mode_pages.index[i];
7143
7144                         /* Look for the right page code */
7145                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7146                                 continue;
7147
7148                         /* Look for the right subpage or the subpage wildcard*/
7149                         if ((page_index->subpage != subpage)
7150                          && (subpage != SMS_SUBPAGE_ALL))
7151                                 continue;
7152
7153                         /* Make sure the page is supported for this dev type */
7154                         if ((control_dev != 0)
7155                          && (page_index->page_flags &
7156                              CTL_PAGE_FLAG_DISK_ONLY))
7157                                 continue;
7158
7159                         /*
7160                          * Call the handler, if it exists, to update the
7161                          * page to the latest values.
7162                          */
7163                         if (page_index->sense_handler != NULL)
7164                                 page_index->sense_handler(ctsio, page_index,pc);
7165
7166                         memcpy(ctsio->kern_data_ptr + data_used,
7167                                page_index->page_data +
7168                                (page_index->page_len * pc),
7169                                page_index->page_len);
7170                         data_used += page_index->page_len;
7171                 }
7172                 break;
7173         }
7174         }
7175
7176         ctsio->scsi_status = SCSI_STATUS_OK;
7177
7178         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7179         ctsio->be_move_done = ctl_config_move_done;
7180         ctl_datamove((union ctl_io *)ctsio);
7181
7182         return (CTL_RETVAL_COMPLETE);
7183 }
7184
7185 int
7186 ctl_read_capacity(struct ctl_scsiio *ctsio)
7187 {
7188         struct scsi_read_capacity *cdb;
7189         struct scsi_read_capacity_data *data;
7190         struct ctl_lun *lun;
7191         uint32_t lba;
7192
7193         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7194
7195         cdb = (struct scsi_read_capacity *)ctsio->cdb;
7196
7197         lba = scsi_4btoul(cdb->addr);
7198         if (((cdb->pmi & SRC_PMI) == 0)
7199          && (lba != 0)) {
7200                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7201                                       /*sks_valid*/ 1,
7202                                       /*command*/ 1,
7203                                       /*field*/ 2,
7204                                       /*bit_valid*/ 0,
7205                                       /*bit*/ 0);
7206                 ctl_done((union ctl_io *)ctsio);
7207                 return (CTL_RETVAL_COMPLETE);
7208         }
7209
7210         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7211
7212         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7213         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7214         ctsio->residual = 0;
7215         ctsio->kern_data_len = sizeof(*data);
7216         ctsio->kern_total_len = sizeof(*data);
7217         ctsio->kern_data_resid = 0;
7218         ctsio->kern_rel_offset = 0;
7219         ctsio->kern_sg_entries = 0;
7220
7221         /*
7222          * If the maximum LBA is greater than 0xfffffffe, the user must
7223          * issue a SERVICE ACTION IN (16) command, with the read capacity
7224          * serivce action set.
7225          */
7226         if (lun->be_lun->maxlba > 0xfffffffe)
7227                 scsi_ulto4b(0xffffffff, data->addr);
7228         else
7229                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7230
7231         /*
7232          * XXX KDM this may not be 512 bytes...
7233          */
7234         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7235
7236         ctsio->scsi_status = SCSI_STATUS_OK;
7237
7238         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7239         ctsio->be_move_done = ctl_config_move_done;
7240         ctl_datamove((union ctl_io *)ctsio);
7241
7242         return (CTL_RETVAL_COMPLETE);
7243 }
7244
7245 int
7246 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7247 {
7248         struct scsi_read_capacity_16 *cdb;
7249         struct scsi_read_capacity_data_long *data;
7250         struct ctl_lun *lun;
7251         uint64_t lba;
7252         uint32_t alloc_len;
7253
7254         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7255
7256         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7257
7258         alloc_len = scsi_4btoul(cdb->alloc_len);
7259         lba = scsi_8btou64(cdb->addr);
7260
7261         if ((cdb->reladr & SRC16_PMI)
7262          && (lba != 0)) {
7263                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7264                                       /*sks_valid*/ 1,
7265                                       /*command*/ 1,
7266                                       /*field*/ 2,
7267                                       /*bit_valid*/ 0,
7268                                       /*bit*/ 0);
7269                 ctl_done((union ctl_io *)ctsio);
7270                 return (CTL_RETVAL_COMPLETE);
7271         }
7272
7273         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7274
7275         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7276         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7277
7278         if (sizeof(*data) < alloc_len) {
7279                 ctsio->residual = alloc_len - sizeof(*data);
7280                 ctsio->kern_data_len = sizeof(*data);
7281                 ctsio->kern_total_len = sizeof(*data);
7282         } else {
7283                 ctsio->residual = 0;
7284                 ctsio->kern_data_len = alloc_len;
7285                 ctsio->kern_total_len = alloc_len;
7286         }
7287         ctsio->kern_data_resid = 0;
7288         ctsio->kern_rel_offset = 0;
7289         ctsio->kern_sg_entries = 0;
7290
7291         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7292         /* XXX KDM this may not be 512 bytes... */
7293         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7294         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7295         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7296         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7297                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7298
7299         ctsio->scsi_status = SCSI_STATUS_OK;
7300
7301         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7302         ctsio->be_move_done = ctl_config_move_done;
7303         ctl_datamove((union ctl_io *)ctsio);
7304
7305         return (CTL_RETVAL_COMPLETE);
7306 }
7307
7308 int
7309 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7310 {
7311         struct scsi_maintenance_in *cdb;
7312         int retval;
7313         int alloc_len, ext, total_len = 0, g, p, pc, pg;
7314         int num_target_port_groups, num_target_ports, single;
7315         struct ctl_lun *lun;
7316         struct ctl_softc *softc;
7317         struct ctl_port *port;
7318         struct scsi_target_group_data *rtg_ptr;
7319         struct scsi_target_group_data_extended *rtg_ext_ptr;
7320         struct scsi_target_port_group_descriptor *tpg_desc;
7321
7322         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7323
7324         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7325         softc = control_softc;
7326         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7327
7328         retval = CTL_RETVAL_COMPLETE;
7329
7330         switch (cdb->byte2 & STG_PDF_MASK) {
7331         case STG_PDF_LENGTH:
7332                 ext = 0;
7333                 break;
7334         case STG_PDF_EXTENDED:
7335                 ext = 1;
7336                 break;
7337         default:
7338                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7339                                       /*sks_valid*/ 1,
7340                                       /*command*/ 1,
7341                                       /*field*/ 2,
7342                                       /*bit_valid*/ 1,
7343                                       /*bit*/ 5);
7344                 ctl_done((union ctl_io *)ctsio);
7345                 return(retval);
7346         }
7347
7348         single = ctl_is_single;
7349         if (single)
7350                 num_target_port_groups = 1;
7351         else
7352                 num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7353         num_target_ports = 0;
7354         mtx_lock(&softc->ctl_lock);
7355         STAILQ_FOREACH(port, &softc->port_list, links) {
7356                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7357                         continue;
7358                 if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7359                         continue;
7360                 num_target_ports++;
7361         }
7362         mtx_unlock(&softc->ctl_lock);
7363
7364         if (ext)
7365                 total_len = sizeof(struct scsi_target_group_data_extended);
7366         else
7367                 total_len = sizeof(struct scsi_target_group_data);
7368         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7369                 num_target_port_groups +
7370             sizeof(struct scsi_target_port_descriptor) *
7371                 num_target_ports * num_target_port_groups;
7372
7373         alloc_len = scsi_4btoul(cdb->length);
7374
7375         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7376
7377         ctsio->kern_sg_entries = 0;
7378
7379         if (total_len < alloc_len) {
7380                 ctsio->residual = alloc_len - total_len;
7381                 ctsio->kern_data_len = total_len;
7382                 ctsio->kern_total_len = total_len;
7383         } else {
7384                 ctsio->residual = 0;
7385                 ctsio->kern_data_len = alloc_len;
7386                 ctsio->kern_total_len = alloc_len;
7387         }
7388         ctsio->kern_data_resid = 0;
7389         ctsio->kern_rel_offset = 0;
7390
7391         if (ext) {
7392                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7393                     ctsio->kern_data_ptr;
7394                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7395                 rtg_ext_ptr->format_type = 0x10;
7396                 rtg_ext_ptr->implicit_transition_time = 0;
7397                 tpg_desc = &rtg_ext_ptr->groups[0];
7398         } else {
7399                 rtg_ptr = (struct scsi_target_group_data *)
7400                     ctsio->kern_data_ptr;
7401                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7402                 tpg_desc = &rtg_ptr->groups[0];
7403         }
7404
7405         pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7406         mtx_lock(&softc->ctl_lock);
7407         for (g = 0; g < num_target_port_groups; g++) {
7408                 if (g == pg)
7409                         tpg_desc->pref_state = TPG_PRIMARY |
7410                             TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7411                 else
7412                         tpg_desc->pref_state =
7413                             TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7414                 tpg_desc->support = TPG_AO_SUP;
7415                 if (!single)
7416                         tpg_desc->support |= TPG_AN_SUP;
7417                 scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7418                 tpg_desc->status = TPG_IMPLICIT;
7419                 pc = 0;
7420                 STAILQ_FOREACH(port, &softc->port_list, links) {
7421                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7422                                 continue;
7423                         if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7424                             CTL_MAX_LUNS)
7425                                 continue;
7426                         p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7427                         scsi_ulto2b(p, tpg_desc->descriptors[pc].
7428                             relative_target_port_identifier);
7429                         pc++;
7430                 }
7431                 tpg_desc->target_port_count = pc;
7432                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7433                     &tpg_desc->descriptors[pc];
7434         }
7435         mtx_unlock(&softc->ctl_lock);
7436
7437         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7438         ctsio->be_move_done = ctl_config_move_done;
7439
7440         CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7441                          ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7442                          ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7443                          ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7444                          ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7445
7446         ctl_datamove((union ctl_io *)ctsio);
7447         return(retval);
7448 }
7449
7450 int
7451 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7452 {
7453         struct ctl_lun *lun;
7454         struct scsi_report_supported_opcodes *cdb;
7455         const struct ctl_cmd_entry *entry, *sentry;
7456         struct scsi_report_supported_opcodes_all *all;
7457         struct scsi_report_supported_opcodes_descr *descr;
7458         struct scsi_report_supported_opcodes_one *one;
7459         int retval;
7460         int alloc_len, total_len;
7461         int opcode, service_action, i, j, num;
7462
7463         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7464
7465         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7466         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7467
7468         retval = CTL_RETVAL_COMPLETE;
7469
7470         opcode = cdb->requested_opcode;
7471         service_action = scsi_2btoul(cdb->requested_service_action);
7472         switch (cdb->options & RSO_OPTIONS_MASK) {
7473         case RSO_OPTIONS_ALL:
7474                 num = 0;
7475                 for (i = 0; i < 256; i++) {
7476                         entry = &ctl_cmd_table[i];
7477                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7478                                 for (j = 0; j < 32; j++) {
7479                                         sentry = &((const struct ctl_cmd_entry *)
7480                                             entry->execute)[j];
7481                                         if (ctl_cmd_applicable(
7482                                             lun->be_lun->lun_type, sentry))
7483                                                 num++;
7484                                 }
7485                         } else {
7486                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7487                                     entry))
7488                                         num++;
7489                         }
7490                 }
7491                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7492                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7493                 break;
7494         case RSO_OPTIONS_OC:
7495                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7496                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7497                                               /*sks_valid*/ 1,
7498                                               /*command*/ 1,
7499                                               /*field*/ 2,
7500                                               /*bit_valid*/ 1,
7501                                               /*bit*/ 2);
7502                         ctl_done((union ctl_io *)ctsio);
7503                         return (CTL_RETVAL_COMPLETE);
7504                 }
7505                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7506                 break;
7507         case RSO_OPTIONS_OC_SA:
7508                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7509                     service_action >= 32) {
7510                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7511                                               /*sks_valid*/ 1,
7512                                               /*command*/ 1,
7513                                               /*field*/ 2,
7514                                               /*bit_valid*/ 1,
7515                                               /*bit*/ 2);
7516                         ctl_done((union ctl_io *)ctsio);
7517                         return (CTL_RETVAL_COMPLETE);
7518                 }
7519                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7520                 break;
7521         default:
7522                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7523                                       /*sks_valid*/ 1,
7524                                       /*command*/ 1,
7525                                       /*field*/ 2,
7526                                       /*bit_valid*/ 1,
7527                                       /*bit*/ 2);
7528                 ctl_done((union ctl_io *)ctsio);
7529                 return (CTL_RETVAL_COMPLETE);
7530         }
7531
7532         alloc_len = scsi_4btoul(cdb->length);
7533
7534         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7535
7536         ctsio->kern_sg_entries = 0;
7537
7538         if (total_len < alloc_len) {
7539                 ctsio->residual = alloc_len - total_len;
7540                 ctsio->kern_data_len = total_len;
7541                 ctsio->kern_total_len = total_len;
7542         } else {
7543                 ctsio->residual = 0;
7544                 ctsio->kern_data_len = alloc_len;
7545                 ctsio->kern_total_len = alloc_len;
7546         }
7547         ctsio->kern_data_resid = 0;
7548         ctsio->kern_rel_offset = 0;
7549
7550         switch (cdb->options & RSO_OPTIONS_MASK) {
7551         case RSO_OPTIONS_ALL:
7552                 all = (struct scsi_report_supported_opcodes_all *)
7553                     ctsio->kern_data_ptr;
7554                 num = 0;
7555                 for (i = 0; i < 256; i++) {
7556                         entry = &ctl_cmd_table[i];
7557                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7558                                 for (j = 0; j < 32; j++) {
7559                                         sentry = &((const struct ctl_cmd_entry *)
7560                                             entry->execute)[j];
7561                                         if (!ctl_cmd_applicable(
7562                                             lun->be_lun->lun_type, sentry))
7563                                                 continue;
7564                                         descr = &all->descr[num++];
7565                                         descr->opcode = i;
7566                                         scsi_ulto2b(j, descr->service_action);
7567                                         descr->flags = RSO_SERVACTV;
7568                                         scsi_ulto2b(sentry->length,
7569                                             descr->cdb_length);
7570                                 }
7571                         } else {
7572                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7573                                     entry))
7574                                         continue;
7575                                 descr = &all->descr[num++];
7576                                 descr->opcode = i;
7577                                 scsi_ulto2b(0, descr->service_action);
7578                                 descr->flags = 0;
7579                                 scsi_ulto2b(entry->length, descr->cdb_length);
7580                         }
7581                 }
7582                 scsi_ulto4b(
7583                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7584                     all->length);
7585                 break;
7586         case RSO_OPTIONS_OC:
7587                 one = (struct scsi_report_supported_opcodes_one *)
7588                     ctsio->kern_data_ptr;
7589                 entry = &ctl_cmd_table[opcode];
7590                 goto fill_one;
7591         case RSO_OPTIONS_OC_SA:
7592                 one = (struct scsi_report_supported_opcodes_one *)
7593                     ctsio->kern_data_ptr;
7594                 entry = &ctl_cmd_table[opcode];
7595                 entry = &((const struct ctl_cmd_entry *)
7596                     entry->execute)[service_action];
7597 fill_one:
7598                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7599                         one->support = 3;
7600                         scsi_ulto2b(entry->length, one->cdb_length);
7601                         one->cdb_usage[0] = opcode;
7602                         memcpy(&one->cdb_usage[1], entry->usage,
7603                             entry->length - 1);
7604                 } else
7605                         one->support = 1;
7606                 break;
7607         }
7608
7609         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7610         ctsio->be_move_done = ctl_config_move_done;
7611
7612         ctl_datamove((union ctl_io *)ctsio);
7613         return(retval);
7614 }
7615
7616 int
7617 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7618 {
7619         struct ctl_lun *lun;
7620         struct scsi_report_supported_tmf *cdb;
7621         struct scsi_report_supported_tmf_data *data;
7622         int retval;
7623         int alloc_len, total_len;
7624
7625         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7626
7627         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7628         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7629
7630         retval = CTL_RETVAL_COMPLETE;
7631
7632         total_len = sizeof(struct scsi_report_supported_tmf_data);
7633         alloc_len = scsi_4btoul(cdb->length);
7634
7635         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7636
7637         ctsio->kern_sg_entries = 0;
7638
7639         if (total_len < alloc_len) {
7640                 ctsio->residual = alloc_len - total_len;
7641                 ctsio->kern_data_len = total_len;
7642                 ctsio->kern_total_len = total_len;
7643         } else {
7644                 ctsio->residual = 0;
7645                 ctsio->kern_data_len = alloc_len;
7646                 ctsio->kern_total_len = alloc_len;
7647         }
7648         ctsio->kern_data_resid = 0;
7649         ctsio->kern_rel_offset = 0;
7650
7651         data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7652         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7653         data->byte2 |= RST_ITNRS;
7654
7655         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7656         ctsio->be_move_done = ctl_config_move_done;
7657
7658         ctl_datamove((union ctl_io *)ctsio);
7659         return (retval);
7660 }
7661
7662 int
7663 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7664 {
7665         struct ctl_lun *lun;
7666         struct scsi_report_timestamp *cdb;
7667         struct scsi_report_timestamp_data *data;
7668         struct timeval tv;
7669         int64_t timestamp;
7670         int retval;
7671         int alloc_len, total_len;
7672
7673         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7674
7675         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7676         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7677
7678         retval = CTL_RETVAL_COMPLETE;
7679
7680         total_len = sizeof(struct scsi_report_timestamp_data);
7681         alloc_len = scsi_4btoul(cdb->length);
7682
7683         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7684
7685         ctsio->kern_sg_entries = 0;
7686
7687         if (total_len < alloc_len) {
7688                 ctsio->residual = alloc_len - total_len;
7689                 ctsio->kern_data_len = total_len;
7690                 ctsio->kern_total_len = total_len;
7691         } else {
7692                 ctsio->residual = 0;
7693                 ctsio->kern_data_len = alloc_len;
7694                 ctsio->kern_total_len = alloc_len;
7695         }
7696         ctsio->kern_data_resid = 0;
7697         ctsio->kern_rel_offset = 0;
7698
7699         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7700         scsi_ulto2b(sizeof(*data) - 2, data->length);
7701         data->origin = RTS_ORIG_OUTSIDE;
7702         getmicrotime(&tv);
7703         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7704         scsi_ulto4b(timestamp >> 16, data->timestamp);
7705         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7706
7707         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7708         ctsio->be_move_done = ctl_config_move_done;
7709
7710         ctl_datamove((union ctl_io *)ctsio);
7711         return (retval);
7712 }
7713
7714 int
7715 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7716 {
7717         struct scsi_per_res_in *cdb;
7718         int alloc_len, total_len = 0;
7719         /* struct scsi_per_res_in_rsrv in_data; */
7720         struct ctl_lun *lun;
7721         struct ctl_softc *softc;
7722
7723         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7724
7725         softc = control_softc;
7726
7727         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7728
7729         alloc_len = scsi_2btoul(cdb->length);
7730
7731         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7732
7733 retry:
7734         mtx_lock(&lun->lun_lock);
7735         switch (cdb->action) {
7736         case SPRI_RK: /* read keys */
7737                 total_len = sizeof(struct scsi_per_res_in_keys) +
7738                         lun->pr_key_count *
7739                         sizeof(struct scsi_per_res_key);
7740                 break;
7741         case SPRI_RR: /* read reservation */
7742                 if (lun->flags & CTL_LUN_PR_RESERVED)
7743                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7744                 else
7745                         total_len = sizeof(struct scsi_per_res_in_header);
7746                 break;
7747         case SPRI_RC: /* report capabilities */
7748                 total_len = sizeof(struct scsi_per_res_cap);
7749                 break;
7750         case SPRI_RS: /* read full status */
7751                 total_len = sizeof(struct scsi_per_res_in_header) +
7752                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7753                     lun->pr_key_count;
7754                 break;
7755         default:
7756                 panic("Invalid PR type %x", cdb->action);
7757         }
7758         mtx_unlock(&lun->lun_lock);
7759
7760         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7761
7762         if (total_len < alloc_len) {
7763                 ctsio->residual = alloc_len - total_len;
7764                 ctsio->kern_data_len = total_len;
7765                 ctsio->kern_total_len = total_len;
7766         } else {
7767                 ctsio->residual = 0;
7768                 ctsio->kern_data_len = alloc_len;
7769                 ctsio->kern_total_len = alloc_len;
7770         }
7771
7772         ctsio->kern_data_resid = 0;
7773         ctsio->kern_rel_offset = 0;
7774         ctsio->kern_sg_entries = 0;
7775
7776         mtx_lock(&lun->lun_lock);
7777         switch (cdb->action) {
7778         case SPRI_RK: { // read keys
7779         struct scsi_per_res_in_keys *res_keys;
7780                 int i, key_count;
7781
7782                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7783
7784                 /*
7785                  * We had to drop the lock to allocate our buffer, which
7786                  * leaves time for someone to come in with another
7787                  * persistent reservation.  (That is unlikely, though,
7788                  * since this should be the only persistent reservation
7789                  * command active right now.)
7790                  */
7791                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7792                     (lun->pr_key_count *
7793                      sizeof(struct scsi_per_res_key)))){
7794                         mtx_unlock(&lun->lun_lock);
7795                         free(ctsio->kern_data_ptr, M_CTL);
7796                         printf("%s: reservation length changed, retrying\n",
7797                                __func__);
7798                         goto retry;
7799                 }
7800
7801                 scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7802
7803                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7804                              lun->pr_key_count, res_keys->header.length);
7805
7806                 for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7807                         if (!lun->per_res[i].registered)
7808                                 continue;
7809
7810                         /*
7811                          * We used lun->pr_key_count to calculate the
7812                          * size to allocate.  If it turns out the number of
7813                          * initiators with the registered flag set is
7814                          * larger than that (i.e. they haven't been kept in
7815                          * sync), we've got a problem.
7816                          */
7817                         if (key_count >= lun->pr_key_count) {
7818 #ifdef NEEDTOPORT
7819                                 csevent_log(CSC_CTL | CSC_SHELF_SW |
7820                                             CTL_PR_ERROR,
7821                                             csevent_LogType_Fault,
7822                                             csevent_AlertLevel_Yellow,
7823                                             csevent_FRU_ShelfController,
7824                                             csevent_FRU_Firmware,
7825                                         csevent_FRU_Unknown,
7826                                             "registered keys %d >= key "
7827                                             "count %d", key_count,
7828                                             lun->pr_key_count);
7829 #endif
7830                                 key_count++;
7831                                 continue;
7832                         }
7833                         memcpy(res_keys->keys[key_count].key,
7834                                lun->per_res[i].res_key.key,
7835                                ctl_min(sizeof(res_keys->keys[key_count].key),
7836                                sizeof(lun->per_res[i].res_key)));
7837                         key_count++;
7838                 }
7839                 break;
7840         }
7841         case SPRI_RR: { // read reservation
7842                 struct scsi_per_res_in_rsrv *res;
7843                 int tmp_len, header_only;
7844
7845                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7846
7847                 scsi_ulto4b(lun->PRGeneration, res->header.generation);
7848
7849                 if (lun->flags & CTL_LUN_PR_RESERVED)
7850                 {
7851                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7852                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7853                                     res->header.length);
7854                         header_only = 0;
7855                 } else {
7856                         tmp_len = sizeof(struct scsi_per_res_in_header);
7857                         scsi_ulto4b(0, res->header.length);
7858                         header_only = 1;
7859                 }
7860
7861                 /*
7862                  * We had to drop the lock to allocate our buffer, which
7863                  * leaves time for someone to come in with another
7864                  * persistent reservation.  (That is unlikely, though,
7865                  * since this should be the only persistent reservation
7866                  * command active right now.)
7867                  */
7868                 if (tmp_len != total_len) {
7869                         mtx_unlock(&lun->lun_lock);
7870                         free(ctsio->kern_data_ptr, M_CTL);
7871                         printf("%s: reservation status changed, retrying\n",
7872                                __func__);
7873                         goto retry;
7874                 }
7875
7876                 /*
7877                  * No reservation held, so we're done.
7878                  */
7879                 if (header_only != 0)
7880                         break;
7881
7882                 /*
7883                  * If the registration is an All Registrants type, the key
7884                  * is 0, since it doesn't really matter.
7885                  */
7886                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7887                         memcpy(res->data.reservation,
7888                                &lun->per_res[lun->pr_res_idx].res_key,
7889                                sizeof(struct scsi_per_res_key));
7890                 }
7891                 res->data.scopetype = lun->res_type;
7892                 break;
7893         }
7894         case SPRI_RC:     //report capabilities
7895         {
7896                 struct scsi_per_res_cap *res_cap;
7897                 uint16_t type_mask;
7898
7899                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7900                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7901                 res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7902                 type_mask = SPRI_TM_WR_EX_AR |
7903                             SPRI_TM_EX_AC_RO |
7904                             SPRI_TM_WR_EX_RO |
7905                             SPRI_TM_EX_AC |
7906                             SPRI_TM_WR_EX |
7907                             SPRI_TM_EX_AC_AR;
7908                 scsi_ulto2b(type_mask, res_cap->type_mask);
7909                 break;
7910         }
7911         case SPRI_RS: { // read full status
7912                 struct scsi_per_res_in_full *res_status;
7913                 struct scsi_per_res_in_full_desc *res_desc;
7914                 struct ctl_port *port;
7915                 int i, len;
7916
7917                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7918
7919                 /*
7920                  * We had to drop the lock to allocate our buffer, which
7921                  * leaves time for someone to come in with another
7922                  * persistent reservation.  (That is unlikely, though,
7923                  * since this should be the only persistent reservation
7924                  * command active right now.)
7925                  */
7926                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7927                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7928                      lun->pr_key_count)){
7929                         mtx_unlock(&lun->lun_lock);
7930                         free(ctsio->kern_data_ptr, M_CTL);
7931                         printf("%s: reservation length changed, retrying\n",
7932                                __func__);
7933                         goto retry;
7934                 }
7935
7936                 scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7937
7938                 res_desc = &res_status->desc[0];
7939                 for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7940                         if (!lun->per_res[i].registered)
7941                                 continue;
7942
7943                         memcpy(&res_desc->res_key, &lun->per_res[i].res_key.key,
7944                             sizeof(res_desc->res_key));
7945                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7946                             (lun->pr_res_idx == i ||
7947                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7948                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7949                                 res_desc->scopetype = lun->res_type;
7950                         }
7951                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7952                             res_desc->rel_trgt_port_id);
7953                         len = 0;
7954                         port = softc->ctl_ports[
7955                             ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7956                         if (port != NULL)
7957                                 len = ctl_create_iid(port,
7958                                     i % CTL_MAX_INIT_PER_PORT,
7959                                     res_desc->transport_id);
7960                         scsi_ulto4b(len, res_desc->additional_length);
7961                         res_desc = (struct scsi_per_res_in_full_desc *)
7962                             &res_desc->transport_id[len];
7963                 }
7964                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7965                     res_status->header.length);
7966                 break;
7967         }
7968         default:
7969                 /*
7970                  * This is a bug, because we just checked for this above,
7971                  * and should have returned an error.
7972                  */
7973                 panic("Invalid PR type %x", cdb->action);
7974                 break; /* NOTREACHED */
7975         }
7976         mtx_unlock(&lun->lun_lock);
7977
7978         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7979         ctsio->be_move_done = ctl_config_move_done;
7980
7981         CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7982                          ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7983                          ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7984                          ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7985                          ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7986
7987         ctl_datamove((union ctl_io *)ctsio);
7988
7989         return (CTL_RETVAL_COMPLETE);
7990 }
7991
7992 /*
7993  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7994  * it should return.
7995  */
7996 static int
7997 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7998                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7999                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
8000                 struct scsi_per_res_out_parms* param)
8001 {
8002         union ctl_ha_msg persis_io;
8003         int retval, i;
8004         int isc_retval;
8005
8006         retval = 0;
8007
8008         mtx_lock(&lun->lun_lock);
8009         if (sa_res_key == 0) {
8010                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8011                         /* validate scope and type */
8012                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8013                              SPR_LU_SCOPE) {
8014                                 mtx_unlock(&lun->lun_lock);
8015                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8016                                                       /*sks_valid*/ 1,
8017                                                       /*command*/ 1,
8018                                                       /*field*/ 2,
8019                                                       /*bit_valid*/ 1,
8020                                                       /*bit*/ 4);
8021                                 ctl_done((union ctl_io *)ctsio);
8022                                 return (1);
8023                         }
8024
8025                         if (type>8 || type==2 || type==4 || type==0) {
8026                                 mtx_unlock(&lun->lun_lock);
8027                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8028                                                       /*sks_valid*/ 1,
8029                                                       /*command*/ 1,
8030                                                       /*field*/ 2,
8031                                                       /*bit_valid*/ 1,
8032                                                       /*bit*/ 0);
8033                                 ctl_done((union ctl_io *)ctsio);
8034                                 return (1);
8035                         }
8036
8037                         /* temporarily unregister this nexus */
8038                         lun->per_res[residx].registered = 0;
8039
8040                         /*
8041                          * Unregister everybody else and build UA for
8042                          * them
8043                          */
8044                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8045                                 if (lun->per_res[i].registered == 0)
8046                                         continue;
8047
8048                                 if (!persis_offset
8049                                  && i <CTL_MAX_INITIATORS)
8050                                         lun->pending_ua[i] |=
8051                                                 CTL_UA_REG_PREEMPT;
8052                                 else if (persis_offset
8053                                       && i >= persis_offset)
8054                                         lun->pending_ua[i-persis_offset] |=
8055                                                 CTL_UA_REG_PREEMPT;
8056                                 lun->per_res[i].registered = 0;
8057                                 memset(&lun->per_res[i].res_key, 0,
8058                                        sizeof(struct scsi_per_res_key));
8059                         }
8060                         lun->per_res[residx].registered = 1;
8061                         lun->pr_key_count = 1;
8062                         lun->res_type = type;
8063                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8064                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8065                                 lun->pr_res_idx = residx;
8066
8067                         /* send msg to other side */
8068                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8069                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8070                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8071                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8072                         persis_io.pr.pr_info.res_type = type;
8073                         memcpy(persis_io.pr.pr_info.sa_res_key,
8074                                param->serv_act_res_key,
8075                                sizeof(param->serv_act_res_key));
8076                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8077                              &persis_io, sizeof(persis_io), 0)) >
8078                              CTL_HA_STATUS_SUCCESS) {
8079                                 printf("CTL:Persis Out error returned "
8080                                        "from ctl_ha_msg_send %d\n",
8081                                        isc_retval);
8082                         }
8083                 } else {
8084                         /* not all registrants */
8085                         mtx_unlock(&lun->lun_lock);
8086                         free(ctsio->kern_data_ptr, M_CTL);
8087                         ctl_set_invalid_field(ctsio,
8088                                               /*sks_valid*/ 1,
8089                                               /*command*/ 0,
8090                                               /*field*/ 8,
8091                                               /*bit_valid*/ 0,
8092                                               /*bit*/ 0);
8093                         ctl_done((union ctl_io *)ctsio);
8094                         return (1);
8095                 }
8096         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8097                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
8098                 int found = 0;
8099
8100                 if (res_key == sa_res_key) {
8101                         /* special case */
8102                         /*
8103                          * The spec implies this is not good but doesn't
8104                          * say what to do. There are two choices either
8105                          * generate a res conflict or check condition
8106                          * with illegal field in parameter data. Since
8107                          * that is what is done when the sa_res_key is
8108                          * zero I'll take that approach since this has
8109                          * to do with the sa_res_key.
8110                          */
8111                         mtx_unlock(&lun->lun_lock);
8112                         free(ctsio->kern_data_ptr, M_CTL);
8113                         ctl_set_invalid_field(ctsio,
8114                                               /*sks_valid*/ 1,
8115                                               /*command*/ 0,
8116                                               /*field*/ 8,
8117                                               /*bit_valid*/ 0,
8118                                               /*bit*/ 0);
8119                         ctl_done((union ctl_io *)ctsio);
8120                         return (1);
8121                 }
8122
8123                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8124                         if (lun->per_res[i].registered
8125                          && memcmp(param->serv_act_res_key,
8126                             lun->per_res[i].res_key.key,
8127                             sizeof(struct scsi_per_res_key)) != 0)
8128                                 continue;
8129
8130                         found = 1;
8131                         lun->per_res[i].registered = 0;
8132                         memset(&lun->per_res[i].res_key, 0,
8133                                sizeof(struct scsi_per_res_key));
8134                         lun->pr_key_count--;
8135
8136                         if (!persis_offset && i < CTL_MAX_INITIATORS)
8137                                 lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8138                         else if (persis_offset && i >= persis_offset)
8139                                 lun->pending_ua[i-persis_offset] |=
8140                                         CTL_UA_REG_PREEMPT;
8141                 }
8142                 if (!found) {
8143                         mtx_unlock(&lun->lun_lock);
8144                         free(ctsio->kern_data_ptr, M_CTL);
8145                         ctl_set_reservation_conflict(ctsio);
8146                         ctl_done((union ctl_io *)ctsio);
8147                         return (CTL_RETVAL_COMPLETE);
8148                 }
8149                 /* send msg to other side */
8150                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8151                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8152                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8153                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
8154                 persis_io.pr.pr_info.res_type = type;
8155                 memcpy(persis_io.pr.pr_info.sa_res_key,
8156                        param->serv_act_res_key,
8157                        sizeof(param->serv_act_res_key));
8158                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8159                      &persis_io, sizeof(persis_io), 0)) >
8160                      CTL_HA_STATUS_SUCCESS) {
8161                         printf("CTL:Persis Out error returned from "
8162                                "ctl_ha_msg_send %d\n", isc_retval);
8163                 }
8164         } else {
8165                 /* Reserved but not all registrants */
8166                 /* sa_res_key is res holder */
8167                 if (memcmp(param->serv_act_res_key,
8168                    lun->per_res[lun->pr_res_idx].res_key.key,
8169                    sizeof(struct scsi_per_res_key)) == 0) {
8170                         /* validate scope and type */
8171                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8172                              SPR_LU_SCOPE) {
8173                                 mtx_unlock(&lun->lun_lock);
8174                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8175                                                       /*sks_valid*/ 1,
8176                                                       /*command*/ 1,
8177                                                       /*field*/ 2,
8178                                                       /*bit_valid*/ 1,
8179                                                       /*bit*/ 4);
8180                                 ctl_done((union ctl_io *)ctsio);
8181                                 return (1);
8182                         }
8183
8184                         if (type>8 || type==2 || type==4 || type==0) {
8185                                 mtx_unlock(&lun->lun_lock);
8186                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8187                                                       /*sks_valid*/ 1,
8188                                                       /*command*/ 1,
8189                                                       /*field*/ 2,
8190                                                       /*bit_valid*/ 1,
8191                                                       /*bit*/ 0);
8192                                 ctl_done((union ctl_io *)ctsio);
8193                                 return (1);
8194                         }
8195
8196                         /*
8197                          * Do the following:
8198                          * if sa_res_key != res_key remove all
8199                          * registrants w/sa_res_key and generate UA
8200                          * for these registrants(Registrations
8201                          * Preempted) if it wasn't an exclusive
8202                          * reservation generate UA(Reservations
8203                          * Preempted) for all other registered nexuses
8204                          * if the type has changed. Establish the new
8205                          * reservation and holder. If res_key and
8206                          * sa_res_key are the same do the above
8207                          * except don't unregister the res holder.
8208                          */
8209
8210                         /*
8211                          * Temporarily unregister so it won't get
8212                          * removed or UA generated
8213                          */
8214                         lun->per_res[residx].registered = 0;
8215                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8216                                 if (lun->per_res[i].registered == 0)
8217                                         continue;
8218
8219                                 if (memcmp(param->serv_act_res_key,
8220                                     lun->per_res[i].res_key.key,
8221                                     sizeof(struct scsi_per_res_key)) == 0) {
8222                                         lun->per_res[i].registered = 0;
8223                                         memset(&lun->per_res[i].res_key,
8224                                                0,
8225                                                sizeof(struct scsi_per_res_key));
8226                                         lun->pr_key_count--;
8227
8228                                         if (!persis_offset
8229                                          && i < CTL_MAX_INITIATORS)
8230                                                 lun->pending_ua[i] |=
8231                                                         CTL_UA_REG_PREEMPT;
8232                                         else if (persis_offset
8233                                               && i >= persis_offset)
8234                                                 lun->pending_ua[i-persis_offset] |=
8235                                                   CTL_UA_REG_PREEMPT;
8236                                 } else if (type != lun->res_type
8237                                         && (lun->res_type == SPR_TYPE_WR_EX_RO
8238                                          || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8239                                                 if (!persis_offset
8240                                                  && i < CTL_MAX_INITIATORS)
8241                                                         lun->pending_ua[i] |=
8242                                                         CTL_UA_RES_RELEASE;
8243                                                 else if (persis_offset
8244                                                       && i >= persis_offset)
8245                                                         lun->pending_ua[
8246                                                         i-persis_offset] |=
8247                                                         CTL_UA_RES_RELEASE;
8248                                 }
8249                         }
8250                         lun->per_res[residx].registered = 1;
8251                         lun->res_type = type;
8252                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8253                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8254                                 lun->pr_res_idx = residx;
8255                         else
8256                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8257
8258                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8259                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8260                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8261                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8262                         persis_io.pr.pr_info.res_type = type;
8263                         memcpy(persis_io.pr.pr_info.sa_res_key,
8264                                param->serv_act_res_key,
8265                                sizeof(param->serv_act_res_key));
8266                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8267                              &persis_io, sizeof(persis_io), 0)) >
8268                              CTL_HA_STATUS_SUCCESS) {
8269                                 printf("CTL:Persis Out error returned "
8270                                        "from ctl_ha_msg_send %d\n",
8271                                        isc_retval);
8272                         }
8273                 } else {
8274                         /*
8275                          * sa_res_key is not the res holder just
8276                          * remove registrants
8277                          */
8278                         int found=0;
8279
8280                         for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8281                                 if (memcmp(param->serv_act_res_key,
8282                                     lun->per_res[i].res_key.key,
8283                                     sizeof(struct scsi_per_res_key)) != 0)
8284                                         continue;
8285
8286                                 found = 1;
8287                                 lun->per_res[i].registered = 0;
8288                                 memset(&lun->per_res[i].res_key, 0,
8289                                        sizeof(struct scsi_per_res_key));
8290                                 lun->pr_key_count--;
8291
8292                                 if (!persis_offset
8293                                  && i < CTL_MAX_INITIATORS)
8294                                         lun->pending_ua[i] |=
8295                                                 CTL_UA_REG_PREEMPT;
8296                                 else if (persis_offset
8297                                       && i >= persis_offset)
8298                                         lun->pending_ua[i-persis_offset] |=
8299                                                 CTL_UA_REG_PREEMPT;
8300                         }
8301
8302                         if (!found) {
8303                                 mtx_unlock(&lun->lun_lock);
8304                                 free(ctsio->kern_data_ptr, M_CTL);
8305                                 ctl_set_reservation_conflict(ctsio);
8306                                 ctl_done((union ctl_io *)ctsio);
8307                                 return (1);
8308                         }
8309                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8310                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8311                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8312                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8313                         persis_io.pr.pr_info.res_type = type;
8314                         memcpy(persis_io.pr.pr_info.sa_res_key,
8315                                param->serv_act_res_key,
8316                                sizeof(param->serv_act_res_key));
8317                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8318                              &persis_io, sizeof(persis_io), 0)) >
8319                              CTL_HA_STATUS_SUCCESS) {
8320                                 printf("CTL:Persis Out error returned "
8321                                        "from ctl_ha_msg_send %d\n",
8322                                 isc_retval);
8323                         }
8324                 }
8325         }
8326
8327         lun->PRGeneration++;
8328         mtx_unlock(&lun->lun_lock);
8329
8330         return (retval);
8331 }
8332
8333 static void
8334 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8335 {
8336         int i;
8337
8338         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8339          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8340          || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8341                    msg->pr.pr_info.sa_res_key,
8342                    sizeof(struct scsi_per_res_key)) != 0) {
8343                 uint64_t sa_res_key;
8344                 sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8345
8346                 if (sa_res_key == 0) {
8347                         /* temporarily unregister this nexus */
8348                         lun->per_res[msg->pr.pr_info.residx].registered = 0;
8349
8350                         /*
8351                          * Unregister everybody else and build UA for
8352                          * them
8353                          */
8354                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8355                                 if (lun->per_res[i].registered == 0)
8356                                         continue;
8357
8358                                 if (!persis_offset
8359                                  && i < CTL_MAX_INITIATORS)
8360                                         lun->pending_ua[i] |=
8361                                                 CTL_UA_REG_PREEMPT;
8362                                 else if (persis_offset && i >= persis_offset)
8363                                         lun->pending_ua[i - persis_offset] |=
8364                                                 CTL_UA_REG_PREEMPT;
8365                                 lun->per_res[i].registered = 0;
8366                                 memset(&lun->per_res[i].res_key, 0,
8367                                        sizeof(struct scsi_per_res_key));
8368                         }
8369
8370                         lun->per_res[msg->pr.pr_info.residx].registered = 1;
8371                         lun->pr_key_count = 1;
8372                         lun->res_type = msg->pr.pr_info.res_type;
8373                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8374                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8375                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8376                 } else {
8377                         for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8378                                 if (memcmp(msg->pr.pr_info.sa_res_key,
8379                                    lun->per_res[i].res_key.key,
8380                                    sizeof(struct scsi_per_res_key)) != 0)
8381                                         continue;
8382
8383                                 lun->per_res[i].registered = 0;
8384                                 memset(&lun->per_res[i].res_key, 0,
8385                                        sizeof(struct scsi_per_res_key));
8386                                 lun->pr_key_count--;
8387
8388                                 if (!persis_offset
8389                                  && i < persis_offset)
8390                                         lun->pending_ua[i] |=
8391                                                 CTL_UA_REG_PREEMPT;
8392                                 else if (persis_offset
8393                                       && i >= persis_offset)
8394                                         lun->pending_ua[i - persis_offset] |=
8395                                                 CTL_UA_REG_PREEMPT;
8396                         }
8397                 }
8398         } else {
8399                 /*
8400                  * Temporarily unregister so it won't get removed
8401                  * or UA generated
8402                  */
8403                 lun->per_res[msg->pr.pr_info.residx].registered = 0;
8404                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8405                         if (lun->per_res[i].registered == 0)
8406                                 continue;
8407
8408                         if (memcmp(msg->pr.pr_info.sa_res_key,
8409                            lun->per_res[i].res_key.key,
8410                            sizeof(struct scsi_per_res_key)) == 0) {
8411                                 lun->per_res[i].registered = 0;
8412                                 memset(&lun->per_res[i].res_key, 0,
8413                                        sizeof(struct scsi_per_res_key));
8414                                 lun->pr_key_count--;
8415                                 if (!persis_offset
8416                                  && i < CTL_MAX_INITIATORS)
8417                                         lun->pending_ua[i] |=
8418                                                 CTL_UA_REG_PREEMPT;
8419                                 else if (persis_offset
8420                                       && i >= persis_offset)
8421                                         lun->pending_ua[i - persis_offset] |=
8422                                                 CTL_UA_REG_PREEMPT;
8423                         } else if (msg->pr.pr_info.res_type != lun->res_type
8424                                 && (lun->res_type == SPR_TYPE_WR_EX_RO
8425                                  || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8426                                         if (!persis_offset
8427                                          && i < persis_offset)
8428                                                 lun->pending_ua[i] |=
8429                                                         CTL_UA_RES_RELEASE;
8430                                         else if (persis_offset
8431                                               && i >= persis_offset)
8432                                         lun->pending_ua[i - persis_offset] |=
8433                                                 CTL_UA_RES_RELEASE;
8434                         }
8435                 }
8436                 lun->per_res[msg->pr.pr_info.residx].registered = 1;
8437                 lun->res_type = msg->pr.pr_info.res_type;
8438                 if (lun->res_type != SPR_TYPE_WR_EX_AR
8439                  && lun->res_type != SPR_TYPE_EX_AC_AR)
8440                         lun->pr_res_idx = msg->pr.pr_info.residx;
8441                 else
8442                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8443         }
8444         lun->PRGeneration++;
8445
8446 }
8447
8448
8449 int
8450 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8451 {
8452         int retval;
8453         int isc_retval;
8454         u_int32_t param_len;
8455         struct scsi_per_res_out *cdb;
8456         struct ctl_lun *lun;
8457         struct scsi_per_res_out_parms* param;
8458         struct ctl_softc *softc;
8459         uint32_t residx;
8460         uint64_t res_key, sa_res_key;
8461         uint8_t type;
8462         union ctl_ha_msg persis_io;
8463         int    i;
8464
8465         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8466
8467         retval = CTL_RETVAL_COMPLETE;
8468
8469         softc = control_softc;
8470
8471         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8472         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8473
8474         /*
8475          * We only support whole-LUN scope.  The scope & type are ignored for
8476          * register, register and ignore existing key and clear.
8477          * We sometimes ignore scope and type on preempts too!!
8478          * Verify reservation type here as well.
8479          */
8480         type = cdb->scope_type & SPR_TYPE_MASK;
8481         if ((cdb->action == SPRO_RESERVE)
8482          || (cdb->action == SPRO_RELEASE)) {
8483                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8484                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8485                                               /*sks_valid*/ 1,
8486                                               /*command*/ 1,
8487                                               /*field*/ 2,
8488                                               /*bit_valid*/ 1,
8489                                               /*bit*/ 4);
8490                         ctl_done((union ctl_io *)ctsio);
8491                         return (CTL_RETVAL_COMPLETE);
8492                 }
8493
8494                 if (type>8 || type==2 || type==4 || type==0) {
8495                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8496                                               /*sks_valid*/ 1,
8497                                               /*command*/ 1,
8498                                               /*field*/ 2,
8499                                               /*bit_valid*/ 1,
8500                                               /*bit*/ 0);
8501                         ctl_done((union ctl_io *)ctsio);
8502                         return (CTL_RETVAL_COMPLETE);
8503                 }
8504         }
8505
8506         param_len = scsi_4btoul(cdb->length);
8507
8508         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8509                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8510                 ctsio->kern_data_len = param_len;
8511                 ctsio->kern_total_len = param_len;
8512                 ctsio->kern_data_resid = 0;
8513                 ctsio->kern_rel_offset = 0;
8514                 ctsio->kern_sg_entries = 0;
8515                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8516                 ctsio->be_move_done = ctl_config_move_done;
8517                 ctl_datamove((union ctl_io *)ctsio);
8518
8519                 return (CTL_RETVAL_COMPLETE);
8520         }
8521
8522         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8523
8524         residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8525         res_key = scsi_8btou64(param->res_key.key);
8526         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8527
8528         /*
8529          * Validate the reservation key here except for SPRO_REG_IGNO
8530          * This must be done for all other service actions
8531          */
8532         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8533                 mtx_lock(&lun->lun_lock);
8534                 if (lun->per_res[residx].registered) {
8535                     if (memcmp(param->res_key.key,
8536                                lun->per_res[residx].res_key.key,
8537                                ctl_min(sizeof(param->res_key),
8538                                sizeof(lun->per_res[residx].res_key))) != 0) {
8539                                 /*
8540                                  * The current key passed in doesn't match
8541                                  * the one the initiator previously
8542                                  * registered.
8543                                  */
8544                                 mtx_unlock(&lun->lun_lock);
8545                                 free(ctsio->kern_data_ptr, M_CTL);
8546                                 ctl_set_reservation_conflict(ctsio);
8547                                 ctl_done((union ctl_io *)ctsio);
8548                                 return (CTL_RETVAL_COMPLETE);
8549                         }
8550                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8551                         /*
8552                          * We are not registered
8553                          */
8554                         mtx_unlock(&lun->lun_lock);
8555                         free(ctsio->kern_data_ptr, M_CTL);
8556                         ctl_set_reservation_conflict(ctsio);
8557                         ctl_done((union ctl_io *)ctsio);
8558                         return (CTL_RETVAL_COMPLETE);
8559                 } else if (res_key != 0) {
8560                         /*
8561                          * We are not registered and trying to register but
8562                          * the register key isn't zero.
8563                          */
8564                         mtx_unlock(&lun->lun_lock);
8565                         free(ctsio->kern_data_ptr, M_CTL);
8566                         ctl_set_reservation_conflict(ctsio);
8567                         ctl_done((union ctl_io *)ctsio);
8568                         return (CTL_RETVAL_COMPLETE);
8569                 }
8570                 mtx_unlock(&lun->lun_lock);
8571         }
8572
8573         switch (cdb->action & SPRO_ACTION_MASK) {
8574         case SPRO_REGISTER:
8575         case SPRO_REG_IGNO: {
8576
8577 #if 0
8578                 printf("Registration received\n");
8579 #endif
8580
8581                 /*
8582                  * We don't support any of these options, as we report in
8583                  * the read capabilities request (see
8584                  * ctl_persistent_reserve_in(), above).
8585                  */
8586                 if ((param->flags & SPR_SPEC_I_PT)
8587                  || (param->flags & SPR_ALL_TG_PT)
8588                  || (param->flags & SPR_APTPL)) {
8589                         int bit_ptr;
8590
8591                         if (param->flags & SPR_APTPL)
8592                                 bit_ptr = 0;
8593                         else if (param->flags & SPR_ALL_TG_PT)
8594                                 bit_ptr = 2;
8595                         else /* SPR_SPEC_I_PT */
8596                                 bit_ptr = 3;
8597
8598                         free(ctsio->kern_data_ptr, M_CTL);
8599                         ctl_set_invalid_field(ctsio,
8600                                               /*sks_valid*/ 1,
8601                                               /*command*/ 0,
8602                                               /*field*/ 20,
8603                                               /*bit_valid*/ 1,
8604                                               /*bit*/ bit_ptr);
8605                         ctl_done((union ctl_io *)ctsio);
8606                         return (CTL_RETVAL_COMPLETE);
8607                 }
8608
8609                 mtx_lock(&lun->lun_lock);
8610
8611                 /*
8612                  * The initiator wants to clear the
8613                  * key/unregister.
8614                  */
8615                 if (sa_res_key == 0) {
8616                         if ((res_key == 0
8617                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8618                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8619                           && !lun->per_res[residx].registered)) {
8620                                 mtx_unlock(&lun->lun_lock);
8621                                 goto done;
8622                         }
8623
8624                         lun->per_res[residx].registered = 0;
8625                         memset(&lun->per_res[residx].res_key,
8626                                0, sizeof(lun->per_res[residx].res_key));
8627                         lun->pr_key_count--;
8628
8629                         if (residx == lun->pr_res_idx) {
8630                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8631                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8632
8633                                 if ((lun->res_type == SPR_TYPE_WR_EX_RO
8634                                   || lun->res_type == SPR_TYPE_EX_AC_RO)
8635                                  && lun->pr_key_count) {
8636                                         /*
8637                                          * If the reservation is a registrants
8638                                          * only type we need to generate a UA
8639                                          * for other registered inits.  The
8640                                          * sense code should be RESERVATIONS
8641                                          * RELEASED
8642                                          */
8643
8644                                         for (i = 0; i < CTL_MAX_INITIATORS;i++){
8645                                                 if (lun->per_res[
8646                                                     i+persis_offset].registered
8647                                                     == 0)
8648                                                         continue;
8649                                                 lun->pending_ua[i] |=
8650                                                         CTL_UA_RES_RELEASE;
8651                                         }
8652                                 }
8653                                 lun->res_type = 0;
8654                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8655                                 if (lun->pr_key_count==0) {
8656                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8657                                         lun->res_type = 0;
8658                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8659                                 }
8660                         }
8661                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8662                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8663                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8664                         persis_io.pr.pr_info.residx = residx;
8665                         if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8666                              &persis_io, sizeof(persis_io), 0 )) >
8667                              CTL_HA_STATUS_SUCCESS) {
8668                                 printf("CTL:Persis Out error returned from "
8669                                        "ctl_ha_msg_send %d\n", isc_retval);
8670                         }
8671                 } else /* sa_res_key != 0 */ {
8672
8673                         /*
8674                          * If we aren't registered currently then increment
8675                          * the key count and set the registered flag.
8676                          */
8677                         if (!lun->per_res[residx].registered) {
8678                                 lun->pr_key_count++;
8679                                 lun->per_res[residx].registered = 1;
8680                         }
8681
8682                         memcpy(&lun->per_res[residx].res_key,
8683                                param->serv_act_res_key,
8684                                ctl_min(sizeof(param->serv_act_res_key),
8685                                sizeof(lun->per_res[residx].res_key)));
8686
8687                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8688                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8689                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8690                         persis_io.pr.pr_info.residx = residx;
8691                         memcpy(persis_io.pr.pr_info.sa_res_key,
8692                                param->serv_act_res_key,
8693                                sizeof(param->serv_act_res_key));
8694                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8695                              &persis_io, sizeof(persis_io), 0)) >
8696                              CTL_HA_STATUS_SUCCESS) {
8697                                 printf("CTL:Persis Out error returned from "
8698                                        "ctl_ha_msg_send %d\n", isc_retval);
8699                         }
8700                 }
8701                 lun->PRGeneration++;
8702                 mtx_unlock(&lun->lun_lock);
8703
8704                 break;
8705         }
8706         case SPRO_RESERVE:
8707 #if 0
8708                 printf("Reserve executed type %d\n", type);
8709 #endif
8710                 mtx_lock(&lun->lun_lock);
8711                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8712                         /*
8713                          * if this isn't the reservation holder and it's
8714                          * not a "all registrants" type or if the type is
8715                          * different then we have a conflict
8716                          */
8717                         if ((lun->pr_res_idx != residx
8718                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8719                          || lun->res_type != type) {
8720                                 mtx_unlock(&lun->lun_lock);
8721                                 free(ctsio->kern_data_ptr, M_CTL);
8722                                 ctl_set_reservation_conflict(ctsio);
8723                                 ctl_done((union ctl_io *)ctsio);
8724                                 return (CTL_RETVAL_COMPLETE);
8725                         }
8726                         mtx_unlock(&lun->lun_lock);
8727                 } else /* create a reservation */ {
8728                         /*
8729                          * If it's not an "all registrants" type record
8730                          * reservation holder
8731                          */
8732                         if (type != SPR_TYPE_WR_EX_AR
8733                          && type != SPR_TYPE_EX_AC_AR)
8734                                 lun->pr_res_idx = residx; /* Res holder */
8735                         else
8736                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8737
8738                         lun->flags |= CTL_LUN_PR_RESERVED;
8739                         lun->res_type = type;
8740
8741                         mtx_unlock(&lun->lun_lock);
8742
8743                         /* send msg to other side */
8744                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8745                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8746                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8747                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8748                         persis_io.pr.pr_info.res_type = type;
8749                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8750                              &persis_io, sizeof(persis_io), 0)) >
8751                              CTL_HA_STATUS_SUCCESS) {
8752                                 printf("CTL:Persis Out error returned from "
8753                                        "ctl_ha_msg_send %d\n", isc_retval);
8754                         }
8755                 }
8756                 break;
8757
8758         case SPRO_RELEASE:
8759                 mtx_lock(&lun->lun_lock);
8760                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8761                         /* No reservation exists return good status */
8762                         mtx_unlock(&lun->lun_lock);
8763                         goto done;
8764                 }
8765                 /*
8766                  * Is this nexus a reservation holder?
8767                  */
8768                 if (lun->pr_res_idx != residx
8769                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8770                         /*
8771                          * not a res holder return good status but
8772                          * do nothing
8773                          */
8774                         mtx_unlock(&lun->lun_lock);
8775                         goto done;
8776                 }
8777
8778                 if (lun->res_type != type) {
8779                         mtx_unlock(&lun->lun_lock);
8780                         free(ctsio->kern_data_ptr, M_CTL);
8781                         ctl_set_illegal_pr_release(ctsio);
8782                         ctl_done((union ctl_io *)ctsio);
8783                         return (CTL_RETVAL_COMPLETE);
8784                 }
8785
8786                 /* okay to release */
8787                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8788                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8789                 lun->res_type = 0;
8790
8791                 /*
8792                  * if this isn't an exclusive access
8793                  * res generate UA for all other
8794                  * registrants.
8795                  */
8796                 if (type != SPR_TYPE_EX_AC
8797                  && type != SPR_TYPE_WR_EX) {
8798                         /*
8799                          * temporarily unregister so we don't generate UA
8800                          */
8801                         lun->per_res[residx].registered = 0;
8802
8803                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8804                                 if (lun->per_res[i+persis_offset].registered
8805                                     == 0)
8806                                         continue;
8807                                 lun->pending_ua[i] |=
8808                                         CTL_UA_RES_RELEASE;
8809                         }
8810
8811                         lun->per_res[residx].registered = 1;
8812                 }
8813                 mtx_unlock(&lun->lun_lock);
8814                 /* Send msg to other side */
8815                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8816                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8817                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8818                 if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8819                      sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8820                         printf("CTL:Persis Out error returned from "
8821                                "ctl_ha_msg_send %d\n", isc_retval);
8822                 }
8823                 break;
8824
8825         case SPRO_CLEAR:
8826                 /* send msg to other side */
8827
8828                 mtx_lock(&lun->lun_lock);
8829                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8830                 lun->res_type = 0;
8831                 lun->pr_key_count = 0;
8832                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8833
8834
8835                 memset(&lun->per_res[residx].res_key,
8836                        0, sizeof(lun->per_res[residx].res_key));
8837                 lun->per_res[residx].registered = 0;
8838
8839                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8840                         if (lun->per_res[i].registered) {
8841                                 if (!persis_offset && i < CTL_MAX_INITIATORS)
8842                                         lun->pending_ua[i] |=
8843                                                 CTL_UA_RES_PREEMPT;
8844                                 else if (persis_offset && i >= persis_offset)
8845                                         lun->pending_ua[i-persis_offset] |=
8846                                             CTL_UA_RES_PREEMPT;
8847
8848                                 memset(&lun->per_res[i].res_key,
8849                                        0, sizeof(struct scsi_per_res_key));
8850                                 lun->per_res[i].registered = 0;
8851                         }
8852                 lun->PRGeneration++;
8853                 mtx_unlock(&lun->lun_lock);
8854                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8855                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8856                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8857                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8858                      sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8859                         printf("CTL:Persis Out error returned from "
8860                                "ctl_ha_msg_send %d\n", isc_retval);
8861                 }
8862                 break;
8863
8864         case SPRO_PREEMPT: {
8865                 int nretval;
8866
8867                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8868                                           residx, ctsio, cdb, param);
8869                 if (nretval != 0)
8870                         return (CTL_RETVAL_COMPLETE);
8871                 break;
8872         }
8873         default:
8874                 panic("Invalid PR type %x", cdb->action);
8875         }
8876
8877 done:
8878         free(ctsio->kern_data_ptr, M_CTL);
8879         ctl_set_success(ctsio);
8880         ctl_done((union ctl_io *)ctsio);
8881
8882         return (retval);
8883 }
8884
8885 /*
8886  * This routine is for handling a message from the other SC pertaining to
8887  * persistent reserve out. All the error checking will have been done
8888  * so only perorming the action need be done here to keep the two
8889  * in sync.
8890  */
8891 static void
8892 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8893 {
8894         struct ctl_lun *lun;
8895         struct ctl_softc *softc;
8896         int i;
8897         uint32_t targ_lun;
8898
8899         softc = control_softc;
8900
8901         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8902         lun = softc->ctl_luns[targ_lun];
8903         mtx_lock(&lun->lun_lock);
8904         switch(msg->pr.pr_info.action) {
8905         case CTL_PR_REG_KEY:
8906                 if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8907                         lun->per_res[msg->pr.pr_info.residx].registered = 1;
8908                         lun->pr_key_count++;
8909                 }
8910                 lun->PRGeneration++;
8911                 memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8912                        msg->pr.pr_info.sa_res_key,
8913                        sizeof(struct scsi_per_res_key));
8914                 break;
8915
8916         case CTL_PR_UNREG_KEY:
8917                 lun->per_res[msg->pr.pr_info.residx].registered = 0;
8918                 memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8919                        0, sizeof(struct scsi_per_res_key));
8920                 lun->pr_key_count--;
8921
8922                 /* XXX Need to see if the reservation has been released */
8923                 /* if so do we need to generate UA? */
8924                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8925                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8926                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8927
8928                         if ((lun->res_type == SPR_TYPE_WR_EX_RO
8929                           || lun->res_type == SPR_TYPE_EX_AC_RO)
8930                          && lun->pr_key_count) {
8931                                 /*
8932                                  * If the reservation is a registrants
8933                                  * only type we need to generate a UA
8934                                  * for other registered inits.  The
8935                                  * sense code should be RESERVATIONS
8936                                  * RELEASED
8937                                  */
8938
8939                                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8940                                         if (lun->per_res[i+
8941                                             persis_offset].registered == 0)
8942                                                 continue;
8943
8944                                         lun->pending_ua[i] |=
8945                                                 CTL_UA_RES_RELEASE;
8946                                 }
8947                         }
8948                         lun->res_type = 0;
8949                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8950                         if (lun->pr_key_count==0) {
8951                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8952                                 lun->res_type = 0;
8953                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8954                         }
8955                 }
8956                 lun->PRGeneration++;
8957                 break;
8958
8959         case CTL_PR_RESERVE:
8960                 lun->flags |= CTL_LUN_PR_RESERVED;
8961                 lun->res_type = msg->pr.pr_info.res_type;
8962                 lun->pr_res_idx = msg->pr.pr_info.residx;
8963
8964                 break;
8965
8966         case CTL_PR_RELEASE:
8967                 /*
8968                  * if this isn't an exclusive access res generate UA for all
8969                  * other registrants.
8970                  */
8971                 if (lun->res_type != SPR_TYPE_EX_AC
8972                  && lun->res_type != SPR_TYPE_WR_EX) {
8973                         for (i = 0; i < CTL_MAX_INITIATORS; i++)
8974                                 if (lun->per_res[i+persis_offset].registered)
8975                                         lun->pending_ua[i] |=
8976                                                 CTL_UA_RES_RELEASE;
8977                 }
8978
8979                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8980                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8981                 lun->res_type = 0;
8982                 break;
8983
8984         case CTL_PR_PREEMPT:
8985                 ctl_pro_preempt_other(lun, msg);
8986                 break;
8987         case CTL_PR_CLEAR:
8988                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8989                 lun->res_type = 0;
8990                 lun->pr_key_count = 0;
8991                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8992
8993                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8994                         if (lun->per_res[i].registered == 0)
8995                                 continue;
8996                         if (!persis_offset
8997                          && i < CTL_MAX_INITIATORS)
8998                                 lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
8999                         else if (persis_offset
9000                               && i >= persis_offset)
9001                                 lun->pending_ua[i-persis_offset] |=
9002                                         CTL_UA_RES_PREEMPT;
9003                         memset(&lun->per_res[i].res_key, 0,
9004                                sizeof(struct scsi_per_res_key));
9005                         lun->per_res[i].registered = 0;
9006                 }
9007                 lun->PRGeneration++;
9008                 break;
9009         }
9010
9011         mtx_unlock(&lun->lun_lock);
9012 }
9013
9014 int
9015 ctl_read_write(struct ctl_scsiio *ctsio)
9016 {
9017         struct ctl_lun *lun;
9018         struct ctl_lba_len_flags *lbalen;
9019         uint64_t lba;
9020         uint32_t num_blocks;
9021         int flags, retval;
9022         int isread;
9023
9024         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9025
9026         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
9027
9028         flags = 0;
9029         retval = CTL_RETVAL_COMPLETE;
9030
9031         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
9032               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
9033         if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
9034                 uint32_t residx;
9035
9036                 /*
9037                  * XXX KDM need a lock here.
9038                  */
9039                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9040                 if ((lun->res_type == SPR_TYPE_EX_AC
9041                   && residx != lun->pr_res_idx)
9042                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
9043                    || lun->res_type == SPR_TYPE_EX_AC_AR)
9044                   && !lun->per_res[residx].registered)) {
9045                         ctl_set_reservation_conflict(ctsio);
9046                         ctl_done((union ctl_io *)ctsio);
9047                         return (CTL_RETVAL_COMPLETE);
9048                 }
9049         }
9050
9051         switch (ctsio->cdb[0]) {
9052         case READ_6:
9053         case WRITE_6: {
9054                 struct scsi_rw_6 *cdb;
9055
9056                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
9057
9058                 lba = scsi_3btoul(cdb->addr);
9059                 /* only 5 bits are valid in the most significant address byte */
9060                 lba &= 0x1fffff;
9061                 num_blocks = cdb->length;
9062                 /*
9063                  * This is correct according to SBC-2.
9064                  */
9065                 if (num_blocks == 0)
9066                         num_blocks = 256;
9067                 break;
9068         }
9069         case READ_10:
9070         case WRITE_10: {
9071                 struct scsi_rw_10 *cdb;
9072
9073                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
9074                 if (cdb->byte2 & SRW10_FUA)
9075                         flags |= CTL_LLF_FUA;
9076                 if (cdb->byte2 & SRW10_DPO)
9077                         flags |= CTL_LLF_DPO;
9078                 lba = scsi_4btoul(cdb->addr);
9079                 num_blocks = scsi_2btoul(cdb->length);
9080                 break;
9081         }
9082         case WRITE_VERIFY_10: {
9083                 struct scsi_write_verify_10 *cdb;
9084
9085                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9086                 flags |= CTL_LLF_FUA;
9087                 if (cdb->byte2 & SWV_DPO)
9088                         flags |= CTL_LLF_DPO;
9089                 lba = scsi_4btoul(cdb->addr);
9090                 num_blocks = scsi_2btoul(cdb->length);
9091                 break;
9092         }
9093         case READ_12:
9094         case WRITE_12: {
9095                 struct scsi_rw_12 *cdb;
9096
9097                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
9098                 if (cdb->byte2 & SRW12_FUA)
9099                         flags |= CTL_LLF_FUA;
9100                 if (cdb->byte2 & SRW12_DPO)
9101                         flags |= CTL_LLF_DPO;
9102                 lba = scsi_4btoul(cdb->addr);
9103                 num_blocks = scsi_4btoul(cdb->length);
9104                 break;
9105         }
9106         case WRITE_VERIFY_12: {
9107                 struct scsi_write_verify_12 *cdb;
9108
9109                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9110                 flags |= CTL_LLF_FUA;
9111                 if (cdb->byte2 & SWV_DPO)
9112                         flags |= CTL_LLF_DPO;
9113                 lba = scsi_4btoul(cdb->addr);
9114                 num_blocks = scsi_4btoul(cdb->length);
9115                 break;
9116         }
9117         case READ_16:
9118         case WRITE_16: {
9119                 struct scsi_rw_16 *cdb;
9120
9121                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9122                 if (cdb->byte2 & SRW12_FUA)
9123                         flags |= CTL_LLF_FUA;
9124                 if (cdb->byte2 & SRW12_DPO)
9125                         flags |= CTL_LLF_DPO;
9126                 lba = scsi_8btou64(cdb->addr);
9127                 num_blocks = scsi_4btoul(cdb->length);
9128                 break;
9129         }
9130         case WRITE_VERIFY_16: {
9131                 struct scsi_write_verify_16 *cdb;
9132
9133                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9134                 flags |= CTL_LLF_FUA;
9135                 if (cdb->byte2 & SWV_DPO)
9136                         flags |= CTL_LLF_DPO;
9137                 lba = scsi_8btou64(cdb->addr);
9138                 num_blocks = scsi_4btoul(cdb->length);
9139                 break;
9140         }
9141         default:
9142                 /*
9143                  * We got a command we don't support.  This shouldn't
9144                  * happen, commands should be filtered out above us.
9145                  */
9146                 ctl_set_invalid_opcode(ctsio);
9147                 ctl_done((union ctl_io *)ctsio);
9148
9149                 return (CTL_RETVAL_COMPLETE);
9150                 break; /* NOTREACHED */
9151         }
9152
9153         /*
9154          * The first check is to make sure we're in bounds, the second
9155          * check is to catch wrap-around problems.  If the lba + num blocks
9156          * is less than the lba, then we've wrapped around and the block
9157          * range is invalid anyway.
9158          */
9159         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9160          || ((lba + num_blocks) < lba)) {
9161                 ctl_set_lba_out_of_range(ctsio);
9162                 ctl_done((union ctl_io *)ctsio);
9163                 return (CTL_RETVAL_COMPLETE);
9164         }
9165
9166         /*
9167          * According to SBC-3, a transfer length of 0 is not an error.
9168          * Note that this cannot happen with WRITE(6) or READ(6), since 0
9169          * translates to 256 blocks for those commands.
9170          */
9171         if (num_blocks == 0) {
9172                 ctl_set_success(ctsio);
9173                 ctl_done((union ctl_io *)ctsio);
9174                 return (CTL_RETVAL_COMPLETE);
9175         }
9176
9177         /* Set FUA and/or DPO if caches are disabled. */
9178         if (isread) {
9179                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9180                     SCP_RCD) != 0)
9181                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9182         } else {
9183                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9184                     SCP_WCE) == 0)
9185                         flags |= CTL_LLF_FUA;
9186         }
9187
9188         lbalen = (struct ctl_lba_len_flags *)
9189             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9190         lbalen->lba = lba;
9191         lbalen->len = num_blocks;
9192         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9193
9194         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9195         ctsio->kern_rel_offset = 0;
9196
9197         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9198
9199         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9200
9201         return (retval);
9202 }
9203
9204 static int
9205 ctl_cnw_cont(union ctl_io *io)
9206 {
9207         struct ctl_scsiio *ctsio;
9208         struct ctl_lun *lun;
9209         struct ctl_lba_len_flags *lbalen;
9210         int retval;
9211
9212         ctsio = &io->scsiio;
9213         ctsio->io_hdr.status = CTL_STATUS_NONE;
9214         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9215         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9216         lbalen = (struct ctl_lba_len_flags *)
9217             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9218         lbalen->flags &= ~CTL_LLF_COMPARE;
9219         lbalen->flags |= CTL_LLF_WRITE;
9220
9221         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9222         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9223         return (retval);
9224 }
9225
9226 int
9227 ctl_cnw(struct ctl_scsiio *ctsio)
9228 {
9229         struct ctl_lun *lun;
9230         struct ctl_lba_len_flags *lbalen;
9231         uint64_t lba;
9232         uint32_t num_blocks;
9233         int flags, retval;
9234
9235         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9236
9237         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9238
9239         flags = 0;
9240         retval = CTL_RETVAL_COMPLETE;
9241
9242         switch (ctsio->cdb[0]) {
9243         case COMPARE_AND_WRITE: {
9244                 struct scsi_compare_and_write *cdb;
9245
9246                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9247                 if (cdb->byte2 & SRW10_FUA)
9248                         flags |= CTL_LLF_FUA;
9249                 if (cdb->byte2 & SRW10_DPO)
9250                         flags |= CTL_LLF_DPO;
9251                 lba = scsi_8btou64(cdb->addr);
9252                 num_blocks = cdb->length;
9253                 break;
9254         }
9255         default:
9256                 /*
9257                  * We got a command we don't support.  This shouldn't
9258                  * happen, commands should be filtered out above us.
9259                  */
9260                 ctl_set_invalid_opcode(ctsio);
9261                 ctl_done((union ctl_io *)ctsio);
9262
9263                 return (CTL_RETVAL_COMPLETE);
9264                 break; /* NOTREACHED */
9265         }
9266
9267         /*
9268          * The first check is to make sure we're in bounds, the second
9269          * check is to catch wrap-around problems.  If the lba + num blocks
9270          * is less than the lba, then we've wrapped around and the block
9271          * range is invalid anyway.
9272          */
9273         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9274          || ((lba + num_blocks) < lba)) {
9275                 ctl_set_lba_out_of_range(ctsio);
9276                 ctl_done((union ctl_io *)ctsio);
9277                 return (CTL_RETVAL_COMPLETE);
9278         }
9279
9280         /*
9281          * According to SBC-3, a transfer length of 0 is not an error.
9282          */
9283         if (num_blocks == 0) {
9284                 ctl_set_success(ctsio);
9285                 ctl_done((union ctl_io *)ctsio);
9286                 return (CTL_RETVAL_COMPLETE);
9287         }
9288
9289         /* Set FUA if write cache is disabled. */
9290         if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9291             SCP_WCE) == 0)
9292                 flags |= CTL_LLF_FUA;
9293
9294         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9295         ctsio->kern_rel_offset = 0;
9296
9297         /*
9298          * Set the IO_CONT flag, so that if this I/O gets passed to
9299          * ctl_data_submit_done(), it'll get passed back to
9300          * ctl_ctl_cnw_cont() for further processing.
9301          */
9302         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9303         ctsio->io_cont = ctl_cnw_cont;
9304
9305         lbalen = (struct ctl_lba_len_flags *)
9306             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9307         lbalen->lba = lba;
9308         lbalen->len = num_blocks;
9309         lbalen->flags = CTL_LLF_COMPARE | flags;
9310
9311         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9312         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9313         return (retval);
9314 }
9315
9316 int
9317 ctl_verify(struct ctl_scsiio *ctsio)
9318 {
9319         struct ctl_lun *lun;
9320         struct ctl_lba_len_flags *lbalen;
9321         uint64_t lba;
9322         uint32_t num_blocks;
9323         int bytchk, flags;
9324         int retval;
9325
9326         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9327
9328         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9329
9330         bytchk = 0;
9331         flags = CTL_LLF_FUA;
9332         retval = CTL_RETVAL_COMPLETE;
9333
9334         switch (ctsio->cdb[0]) {
9335         case VERIFY_10: {
9336                 struct scsi_verify_10 *cdb;
9337
9338                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
9339                 if (cdb->byte2 & SVFY_BYTCHK)
9340                         bytchk = 1;
9341                 if (cdb->byte2 & SVFY_DPO)
9342                         flags |= CTL_LLF_DPO;
9343                 lba = scsi_4btoul(cdb->addr);
9344                 num_blocks = scsi_2btoul(cdb->length);
9345                 break;
9346         }
9347         case VERIFY_12: {
9348                 struct scsi_verify_12 *cdb;
9349
9350                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
9351                 if (cdb->byte2 & SVFY_BYTCHK)
9352                         bytchk = 1;
9353                 if (cdb->byte2 & SVFY_DPO)
9354                         flags |= CTL_LLF_DPO;
9355                 lba = scsi_4btoul(cdb->addr);
9356                 num_blocks = scsi_4btoul(cdb->length);
9357                 break;
9358         }
9359         case VERIFY_16: {
9360                 struct scsi_rw_16 *cdb;
9361
9362                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9363                 if (cdb->byte2 & SVFY_BYTCHK)
9364                         bytchk = 1;
9365                 if (cdb->byte2 & SVFY_DPO)
9366                         flags |= CTL_LLF_DPO;
9367                 lba = scsi_8btou64(cdb->addr);
9368                 num_blocks = scsi_4btoul(cdb->length);
9369                 break;
9370         }
9371         default:
9372                 /*
9373                  * We got a command we don't support.  This shouldn't
9374                  * happen, commands should be filtered out above us.
9375                  */
9376                 ctl_set_invalid_opcode(ctsio);
9377                 ctl_done((union ctl_io *)ctsio);
9378                 return (CTL_RETVAL_COMPLETE);
9379         }
9380
9381         /*
9382          * The first check is to make sure we're in bounds, the second
9383          * check is to catch wrap-around problems.  If the lba + num blocks
9384          * is less than the lba, then we've wrapped around and the block
9385          * range is invalid anyway.
9386          */
9387         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9388          || ((lba + num_blocks) < lba)) {
9389                 ctl_set_lba_out_of_range(ctsio);
9390                 ctl_done((union ctl_io *)ctsio);
9391                 return (CTL_RETVAL_COMPLETE);
9392         }
9393
9394         /*
9395          * According to SBC-3, a transfer length of 0 is not an error.
9396          */
9397         if (num_blocks == 0) {
9398                 ctl_set_success(ctsio);
9399                 ctl_done((union ctl_io *)ctsio);
9400                 return (CTL_RETVAL_COMPLETE);
9401         }
9402
9403         lbalen = (struct ctl_lba_len_flags *)
9404             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9405         lbalen->lba = lba;
9406         lbalen->len = num_blocks;
9407         if (bytchk) {
9408                 lbalen->flags = CTL_LLF_COMPARE | flags;
9409                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9410         } else {
9411                 lbalen->flags = CTL_LLF_VERIFY | flags;
9412                 ctsio->kern_total_len = 0;
9413         }
9414         ctsio->kern_rel_offset = 0;
9415
9416         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9417         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9418         return (retval);
9419 }
9420
9421 int
9422 ctl_report_luns(struct ctl_scsiio *ctsio)
9423 {
9424         struct scsi_report_luns *cdb;
9425         struct scsi_report_luns_data *lun_data;
9426         struct ctl_lun *lun, *request_lun;
9427         int num_luns, retval;
9428         uint32_t alloc_len, lun_datalen;
9429         int num_filled, well_known;
9430         uint32_t initidx, targ_lun_id, lun_id;
9431
9432         retval = CTL_RETVAL_COMPLETE;
9433         well_known = 0;
9434
9435         cdb = (struct scsi_report_luns *)ctsio->cdb;
9436
9437         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9438
9439         mtx_lock(&control_softc->ctl_lock);
9440         num_luns = control_softc->num_luns;
9441         mtx_unlock(&control_softc->ctl_lock);
9442
9443         switch (cdb->select_report) {
9444         case RPL_REPORT_DEFAULT:
9445         case RPL_REPORT_ALL:
9446                 break;
9447         case RPL_REPORT_WELLKNOWN:
9448                 well_known = 1;
9449                 num_luns = 0;
9450                 break;
9451         default:
9452                 ctl_set_invalid_field(ctsio,
9453                                       /*sks_valid*/ 1,
9454                                       /*command*/ 1,
9455                                       /*field*/ 2,
9456                                       /*bit_valid*/ 0,
9457                                       /*bit*/ 0);
9458                 ctl_done((union ctl_io *)ctsio);
9459                 return (retval);
9460                 break; /* NOTREACHED */
9461         }
9462
9463         alloc_len = scsi_4btoul(cdb->length);
9464         /*
9465          * The initiator has to allocate at least 16 bytes for this request,
9466          * so he can at least get the header and the first LUN.  Otherwise
9467          * we reject the request (per SPC-3 rev 14, section 6.21).
9468          */
9469         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9470             sizeof(struct scsi_report_luns_lundata))) {
9471                 ctl_set_invalid_field(ctsio,
9472                                       /*sks_valid*/ 1,
9473                                       /*command*/ 1,
9474                                       /*field*/ 6,
9475                                       /*bit_valid*/ 0,
9476                                       /*bit*/ 0);
9477                 ctl_done((union ctl_io *)ctsio);
9478                 return (retval);
9479         }
9480
9481         request_lun = (struct ctl_lun *)
9482                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9483
9484         lun_datalen = sizeof(*lun_data) +
9485                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9486
9487         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9488         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9489         ctsio->kern_sg_entries = 0;
9490
9491         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9492
9493         mtx_lock(&control_softc->ctl_lock);
9494         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9495                 lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9496                 if (lun_id >= CTL_MAX_LUNS)
9497                         continue;
9498                 lun = control_softc->ctl_luns[lun_id];
9499                 if (lun == NULL)
9500                         continue;
9501
9502                 if (targ_lun_id <= 0xff) {
9503                         /*
9504                          * Peripheral addressing method, bus number 0.
9505                          */
9506                         lun_data->luns[num_filled].lundata[0] =
9507                                 RPL_LUNDATA_ATYP_PERIPH;
9508                         lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9509                         num_filled++;
9510                 } else if (targ_lun_id <= 0x3fff) {
9511                         /*
9512                          * Flat addressing method.
9513                          */
9514                         lun_data->luns[num_filled].lundata[0] =
9515                                 RPL_LUNDATA_ATYP_FLAT |
9516                                 (targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9517 #ifdef OLDCTLHEADERS
9518                                 (SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9519                                 (targ_lun_id & SRLD_BUS_LUN_MASK);
9520 #endif
9521                         lun_data->luns[num_filled].lundata[1] =
9522 #ifdef OLDCTLHEADERS
9523                                 targ_lun_id >> SRLD_BUS_LUN_BITS;
9524 #endif
9525                                 targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9526                         num_filled++;
9527                 } else {
9528                         printf("ctl_report_luns: bogus LUN number %jd, "
9529                                "skipping\n", (intmax_t)targ_lun_id);
9530                 }
9531                 /*
9532                  * According to SPC-3, rev 14 section 6.21:
9533                  *
9534                  * "The execution of a REPORT LUNS command to any valid and
9535                  * installed logical unit shall clear the REPORTED LUNS DATA
9536                  * HAS CHANGED unit attention condition for all logical
9537                  * units of that target with respect to the requesting
9538                  * initiator. A valid and installed logical unit is one
9539                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9540                  * INQUIRY data (see 6.4.2)."
9541                  *
9542                  * If request_lun is NULL, the LUN this report luns command
9543                  * was issued to is either disabled or doesn't exist. In that
9544                  * case, we shouldn't clear any pending lun change unit
9545                  * attention.
9546                  */
9547                 if (request_lun != NULL) {
9548                         mtx_lock(&lun->lun_lock);
9549                         lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9550                         mtx_unlock(&lun->lun_lock);
9551                 }
9552         }
9553         mtx_unlock(&control_softc->ctl_lock);
9554
9555         /*
9556          * It's quite possible that we've returned fewer LUNs than we allocated
9557          * space for.  Trim it.
9558          */
9559         lun_datalen = sizeof(*lun_data) +
9560                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9561
9562         if (lun_datalen < alloc_len) {
9563                 ctsio->residual = alloc_len - lun_datalen;
9564                 ctsio->kern_data_len = lun_datalen;
9565                 ctsio->kern_total_len = lun_datalen;
9566         } else {
9567                 ctsio->residual = 0;
9568                 ctsio->kern_data_len = alloc_len;
9569                 ctsio->kern_total_len = alloc_len;
9570         }
9571         ctsio->kern_data_resid = 0;
9572         ctsio->kern_rel_offset = 0;
9573         ctsio->kern_sg_entries = 0;
9574
9575         /*
9576          * We set this to the actual data length, regardless of how much
9577          * space we actually have to return results.  If the user looks at
9578          * this value, he'll know whether or not he allocated enough space
9579          * and reissue the command if necessary.  We don't support well
9580          * known logical units, so if the user asks for that, return none.
9581          */
9582         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9583
9584         /*
9585          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9586          * this request.
9587          */
9588         ctsio->scsi_status = SCSI_STATUS_OK;
9589
9590         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9591         ctsio->be_move_done = ctl_config_move_done;
9592         ctl_datamove((union ctl_io *)ctsio);
9593
9594         return (retval);
9595 }
9596
9597 int
9598 ctl_request_sense(struct ctl_scsiio *ctsio)
9599 {
9600         struct scsi_request_sense *cdb;
9601         struct scsi_sense_data *sense_ptr;
9602         struct ctl_lun *lun;
9603         uint32_t initidx;
9604         int have_error;
9605         scsi_sense_data_type sense_format;
9606
9607         cdb = (struct scsi_request_sense *)ctsio->cdb;
9608
9609         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9610
9611         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9612
9613         /*
9614          * Determine which sense format the user wants.
9615          */
9616         if (cdb->byte2 & SRS_DESC)
9617                 sense_format = SSD_TYPE_DESC;
9618         else
9619                 sense_format = SSD_TYPE_FIXED;
9620
9621         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9622         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9623         ctsio->kern_sg_entries = 0;
9624
9625         /*
9626          * struct scsi_sense_data, which is currently set to 256 bytes, is
9627          * larger than the largest allowed value for the length field in the
9628          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9629          */
9630         ctsio->residual = 0;
9631         ctsio->kern_data_len = cdb->length;
9632         ctsio->kern_total_len = cdb->length;
9633
9634         ctsio->kern_data_resid = 0;
9635         ctsio->kern_rel_offset = 0;
9636         ctsio->kern_sg_entries = 0;
9637
9638         /*
9639          * If we don't have a LUN, we don't have any pending sense.
9640          */
9641         if (lun == NULL)
9642                 goto no_sense;
9643
9644         have_error = 0;
9645         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9646         /*
9647          * Check for pending sense, and then for pending unit attentions.
9648          * Pending sense gets returned first, then pending unit attentions.
9649          */
9650         mtx_lock(&lun->lun_lock);
9651 #ifdef CTL_WITH_CA
9652         if (ctl_is_set(lun->have_ca, initidx)) {
9653                 scsi_sense_data_type stored_format;
9654
9655                 /*
9656                  * Check to see which sense format was used for the stored
9657                  * sense data.
9658                  */
9659                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9660
9661                 /*
9662                  * If the user requested a different sense format than the
9663                  * one we stored, then we need to convert it to the other
9664                  * format.  If we're going from descriptor to fixed format
9665                  * sense data, we may lose things in translation, depending
9666                  * on what options were used.
9667                  *
9668                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9669                  * for some reason we'll just copy it out as-is.
9670                  */
9671                 if ((stored_format == SSD_TYPE_FIXED)
9672                  && (sense_format == SSD_TYPE_DESC))
9673                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9674                             &lun->pending_sense[initidx],
9675                             (struct scsi_sense_data_desc *)sense_ptr);
9676                 else if ((stored_format == SSD_TYPE_DESC)
9677                       && (sense_format == SSD_TYPE_FIXED))
9678                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9679                             &lun->pending_sense[initidx],
9680                             (struct scsi_sense_data_fixed *)sense_ptr);
9681                 else
9682                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9683                                ctl_min(sizeof(*sense_ptr),
9684                                sizeof(lun->pending_sense[initidx])));
9685
9686                 ctl_clear_mask(lun->have_ca, initidx);
9687                 have_error = 1;
9688         } else
9689 #endif
9690         if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9691                 ctl_ua_type ua_type;
9692
9693                 ua_type = ctl_build_ua(&lun->pending_ua[initidx],
9694                                        sense_ptr, sense_format);
9695                 if (ua_type != CTL_UA_NONE)
9696                         have_error = 1;
9697         }
9698         mtx_unlock(&lun->lun_lock);
9699
9700         /*
9701          * We already have a pending error, return it.
9702          */
9703         if (have_error != 0) {
9704                 /*
9705                  * We report the SCSI status as OK, since the status of the
9706                  * request sense command itself is OK.
9707                  */
9708                 ctsio->scsi_status = SCSI_STATUS_OK;
9709
9710                 /*
9711                  * We report 0 for the sense length, because we aren't doing
9712                  * autosense in this case.  We're reporting sense as
9713                  * parameter data.
9714                  */
9715                 ctsio->sense_len = 0;
9716                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9717                 ctsio->be_move_done = ctl_config_move_done;
9718                 ctl_datamove((union ctl_io *)ctsio);
9719
9720                 return (CTL_RETVAL_COMPLETE);
9721         }
9722
9723 no_sense:
9724
9725         /*
9726          * No sense information to report, so we report that everything is
9727          * okay.
9728          */
9729         ctl_set_sense_data(sense_ptr,
9730                            lun,
9731                            sense_format,
9732                            /*current_error*/ 1,
9733                            /*sense_key*/ SSD_KEY_NO_SENSE,
9734                            /*asc*/ 0x00,
9735                            /*ascq*/ 0x00,
9736                            SSD_ELEM_NONE);
9737
9738         ctsio->scsi_status = SCSI_STATUS_OK;
9739
9740         /*
9741          * We report 0 for the sense length, because we aren't doing
9742          * autosense in this case.  We're reporting sense as parameter data.
9743          */
9744         ctsio->sense_len = 0;
9745         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9746         ctsio->be_move_done = ctl_config_move_done;
9747         ctl_datamove((union ctl_io *)ctsio);
9748
9749         return (CTL_RETVAL_COMPLETE);
9750 }
9751
9752 int
9753 ctl_tur(struct ctl_scsiio *ctsio)
9754 {
9755         struct ctl_lun *lun;
9756
9757         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9758
9759         CTL_DEBUG_PRINT(("ctl_tur\n"));
9760
9761         if (lun == NULL)
9762                 return (EINVAL);
9763
9764         ctsio->scsi_status = SCSI_STATUS_OK;
9765         ctsio->io_hdr.status = CTL_SUCCESS;
9766
9767         ctl_done((union ctl_io *)ctsio);
9768
9769         return (CTL_RETVAL_COMPLETE);
9770 }
9771
9772 #ifdef notyet
9773 static int
9774 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9775 {
9776
9777 }
9778 #endif
9779
9780 static int
9781 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9782 {
9783         struct scsi_vpd_supported_pages *pages;
9784         int sup_page_size;
9785         struct ctl_lun *lun;
9786
9787         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9788
9789         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9790             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9791         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9792         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9793         ctsio->kern_sg_entries = 0;
9794
9795         if (sup_page_size < alloc_len) {
9796                 ctsio->residual = alloc_len - sup_page_size;
9797                 ctsio->kern_data_len = sup_page_size;
9798                 ctsio->kern_total_len = sup_page_size;
9799         } else {
9800                 ctsio->residual = 0;
9801                 ctsio->kern_data_len = alloc_len;
9802                 ctsio->kern_total_len = alloc_len;
9803         }
9804         ctsio->kern_data_resid = 0;
9805         ctsio->kern_rel_offset = 0;
9806         ctsio->kern_sg_entries = 0;
9807
9808         /*
9809          * The control device is always connected.  The disk device, on the
9810          * other hand, may not be online all the time.  Need to change this
9811          * to figure out whether the disk device is actually online or not.
9812          */
9813         if (lun != NULL)
9814                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9815                                 lun->be_lun->lun_type;
9816         else
9817                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9818
9819         pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9820         /* Supported VPD pages */
9821         pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9822         /* Serial Number */
9823         pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9824         /* Device Identification */
9825         pages->page_list[2] = SVPD_DEVICE_ID;
9826         /* Extended INQUIRY Data */
9827         pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9828         /* Mode Page Policy */
9829         pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9830         /* SCSI Ports */
9831         pages->page_list[5] = SVPD_SCSI_PORTS;
9832         /* Third-party Copy */
9833         pages->page_list[6] = SVPD_SCSI_TPC;
9834         /* Block limits */
9835         pages->page_list[7] = SVPD_BLOCK_LIMITS;
9836         /* Block Device Characteristics */
9837         pages->page_list[8] = SVPD_BDC;
9838         /* Logical Block Provisioning */
9839         pages->page_list[9] = SVPD_LBP;
9840
9841         ctsio->scsi_status = SCSI_STATUS_OK;
9842
9843         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9844         ctsio->be_move_done = ctl_config_move_done;
9845         ctl_datamove((union ctl_io *)ctsio);
9846
9847         return (CTL_RETVAL_COMPLETE);
9848 }
9849
9850 static int
9851 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9852 {
9853         struct scsi_vpd_unit_serial_number *sn_ptr;
9854         struct ctl_lun *lun;
9855
9856         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9857
9858         ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9859         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9860         ctsio->kern_sg_entries = 0;
9861
9862         if (sizeof(*sn_ptr) < alloc_len) {
9863                 ctsio->residual = alloc_len - sizeof(*sn_ptr);
9864                 ctsio->kern_data_len = sizeof(*sn_ptr);
9865                 ctsio->kern_total_len = sizeof(*sn_ptr);
9866         } else {
9867                 ctsio->residual = 0;
9868                 ctsio->kern_data_len = alloc_len;
9869                 ctsio->kern_total_len = alloc_len;
9870         }
9871         ctsio->kern_data_resid = 0;
9872         ctsio->kern_rel_offset = 0;
9873         ctsio->kern_sg_entries = 0;
9874
9875         /*
9876          * The control device is always connected.  The disk device, on the
9877          * other hand, may not be online all the time.  Need to change this
9878          * to figure out whether the disk device is actually online or not.
9879          */
9880         if (lun != NULL)
9881                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9882                                   lun->be_lun->lun_type;
9883         else
9884                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9885
9886         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9887         sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9888         /*
9889          * If we don't have a LUN, we just leave the serial number as
9890          * all spaces.
9891          */
9892         memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9893         if (lun != NULL) {
9894                 strncpy((char *)sn_ptr->serial_num,
9895                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9896         }
9897         ctsio->scsi_status = SCSI_STATUS_OK;
9898
9899         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9900         ctsio->be_move_done = ctl_config_move_done;
9901         ctl_datamove((union ctl_io *)ctsio);
9902
9903         return (CTL_RETVAL_COMPLETE);
9904 }
9905
9906
9907 static int
9908 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9909 {
9910         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9911         struct ctl_lun *lun;
9912         int data_len;
9913
9914         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9915
9916         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9917             sizeof(struct scsi_vpd_mode_page_policy_descr);
9918
9919         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9920         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9921         ctsio->kern_sg_entries = 0;
9922
9923         if (data_len < alloc_len) {
9924                 ctsio->residual = alloc_len - data_len;
9925                 ctsio->kern_data_len = data_len;
9926                 ctsio->kern_total_len = data_len;
9927         } else {
9928                 ctsio->residual = 0;
9929                 ctsio->kern_data_len = alloc_len;
9930                 ctsio->kern_total_len = alloc_len;
9931         }
9932         ctsio->kern_data_resid = 0;
9933         ctsio->kern_rel_offset = 0;
9934         ctsio->kern_sg_entries = 0;
9935
9936         /*
9937          * The control device is always connected.  The disk device, on the
9938          * other hand, may not be online all the time.
9939          */
9940         if (lun != NULL)
9941                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9942                                      lun->be_lun->lun_type;
9943         else
9944                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9945         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9946         eid_ptr->page_length = data_len - 4;
9947         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9948         eid_ptr->flags3 = SVPD_EID_V_SUP;
9949
9950         ctsio->scsi_status = SCSI_STATUS_OK;
9951         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9952         ctsio->be_move_done = ctl_config_move_done;
9953         ctl_datamove((union ctl_io *)ctsio);
9954
9955         return (CTL_RETVAL_COMPLETE);
9956 }
9957
9958 static int
9959 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9960 {
9961         struct scsi_vpd_mode_page_policy *mpp_ptr;
9962         struct ctl_lun *lun;
9963         int data_len;
9964
9965         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9966
9967         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9968             sizeof(struct scsi_vpd_mode_page_policy_descr);
9969
9970         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9971         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9972         ctsio->kern_sg_entries = 0;
9973
9974         if (data_len < alloc_len) {
9975                 ctsio->residual = alloc_len - data_len;
9976                 ctsio->kern_data_len = data_len;
9977                 ctsio->kern_total_len = data_len;
9978         } else {
9979                 ctsio->residual = 0;
9980                 ctsio->kern_data_len = alloc_len;
9981                 ctsio->kern_total_len = alloc_len;
9982         }
9983         ctsio->kern_data_resid = 0;
9984         ctsio->kern_rel_offset = 0;
9985         ctsio->kern_sg_entries = 0;
9986
9987         /*
9988          * The control device is always connected.  The disk device, on the
9989          * other hand, may not be online all the time.
9990          */
9991         if (lun != NULL)
9992                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9993                                      lun->be_lun->lun_type;
9994         else
9995                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9996         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9997         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9998         mpp_ptr->descr[0].page_code = 0x3f;
9999         mpp_ptr->descr[0].subpage_code = 0xff;
10000         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
10001
10002         ctsio->scsi_status = SCSI_STATUS_OK;
10003         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10004         ctsio->be_move_done = ctl_config_move_done;
10005         ctl_datamove((union ctl_io *)ctsio);
10006
10007         return (CTL_RETVAL_COMPLETE);
10008 }
10009
10010 static int
10011 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
10012 {
10013         struct scsi_vpd_device_id *devid_ptr;
10014         struct scsi_vpd_id_descriptor *desc;
10015         struct ctl_softc *ctl_softc;
10016         struct ctl_lun *lun;
10017         struct ctl_port *port;
10018         int data_len;
10019         uint8_t proto;
10020
10021         ctl_softc = control_softc;
10022
10023         port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
10024         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10025
10026         data_len = sizeof(struct scsi_vpd_device_id) +
10027             sizeof(struct scsi_vpd_id_descriptor) +
10028                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
10029             sizeof(struct scsi_vpd_id_descriptor) +
10030                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
10031         if (lun && lun->lun_devid)
10032                 data_len += lun->lun_devid->len;
10033         if (port->port_devid)
10034                 data_len += port->port_devid->len;
10035         if (port->target_devid)
10036                 data_len += port->target_devid->len;
10037
10038         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10039         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
10040         ctsio->kern_sg_entries = 0;
10041
10042         if (data_len < alloc_len) {
10043                 ctsio->residual = alloc_len - data_len;
10044                 ctsio->kern_data_len = data_len;
10045                 ctsio->kern_total_len = data_len;
10046         } else {
10047                 ctsio->residual = 0;
10048                 ctsio->kern_data_len = alloc_len;
10049                 ctsio->kern_total_len = alloc_len;
10050         }
10051         ctsio->kern_data_resid = 0;
10052         ctsio->kern_rel_offset = 0;
10053         ctsio->kern_sg_entries = 0;
10054
10055         /*
10056          * The control device is always connected.  The disk device, on the
10057          * other hand, may not be online all the time.
10058          */
10059         if (lun != NULL)
10060                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10061                                      lun->be_lun->lun_type;
10062         else
10063                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10064         devid_ptr->page_code = SVPD_DEVICE_ID;
10065         scsi_ulto2b(data_len - 4, devid_ptr->length);
10066
10067         if (port->port_type == CTL_PORT_FC)
10068                 proto = SCSI_PROTO_FC << 4;
10069         else if (port->port_type == CTL_PORT_ISCSI)
10070                 proto = SCSI_PROTO_ISCSI << 4;
10071         else
10072                 proto = SCSI_PROTO_SPI << 4;
10073         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
10074
10075         /*
10076          * We're using a LUN association here.  i.e., this device ID is a
10077          * per-LUN identifier.
10078          */
10079         if (lun && lun->lun_devid) {
10080                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
10081                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10082                     lun->lun_devid->len);
10083         }
10084
10085         /*
10086          * This is for the WWPN which is a port association.
10087          */
10088         if (port->port_devid) {
10089                 memcpy(desc, port->port_devid->data, port->port_devid->len);
10090                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10091                     port->port_devid->len);
10092         }
10093
10094         /*
10095          * This is for the Relative Target Port(type 4h) identifier
10096          */
10097         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10098         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10099             SVPD_ID_TYPE_RELTARG;
10100         desc->length = 4;
10101         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
10102         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10103             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
10104
10105         /*
10106          * This is for the Target Port Group(type 5h) identifier
10107          */
10108         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10109         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10110             SVPD_ID_TYPE_TPORTGRP;
10111         desc->length = 4;
10112         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
10113             &desc->identifier[2]);
10114         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10115             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10116
10117         /*
10118          * This is for the Target identifier
10119          */
10120         if (port->target_devid) {
10121                 memcpy(desc, port->target_devid->data, port->target_devid->len);
10122         }
10123
10124         ctsio->scsi_status = SCSI_STATUS_OK;
10125         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10126         ctsio->be_move_done = ctl_config_move_done;
10127         ctl_datamove((union ctl_io *)ctsio);
10128
10129         return (CTL_RETVAL_COMPLETE);
10130 }
10131
10132 static int
10133 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10134 {
10135         struct ctl_softc *softc = control_softc;
10136         struct scsi_vpd_scsi_ports *sp;
10137         struct scsi_vpd_port_designation *pd;
10138         struct scsi_vpd_port_designation_cont *pdc;
10139         struct ctl_lun *lun;
10140         struct ctl_port *port;
10141         int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10142         int num_target_port_groups, single;
10143
10144         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10145
10146         single = ctl_is_single;
10147         if (single)
10148                 num_target_port_groups = 1;
10149         else
10150                 num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10151         num_target_ports = 0;
10152         iid_len = 0;
10153         id_len = 0;
10154         mtx_lock(&softc->ctl_lock);
10155         STAILQ_FOREACH(port, &softc->port_list, links) {
10156                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10157                         continue;
10158                 if (lun != NULL &&
10159                     ctl_map_lun_back(port->targ_port, lun->lun) >=
10160                     CTL_MAX_LUNS)
10161                         continue;
10162                 num_target_ports++;
10163                 if (port->init_devid)
10164                         iid_len += port->init_devid->len;
10165                 if (port->port_devid)
10166                         id_len += port->port_devid->len;
10167         }
10168         mtx_unlock(&softc->ctl_lock);
10169
10170         data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10171             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10172              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10173         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10174         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10175         ctsio->kern_sg_entries = 0;
10176
10177         if (data_len < alloc_len) {
10178                 ctsio->residual = alloc_len - data_len;
10179                 ctsio->kern_data_len = data_len;
10180                 ctsio->kern_total_len = data_len;
10181         } else {
10182                 ctsio->residual = 0;
10183                 ctsio->kern_data_len = alloc_len;
10184                 ctsio->kern_total_len = alloc_len;
10185         }
10186         ctsio->kern_data_resid = 0;
10187         ctsio->kern_rel_offset = 0;
10188         ctsio->kern_sg_entries = 0;
10189
10190         /*
10191          * The control device is always connected.  The disk device, on the
10192          * other hand, may not be online all the time.  Need to change this
10193          * to figure out whether the disk device is actually online or not.
10194          */
10195         if (lun != NULL)
10196                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10197                                   lun->be_lun->lun_type;
10198         else
10199                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10200
10201         sp->page_code = SVPD_SCSI_PORTS;
10202         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10203             sp->page_length);
10204         pd = &sp->design[0];
10205
10206         mtx_lock(&softc->ctl_lock);
10207         if (softc->flags & CTL_FLAG_MASTER_SHELF)
10208                 pg = 0;
10209         else
10210                 pg = 1;
10211         for (g = 0; g < num_target_port_groups; g++) {
10212                 STAILQ_FOREACH(port, &softc->port_list, links) {
10213                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10214                                 continue;
10215                         if (lun != NULL &&
10216                             ctl_map_lun_back(port->targ_port, lun->lun) >=
10217                             CTL_MAX_LUNS)
10218                                 continue;
10219                         p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10220                         scsi_ulto2b(p, pd->relative_port_id);
10221                         if (port->init_devid && g == pg) {
10222                                 iid_len = port->init_devid->len;
10223                                 memcpy(pd->initiator_transportid,
10224                                     port->init_devid->data, port->init_devid->len);
10225                         } else
10226                                 iid_len = 0;
10227                         scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10228                         pdc = (struct scsi_vpd_port_designation_cont *)
10229                             (&pd->initiator_transportid[iid_len]);
10230                         if (port->port_devid && g == pg) {
10231                                 id_len = port->port_devid->len;
10232                                 memcpy(pdc->target_port_descriptors,
10233                                     port->port_devid->data, port->port_devid->len);
10234                         } else
10235                                 id_len = 0;
10236                         scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10237                         pd = (struct scsi_vpd_port_designation *)
10238                             ((uint8_t *)pdc->target_port_descriptors + id_len);
10239                 }
10240         }
10241         mtx_unlock(&softc->ctl_lock);
10242
10243         ctsio->scsi_status = SCSI_STATUS_OK;
10244         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10245         ctsio->be_move_done = ctl_config_move_done;
10246         ctl_datamove((union ctl_io *)ctsio);
10247
10248         return (CTL_RETVAL_COMPLETE);
10249 }
10250
10251 static int
10252 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10253 {
10254         struct scsi_vpd_block_limits *bl_ptr;
10255         struct ctl_lun *lun;
10256         int bs;
10257
10258         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10259
10260         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10261         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10262         ctsio->kern_sg_entries = 0;
10263
10264         if (sizeof(*bl_ptr) < alloc_len) {
10265                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
10266                 ctsio->kern_data_len = sizeof(*bl_ptr);
10267                 ctsio->kern_total_len = sizeof(*bl_ptr);
10268         } else {
10269                 ctsio->residual = 0;
10270                 ctsio->kern_data_len = alloc_len;
10271                 ctsio->kern_total_len = alloc_len;
10272         }
10273         ctsio->kern_data_resid = 0;
10274         ctsio->kern_rel_offset = 0;
10275         ctsio->kern_sg_entries = 0;
10276
10277         /*
10278          * The control device is always connected.  The disk device, on the
10279          * other hand, may not be online all the time.  Need to change this
10280          * to figure out whether the disk device is actually online or not.
10281          */
10282         if (lun != NULL)
10283                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10284                                   lun->be_lun->lun_type;
10285         else
10286                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10287
10288         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10289         scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
10290         bl_ptr->max_cmp_write_len = 0xff;
10291         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10292         if (lun != NULL) {
10293                 bs = lun->be_lun->blocksize;
10294                 scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10295                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10296                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10297                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10298                         if (lun->be_lun->pblockexp != 0) {
10299                                 scsi_ulto4b((1 << lun->be_lun->pblockexp),
10300                                     bl_ptr->opt_unmap_grain);
10301                                 scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10302                                     bl_ptr->unmap_grain_align);
10303                         }
10304                 }
10305         }
10306         scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10307
10308         ctsio->scsi_status = SCSI_STATUS_OK;
10309         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10310         ctsio->be_move_done = ctl_config_move_done;
10311         ctl_datamove((union ctl_io *)ctsio);
10312
10313         return (CTL_RETVAL_COMPLETE);
10314 }
10315
10316 static int
10317 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10318 {
10319         struct scsi_vpd_block_device_characteristics *bdc_ptr;
10320         struct ctl_lun *lun;
10321
10322         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10323
10324         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10325         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10326         ctsio->kern_sg_entries = 0;
10327
10328         if (sizeof(*bdc_ptr) < alloc_len) {
10329                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10330                 ctsio->kern_data_len = sizeof(*bdc_ptr);
10331                 ctsio->kern_total_len = sizeof(*bdc_ptr);
10332         } else {
10333                 ctsio->residual = 0;
10334                 ctsio->kern_data_len = alloc_len;
10335                 ctsio->kern_total_len = alloc_len;
10336         }
10337         ctsio->kern_data_resid = 0;
10338         ctsio->kern_rel_offset = 0;
10339         ctsio->kern_sg_entries = 0;
10340
10341         /*
10342          * The control device is always connected.  The disk device, on the
10343          * other hand, may not be online all the time.  Need to change this
10344          * to figure out whether the disk device is actually online or not.
10345          */
10346         if (lun != NULL)
10347                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10348                                   lun->be_lun->lun_type;
10349         else
10350                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10351         bdc_ptr->page_code = SVPD_BDC;
10352         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10353         scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10354         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10355
10356         ctsio->scsi_status = SCSI_STATUS_OK;
10357         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10358         ctsio->be_move_done = ctl_config_move_done;
10359         ctl_datamove((union ctl_io *)ctsio);
10360
10361         return (CTL_RETVAL_COMPLETE);
10362 }
10363
10364 static int
10365 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10366 {
10367         struct scsi_vpd_logical_block_prov *lbp_ptr;
10368         struct ctl_lun *lun;
10369
10370         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10371
10372         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10373         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10374         ctsio->kern_sg_entries = 0;
10375
10376         if (sizeof(*lbp_ptr) < alloc_len) {
10377                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10378                 ctsio->kern_data_len = sizeof(*lbp_ptr);
10379                 ctsio->kern_total_len = sizeof(*lbp_ptr);
10380         } else {
10381                 ctsio->residual = 0;
10382                 ctsio->kern_data_len = alloc_len;
10383                 ctsio->kern_total_len = alloc_len;
10384         }
10385         ctsio->kern_data_resid = 0;
10386         ctsio->kern_rel_offset = 0;
10387         ctsio->kern_sg_entries = 0;
10388
10389         /*
10390          * The control device is always connected.  The disk device, on the
10391          * other hand, may not be online all the time.  Need to change this
10392          * to figure out whether the disk device is actually online or not.
10393          */
10394         if (lun != NULL)
10395                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10396                                   lun->be_lun->lun_type;
10397         else
10398                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10399
10400         lbp_ptr->page_code = SVPD_LBP;
10401         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10402         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10403                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10404                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10405                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10406         }
10407
10408         ctsio->scsi_status = SCSI_STATUS_OK;
10409         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10410         ctsio->be_move_done = ctl_config_move_done;
10411         ctl_datamove((union ctl_io *)ctsio);
10412
10413         return (CTL_RETVAL_COMPLETE);
10414 }
10415
10416 static int
10417 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10418 {
10419         struct scsi_inquiry *cdb;
10420         struct ctl_lun *lun;
10421         int alloc_len, retval;
10422
10423         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10424         cdb = (struct scsi_inquiry *)ctsio->cdb;
10425
10426         retval = CTL_RETVAL_COMPLETE;
10427
10428         alloc_len = scsi_2btoul(cdb->length);
10429
10430         switch (cdb->page_code) {
10431         case SVPD_SUPPORTED_PAGES:
10432                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10433                 break;
10434         case SVPD_UNIT_SERIAL_NUMBER:
10435                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10436                 break;
10437         case SVPD_DEVICE_ID:
10438                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10439                 break;
10440         case SVPD_EXTENDED_INQUIRY_DATA:
10441                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10442                 break;
10443         case SVPD_MODE_PAGE_POLICY:
10444                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10445                 break;
10446         case SVPD_SCSI_PORTS:
10447                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10448                 break;
10449         case SVPD_SCSI_TPC:
10450                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10451                 break;
10452         case SVPD_BLOCK_LIMITS:
10453                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10454                 break;
10455         case SVPD_BDC:
10456                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10457                 break;
10458         case SVPD_LBP:
10459                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10460                 break;
10461         default:
10462                 ctl_set_invalid_field(ctsio,
10463                                       /*sks_valid*/ 1,
10464                                       /*command*/ 1,
10465                                       /*field*/ 2,
10466                                       /*bit_valid*/ 0,
10467                                       /*bit*/ 0);
10468                 ctl_done((union ctl_io *)ctsio);
10469                 retval = CTL_RETVAL_COMPLETE;
10470                 break;
10471         }
10472
10473         return (retval);
10474 }
10475
10476 static int
10477 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10478 {
10479         struct scsi_inquiry_data *inq_ptr;
10480         struct scsi_inquiry *cdb;
10481         struct ctl_softc *ctl_softc;
10482         struct ctl_lun *lun;
10483         char *val;
10484         uint32_t alloc_len;
10485         ctl_port_type port_type;
10486
10487         ctl_softc = control_softc;
10488
10489         /*
10490          * Figure out whether we're talking to a Fibre Channel port or not.
10491          * We treat the ioctl front end, and any SCSI adapters, as packetized
10492          * SCSI front ends.
10493          */
10494         port_type = ctl_softc->ctl_ports[
10495             ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10496         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10497                 port_type = CTL_PORT_SCSI;
10498
10499         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10500         cdb = (struct scsi_inquiry *)ctsio->cdb;
10501         alloc_len = scsi_2btoul(cdb->length);
10502
10503         /*
10504          * We malloc the full inquiry data size here and fill it
10505          * in.  If the user only asks for less, we'll give him
10506          * that much.
10507          */
10508         ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
10509         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10510         ctsio->kern_sg_entries = 0;
10511         ctsio->kern_data_resid = 0;
10512         ctsio->kern_rel_offset = 0;
10513
10514         if (sizeof(*inq_ptr) < alloc_len) {
10515                 ctsio->residual = alloc_len - sizeof(*inq_ptr);
10516                 ctsio->kern_data_len = sizeof(*inq_ptr);
10517                 ctsio->kern_total_len = sizeof(*inq_ptr);
10518         } else {
10519                 ctsio->residual = 0;
10520                 ctsio->kern_data_len = alloc_len;
10521                 ctsio->kern_total_len = alloc_len;
10522         }
10523
10524         /*
10525          * If we have a LUN configured, report it as connected.  Otherwise,
10526          * report that it is offline or no device is supported, depending
10527          * on the value of inquiry_pq_no_lun.
10528          *
10529          * According to the spec (SPC-4 r34), the peripheral qualifier
10530          * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10531          *
10532          * "A peripheral device having the specified peripheral device type
10533          * is not connected to this logical unit. However, the device
10534          * server is capable of supporting the specified peripheral device
10535          * type on this logical unit."
10536          *
10537          * According to the same spec, the peripheral qualifier
10538          * SID_QUAL_BAD_LU (011b) is used in this scenario:
10539          *
10540          * "The device server is not capable of supporting a peripheral
10541          * device on this logical unit. For this peripheral qualifier the
10542          * peripheral device type shall be set to 1Fh. All other peripheral
10543          * device type values are reserved for this peripheral qualifier."
10544          *
10545          * Given the text, it would seem that we probably want to report that
10546          * the LUN is offline here.  There is no LUN connected, but we can
10547          * support a LUN at the given LUN number.
10548          *
10549          * In the real world, though, it sounds like things are a little
10550          * different:
10551          *
10552          * - Linux, when presented with a LUN with the offline peripheral
10553          *   qualifier, will create an sg driver instance for it.  So when
10554          *   you attach it to CTL, you wind up with a ton of sg driver
10555          *   instances.  (One for every LUN that Linux bothered to probe.)
10556          *   Linux does this despite the fact that it issues a REPORT LUNs
10557          *   to LUN 0 to get the inventory of supported LUNs.
10558          *
10559          * - There is other anecdotal evidence (from Emulex folks) about
10560          *   arrays that use the offline peripheral qualifier for LUNs that
10561          *   are on the "passive" path in an active/passive array.
10562          *
10563          * So the solution is provide a hopefully reasonable default
10564          * (return bad/no LUN) and allow the user to change the behavior
10565          * with a tunable/sysctl variable.
10566          */
10567         if (lun != NULL)
10568                 inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10569                                   lun->be_lun->lun_type;
10570         else if (ctl_softc->inquiry_pq_no_lun == 0)
10571                 inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10572         else
10573                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10574
10575         /* RMB in byte 2 is 0 */
10576         inq_ptr->version = SCSI_REV_SPC4;
10577
10578         /*
10579          * According to SAM-3, even if a device only supports a single
10580          * level of LUN addressing, it should still set the HISUP bit:
10581          *
10582          * 4.9.1 Logical unit numbers overview
10583          *
10584          * All logical unit number formats described in this standard are
10585          * hierarchical in structure even when only a single level in that
10586          * hierarchy is used. The HISUP bit shall be set to one in the
10587          * standard INQUIRY data (see SPC-2) when any logical unit number
10588          * format described in this standard is used.  Non-hierarchical
10589          * formats are outside the scope of this standard.
10590          *
10591          * Therefore we set the HiSup bit here.
10592          *
10593          * The reponse format is 2, per SPC-3.
10594          */
10595         inq_ptr->response_format = SID_HiSup | 2;
10596
10597         inq_ptr->additional_length =
10598             offsetof(struct scsi_inquiry_data, vendor_specific1) -
10599             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10600         CTL_DEBUG_PRINT(("additional_length = %d\n",
10601                          inq_ptr->additional_length));
10602
10603         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10604         /* 16 bit addressing */
10605         if (port_type == CTL_PORT_SCSI)
10606                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10607         /* XXX set the SID_MultiP bit here if we're actually going to
10608            respond on multiple ports */
10609         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10610
10611         /* 16 bit data bus, synchronous transfers */
10612         if (port_type == CTL_PORT_SCSI)
10613                 inq_ptr->flags = SID_WBus16 | SID_Sync;
10614         /*
10615          * XXX KDM do we want to support tagged queueing on the control
10616          * device at all?
10617          */
10618         if ((lun == NULL)
10619          || (lun->be_lun->lun_type != T_PROCESSOR))
10620                 inq_ptr->flags |= SID_CmdQue;
10621         /*
10622          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10623          * We have 8 bytes for the vendor name, and 16 bytes for the device
10624          * name and 4 bytes for the revision.
10625          */
10626         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10627             "vendor")) == NULL) {
10628                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10629         } else {
10630                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10631                 strncpy(inq_ptr->vendor, val,
10632                     min(sizeof(inq_ptr->vendor), strlen(val)));
10633         }
10634         if (lun == NULL) {
10635                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10636                     sizeof(inq_ptr->product));
10637         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10638                 switch (lun->be_lun->lun_type) {
10639                 case T_DIRECT:
10640                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10641                             sizeof(inq_ptr->product));
10642                         break;
10643                 case T_PROCESSOR:
10644                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10645                             sizeof(inq_ptr->product));
10646                         break;
10647                 default:
10648                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10649                             sizeof(inq_ptr->product));
10650                         break;
10651                 }
10652         } else {
10653                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10654                 strncpy(inq_ptr->product, val,
10655                     min(sizeof(inq_ptr->product), strlen(val)));
10656         }
10657
10658         /*
10659          * XXX make this a macro somewhere so it automatically gets
10660          * incremented when we make changes.
10661          */
10662         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10663             "revision")) == NULL) {
10664                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10665         } else {
10666                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10667                 strncpy(inq_ptr->revision, val,
10668                     min(sizeof(inq_ptr->revision), strlen(val)));
10669         }
10670
10671         /*
10672          * For parallel SCSI, we support double transition and single
10673          * transition clocking.  We also support QAS (Quick Arbitration
10674          * and Selection) and Information Unit transfers on both the
10675          * control and array devices.
10676          */
10677         if (port_type == CTL_PORT_SCSI)
10678                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10679                                     SID_SPI_IUS;
10680
10681         /* SAM-5 (no version claimed) */
10682         scsi_ulto2b(0x00A0, inq_ptr->version1);
10683         /* SPC-4 (no version claimed) */
10684         scsi_ulto2b(0x0460, inq_ptr->version2);
10685         if (port_type == CTL_PORT_FC) {
10686                 /* FCP-2 ANSI INCITS.350:2003 */
10687                 scsi_ulto2b(0x0917, inq_ptr->version3);
10688         } else if (port_type == CTL_PORT_SCSI) {
10689                 /* SPI-4 ANSI INCITS.362:200x */
10690                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10691         } else if (port_type == CTL_PORT_ISCSI) {
10692                 /* iSCSI (no version claimed) */
10693                 scsi_ulto2b(0x0960, inq_ptr->version3);
10694         } else if (port_type == CTL_PORT_SAS) {
10695                 /* SAS (no version claimed) */
10696                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10697         }
10698
10699         if (lun == NULL) {
10700                 /* SBC-3 (no version claimed) */
10701                 scsi_ulto2b(0x04C0, inq_ptr->version4);
10702         } else {
10703                 switch (lun->be_lun->lun_type) {
10704                 case T_DIRECT:
10705                         /* SBC-3 (no version claimed) */
10706                         scsi_ulto2b(0x04C0, inq_ptr->version4);
10707                         break;
10708                 case T_PROCESSOR:
10709                 default:
10710                         break;
10711                 }
10712         }
10713
10714         ctsio->scsi_status = SCSI_STATUS_OK;
10715         if (ctsio->kern_data_len > 0) {
10716                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10717                 ctsio->be_move_done = ctl_config_move_done;
10718                 ctl_datamove((union ctl_io *)ctsio);
10719         } else {
10720                 ctsio->io_hdr.status = CTL_SUCCESS;
10721                 ctl_done((union ctl_io *)ctsio);
10722         }
10723
10724         return (CTL_RETVAL_COMPLETE);
10725 }
10726
10727 int
10728 ctl_inquiry(struct ctl_scsiio *ctsio)
10729 {
10730         struct scsi_inquiry *cdb;
10731         int retval;
10732
10733         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10734
10735         cdb = (struct scsi_inquiry *)ctsio->cdb;
10736         if (cdb->byte2 & SI_EVPD)
10737                 retval = ctl_inquiry_evpd(ctsio);
10738         else if (cdb->page_code == 0)
10739                 retval = ctl_inquiry_std(ctsio);
10740         else {
10741                 ctl_set_invalid_field(ctsio,
10742                                       /*sks_valid*/ 1,
10743                                       /*command*/ 1,
10744                                       /*field*/ 2,
10745                                       /*bit_valid*/ 0,
10746                                       /*bit*/ 0);
10747                 ctl_done((union ctl_io *)ctsio);
10748                 return (CTL_RETVAL_COMPLETE);
10749         }
10750
10751         return (retval);
10752 }
10753
10754 /*
10755  * For known CDB types, parse the LBA and length.
10756  */
10757 static int
10758 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10759 {
10760         if (io->io_hdr.io_type != CTL_IO_SCSI)
10761                 return (1);
10762
10763         switch (io->scsiio.cdb[0]) {
10764         case COMPARE_AND_WRITE: {
10765                 struct scsi_compare_and_write *cdb;
10766
10767                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10768
10769                 *lba = scsi_8btou64(cdb->addr);
10770                 *len = cdb->length;
10771                 break;
10772         }
10773         case READ_6:
10774         case WRITE_6: {
10775                 struct scsi_rw_6 *cdb;
10776
10777                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10778
10779                 *lba = scsi_3btoul(cdb->addr);
10780                 /* only 5 bits are valid in the most significant address byte */
10781                 *lba &= 0x1fffff;
10782                 *len = cdb->length;
10783                 break;
10784         }
10785         case READ_10:
10786         case WRITE_10: {
10787                 struct scsi_rw_10 *cdb;
10788
10789                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10790
10791                 *lba = scsi_4btoul(cdb->addr);
10792                 *len = scsi_2btoul(cdb->length);
10793                 break;
10794         }
10795         case WRITE_VERIFY_10: {
10796                 struct scsi_write_verify_10 *cdb;
10797
10798                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10799
10800                 *lba = scsi_4btoul(cdb->addr);
10801                 *len = scsi_2btoul(cdb->length);
10802                 break;
10803         }
10804         case READ_12:
10805         case WRITE_12: {
10806                 struct scsi_rw_12 *cdb;
10807
10808                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10809
10810                 *lba = scsi_4btoul(cdb->addr);
10811                 *len = scsi_4btoul(cdb->length);
10812                 break;
10813         }
10814         case WRITE_VERIFY_12: {
10815                 struct scsi_write_verify_12 *cdb;
10816
10817                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10818
10819                 *lba = scsi_4btoul(cdb->addr);
10820                 *len = scsi_4btoul(cdb->length);
10821                 break;
10822         }
10823         case READ_16:
10824         case WRITE_16: {
10825                 struct scsi_rw_16 *cdb;
10826
10827                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10828
10829                 *lba = scsi_8btou64(cdb->addr);
10830                 *len = scsi_4btoul(cdb->length);
10831                 break;
10832         }
10833         case WRITE_VERIFY_16: {
10834                 struct scsi_write_verify_16 *cdb;
10835
10836                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10837
10838
10839                 *lba = scsi_8btou64(cdb->addr);
10840                 *len = scsi_4btoul(cdb->length);
10841                 break;
10842         }
10843         case WRITE_SAME_10: {
10844                 struct scsi_write_same_10 *cdb;
10845
10846                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10847
10848                 *lba = scsi_4btoul(cdb->addr);
10849                 *len = scsi_2btoul(cdb->length);
10850                 break;
10851         }
10852         case WRITE_SAME_16: {
10853                 struct scsi_write_same_16 *cdb;
10854
10855                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10856
10857                 *lba = scsi_8btou64(cdb->addr);
10858                 *len = scsi_4btoul(cdb->length);
10859                 break;
10860         }
10861         case VERIFY_10: {
10862                 struct scsi_verify_10 *cdb;
10863
10864                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10865
10866                 *lba = scsi_4btoul(cdb->addr);
10867                 *len = scsi_2btoul(cdb->length);
10868                 break;
10869         }
10870         case VERIFY_12: {
10871                 struct scsi_verify_12 *cdb;
10872
10873                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10874
10875                 *lba = scsi_4btoul(cdb->addr);
10876                 *len = scsi_4btoul(cdb->length);
10877                 break;
10878         }
10879         case VERIFY_16: {
10880                 struct scsi_verify_16 *cdb;
10881
10882                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10883
10884                 *lba = scsi_8btou64(cdb->addr);
10885                 *len = scsi_4btoul(cdb->length);
10886                 break;
10887         }
10888         case UNMAP: {
10889                 *lba = 0;
10890                 *len = UINT64_MAX;
10891                 break;
10892         }
10893         default:
10894                 return (1);
10895                 break; /* NOTREACHED */
10896         }
10897
10898         return (0);
10899 }
10900
10901 static ctl_action
10902 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10903 {
10904         uint64_t endlba1, endlba2;
10905
10906         endlba1 = lba1 + len1 - 1;
10907         endlba2 = lba2 + len2 - 1;
10908
10909         if ((endlba1 < lba2)
10910          || (endlba2 < lba1))
10911                 return (CTL_ACTION_PASS);
10912         else
10913                 return (CTL_ACTION_BLOCK);
10914 }
10915
10916 static int
10917 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10918 {
10919         struct ctl_ptr_len_flags *ptrlen;
10920         struct scsi_unmap_desc *buf, *end, *range;
10921         uint64_t lba;
10922         uint32_t len;
10923
10924         /* If not UNMAP -- go other way. */
10925         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10926             io->scsiio.cdb[0] != UNMAP)
10927                 return (CTL_ACTION_ERROR);
10928
10929         /* If UNMAP without data -- block and wait for data. */
10930         ptrlen = (struct ctl_ptr_len_flags *)
10931             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10932         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10933             ptrlen->ptr == NULL)
10934                 return (CTL_ACTION_BLOCK);
10935
10936         /* UNMAP with data -- check for collision. */
10937         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10938         end = buf + ptrlen->len / sizeof(*buf);
10939         for (range = buf; range < end; range++) {
10940                 lba = scsi_8btou64(range->lba);
10941                 len = scsi_4btoul(range->length);
10942                 if ((lba < lba2 + len2) && (lba + len > lba2))
10943                         return (CTL_ACTION_BLOCK);
10944         }
10945         return (CTL_ACTION_PASS);
10946 }
10947
10948 static ctl_action
10949 ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10950 {
10951         uint64_t lba1, lba2;
10952         uint64_t len1, len2;
10953         int retval;
10954
10955         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10956                 return (CTL_ACTION_ERROR);
10957
10958         retval = ctl_extent_check_unmap(io2, lba1, len1);
10959         if (retval != CTL_ACTION_ERROR)
10960                 return (retval);
10961
10962         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10963                 return (CTL_ACTION_ERROR);
10964
10965         return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10966 }
10967
10968 static ctl_action
10969 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10970     union ctl_io *ooa_io)
10971 {
10972         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10973         ctl_serialize_action *serialize_row;
10974
10975         /*
10976          * The initiator attempted multiple untagged commands at the same
10977          * time.  Can't do that.
10978          */
10979         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10980          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10981          && ((pending_io->io_hdr.nexus.targ_port ==
10982               ooa_io->io_hdr.nexus.targ_port)
10983           && (pending_io->io_hdr.nexus.initid.id ==
10984               ooa_io->io_hdr.nexus.initid.id))
10985          && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10986                 return (CTL_ACTION_OVERLAP);
10987
10988         /*
10989          * The initiator attempted to send multiple tagged commands with
10990          * the same ID.  (It's fine if different initiators have the same
10991          * tag ID.)
10992          *
10993          * Even if all of those conditions are true, we don't kill the I/O
10994          * if the command ahead of us has been aborted.  We won't end up
10995          * sending it to the FETD, and it's perfectly legal to resend a
10996          * command with the same tag number as long as the previous
10997          * instance of this tag number has been aborted somehow.
10998          */
10999         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11000          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11001          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11002          && ((pending_io->io_hdr.nexus.targ_port ==
11003               ooa_io->io_hdr.nexus.targ_port)
11004           && (pending_io->io_hdr.nexus.initid.id ==
11005               ooa_io->io_hdr.nexus.initid.id))
11006          && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
11007                 return (CTL_ACTION_OVERLAP_TAG);
11008
11009         /*
11010          * If we get a head of queue tag, SAM-3 says that we should
11011          * immediately execute it.
11012          *
11013          * What happens if this command would normally block for some other
11014          * reason?  e.g. a request sense with a head of queue tag
11015          * immediately after a write.  Normally that would block, but this
11016          * will result in its getting executed immediately...
11017          *
11018          * We currently return "pass" instead of "skip", so we'll end up
11019          * going through the rest of the queue to check for overlapped tags.
11020          *
11021          * XXX KDM check for other types of blockage first??
11022          */
11023         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11024                 return (CTL_ACTION_PASS);
11025
11026         /*
11027          * Ordered tags have to block until all items ahead of them
11028          * have completed.  If we get called with an ordered tag, we always
11029          * block, if something else is ahead of us in the queue.
11030          */
11031         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11032                 return (CTL_ACTION_BLOCK);
11033
11034         /*
11035          * Simple tags get blocked until all head of queue and ordered tags
11036          * ahead of them have completed.  I'm lumping untagged commands in
11037          * with simple tags here.  XXX KDM is that the right thing to do?
11038          */
11039         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11040           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11041          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11042           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11043                 return (CTL_ACTION_BLOCK);
11044
11045         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11046         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11047
11048         serialize_row = ctl_serialize_table[ooa_entry->seridx];
11049
11050         switch (serialize_row[pending_entry->seridx]) {
11051         case CTL_SER_BLOCK:
11052                 return (CTL_ACTION_BLOCK);
11053         case CTL_SER_EXTENT:
11054                 return (ctl_extent_check(pending_io, ooa_io));
11055         case CTL_SER_EXTENTOPT:
11056                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11057                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11058                         return (ctl_extent_check(pending_io, ooa_io));
11059                 /* FALLTHROUGH */
11060         case CTL_SER_PASS:
11061                 return (CTL_ACTION_PASS);
11062         case CTL_SER_BLOCKOPT:
11063                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11064                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11065                         return (CTL_ACTION_BLOCK);
11066                 return (CTL_ACTION_PASS);
11067         case CTL_SER_SKIP:
11068                 return (CTL_ACTION_SKIP);
11069         default:
11070                 panic("invalid serialization value %d",
11071                       serialize_row[pending_entry->seridx]);
11072         }
11073
11074         return (CTL_ACTION_ERROR);
11075 }
11076
11077 /*
11078  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11079  * Assumptions:
11080  * - pending_io is generally either incoming, or on the blocked queue
11081  * - starting I/O is the I/O we want to start the check with.
11082  */
11083 static ctl_action
11084 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11085               union ctl_io *starting_io)
11086 {
11087         union ctl_io *ooa_io;
11088         ctl_action action;
11089
11090         mtx_assert(&lun->lun_lock, MA_OWNED);
11091
11092         /*
11093          * Run back along the OOA queue, starting with the current
11094          * blocked I/O and going through every I/O before it on the
11095          * queue.  If starting_io is NULL, we'll just end up returning
11096          * CTL_ACTION_PASS.
11097          */
11098         for (ooa_io = starting_io; ooa_io != NULL;
11099              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11100              ooa_links)){
11101
11102                 /*
11103                  * This routine just checks to see whether
11104                  * cur_blocked is blocked by ooa_io, which is ahead
11105                  * of it in the queue.  It doesn't queue/dequeue
11106                  * cur_blocked.
11107                  */
11108                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11109                 switch (action) {
11110                 case CTL_ACTION_BLOCK:
11111                 case CTL_ACTION_OVERLAP:
11112                 case CTL_ACTION_OVERLAP_TAG:
11113                 case CTL_ACTION_SKIP:
11114                 case CTL_ACTION_ERROR:
11115                         return (action);
11116                         break; /* NOTREACHED */
11117                 case CTL_ACTION_PASS:
11118                         break;
11119                 default:
11120                         panic("invalid action %d", action);
11121                         break;  /* NOTREACHED */
11122                 }
11123         }
11124
11125         return (CTL_ACTION_PASS);
11126 }
11127
11128 /*
11129  * Assumptions:
11130  * - An I/O has just completed, and has been removed from the per-LUN OOA
11131  *   queue, so some items on the blocked queue may now be unblocked.
11132  */
11133 static int
11134 ctl_check_blocked(struct ctl_lun *lun)
11135 {
11136         union ctl_io *cur_blocked, *next_blocked;
11137
11138         mtx_assert(&lun->lun_lock, MA_OWNED);
11139
11140         /*
11141          * Run forward from the head of the blocked queue, checking each
11142          * entry against the I/Os prior to it on the OOA queue to see if
11143          * there is still any blockage.
11144          *
11145          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11146          * with our removing a variable on it while it is traversing the
11147          * list.
11148          */
11149         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11150              cur_blocked != NULL; cur_blocked = next_blocked) {
11151                 union ctl_io *prev_ooa;
11152                 ctl_action action;
11153
11154                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11155                                                           blocked_links);
11156
11157                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11158                                                       ctl_ooaq, ooa_links);
11159
11160                 /*
11161                  * If cur_blocked happens to be the first item in the OOA
11162                  * queue now, prev_ooa will be NULL, and the action
11163                  * returned will just be CTL_ACTION_PASS.
11164                  */
11165                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11166
11167                 switch (action) {
11168                 case CTL_ACTION_BLOCK:
11169                         /* Nothing to do here, still blocked */
11170                         break;
11171                 case CTL_ACTION_OVERLAP:
11172                 case CTL_ACTION_OVERLAP_TAG:
11173                         /*
11174                          * This shouldn't happen!  In theory we've already
11175                          * checked this command for overlap...
11176                          */
11177                         break;
11178                 case CTL_ACTION_PASS:
11179                 case CTL_ACTION_SKIP: {
11180                         struct ctl_softc *softc;
11181                         const struct ctl_cmd_entry *entry;
11182                         uint32_t initidx;
11183                         int isc_retval;
11184
11185                         /*
11186                          * The skip case shouldn't happen, this transaction
11187                          * should have never made it onto the blocked queue.
11188                          */
11189                         /*
11190                          * This I/O is no longer blocked, we can remove it
11191                          * from the blocked queue.  Since this is a TAILQ
11192                          * (doubly linked list), we can do O(1) removals
11193                          * from any place on the list.
11194                          */
11195                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11196                                      blocked_links);
11197                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11198
11199                         if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11200                                 /*
11201                                  * Need to send IO back to original side to
11202                                  * run
11203                                  */
11204                                 union ctl_ha_msg msg_info;
11205
11206                                 msg_info.hdr.original_sc =
11207                                         cur_blocked->io_hdr.original_sc;
11208                                 msg_info.hdr.serializing_sc = cur_blocked;
11209                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11210                                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11211                                      &msg_info, sizeof(msg_info), 0)) >
11212                                      CTL_HA_STATUS_SUCCESS) {
11213                                         printf("CTL:Check Blocked error from "
11214                                                "ctl_ha_msg_send %d\n",
11215                                                isc_retval);
11216                                 }
11217                                 break;
11218                         }
11219                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11220                         softc = control_softc;
11221
11222                         initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11223
11224                         /*
11225                          * Check this I/O for LUN state changes that may
11226                          * have happened while this command was blocked.
11227                          * The LUN state may have been changed by a command
11228                          * ahead of us in the queue, so we need to re-check
11229                          * for any states that can be caused by SCSI
11230                          * commands.
11231                          */
11232                         if (ctl_scsiio_lun_check(softc, lun, entry,
11233                                                  &cur_blocked->scsiio) == 0) {
11234                                 cur_blocked->io_hdr.flags |=
11235                                                       CTL_FLAG_IS_WAS_ON_RTR;
11236                                 ctl_enqueue_rtr(cur_blocked);
11237                         } else
11238                                 ctl_done(cur_blocked);
11239                         break;
11240                 }
11241                 default:
11242                         /*
11243                          * This probably shouldn't happen -- we shouldn't
11244                          * get CTL_ACTION_ERROR, or anything else.
11245                          */
11246                         break;
11247                 }
11248         }
11249
11250         return (CTL_RETVAL_COMPLETE);
11251 }
11252
11253 /*
11254  * This routine (with one exception) checks LUN flags that can be set by
11255  * commands ahead of us in the OOA queue.  These flags have to be checked
11256  * when a command initially comes in, and when we pull a command off the
11257  * blocked queue and are preparing to execute it.  The reason we have to
11258  * check these flags for commands on the blocked queue is that the LUN
11259  * state may have been changed by a command ahead of us while we're on the
11260  * blocked queue.
11261  *
11262  * Ordering is somewhat important with these checks, so please pay
11263  * careful attention to the placement of any new checks.
11264  */
11265 static int
11266 ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11267     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11268 {
11269         int retval;
11270         uint32_t residx;
11271
11272         retval = 0;
11273
11274         mtx_assert(&lun->lun_lock, MA_OWNED);
11275
11276         /*
11277          * If this shelf is a secondary shelf controller, we have to reject
11278          * any media access commands.
11279          */
11280 #if 0
11281         /* No longer needed for HA */
11282         if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11283          && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11284                 ctl_set_lun_standby(ctsio);
11285                 retval = 1;
11286                 goto bailout;
11287         }
11288 #endif
11289
11290         /*
11291          * Check for a reservation conflict.  If this command isn't allowed
11292          * even on reserved LUNs, and if this initiator isn't the one who
11293          * reserved us, reject the command with a reservation conflict.
11294          */
11295         residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11296         if ((lun->flags & CTL_LUN_RESERVED)
11297          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11298                 if (lun->res_idx != residx) {
11299                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11300                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
11301                         retval = 1;
11302                         goto bailout;
11303                 }
11304         }
11305
11306         if ((lun->flags & CTL_LUN_PR_RESERVED)
11307          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11308                 /*
11309                  * if we aren't registered or it's a res holder type
11310                  * reservation and this isn't the res holder then set a
11311                  * conflict.
11312                  * NOTE: Commands which might be allowed on write exclusive
11313                  * type reservations are checked in the particular command
11314                  * for a conflict. Read and SSU are the only ones.
11315                  */
11316                 if (!lun->per_res[residx].registered
11317                  || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11318                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11319                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
11320                         retval = 1;
11321                         goto bailout;
11322                 }
11323
11324         }
11325
11326         if ((lun->flags & CTL_LUN_OFFLINE)
11327          && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11328                 ctl_set_lun_not_ready(ctsio);
11329                 retval = 1;
11330                 goto bailout;
11331         }
11332
11333         /*
11334          * If the LUN is stopped, see if this particular command is allowed
11335          * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11336          */
11337         if ((lun->flags & CTL_LUN_STOPPED)
11338          && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11339                 /* "Logical unit not ready, initializing cmd. required" */
11340                 ctl_set_lun_stopped(ctsio);
11341                 retval = 1;
11342                 goto bailout;
11343         }
11344
11345         if ((lun->flags & CTL_LUN_INOPERABLE)
11346          && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11347                 /* "Medium format corrupted" */
11348                 ctl_set_medium_format_corrupted(ctsio);
11349                 retval = 1;
11350                 goto bailout;
11351         }
11352
11353 bailout:
11354         return (retval);
11355
11356 }
11357
11358 static void
11359 ctl_failover_io(union ctl_io *io, int have_lock)
11360 {
11361         ctl_set_busy(&io->scsiio);
11362         ctl_done(io);
11363 }
11364
11365 static void
11366 ctl_failover(void)
11367 {
11368         struct ctl_lun *lun;
11369         struct ctl_softc *ctl_softc;
11370         union ctl_io *next_io, *pending_io;
11371         union ctl_io *io;
11372         int lun_idx;
11373         int i;
11374
11375         ctl_softc = control_softc;
11376
11377         mtx_lock(&ctl_softc->ctl_lock);
11378         /*
11379          * Remove any cmds from the other SC from the rtr queue.  These
11380          * will obviously only be for LUNs for which we're the primary.
11381          * We can't send status or get/send data for these commands.
11382          * Since they haven't been executed yet, we can just remove them.
11383          * We'll either abort them or delete them below, depending on
11384          * which HA mode we're in.
11385          */
11386 #ifdef notyet
11387         mtx_lock(&ctl_softc->queue_lock);
11388         for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11389              io != NULL; io = next_io) {
11390                 next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11391                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11392                         STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11393                                       ctl_io_hdr, links);
11394         }
11395         mtx_unlock(&ctl_softc->queue_lock);
11396 #endif
11397
11398         for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11399                 lun = ctl_softc->ctl_luns[lun_idx];
11400                 if (lun==NULL)
11401                         continue;
11402
11403                 /*
11404                  * Processor LUNs are primary on both sides.
11405                  * XXX will this always be true?
11406                  */
11407                 if (lun->be_lun->lun_type == T_PROCESSOR)
11408                         continue;
11409
11410                 if ((lun->flags & CTL_LUN_PRIMARY_SC)
11411                  && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11412                         printf("FAILOVER: primary lun %d\n", lun_idx);
11413                         /*
11414                          * Remove all commands from the other SC. First from the
11415                          * blocked queue then from the ooa queue. Once we have
11416                          * removed them. Call ctl_check_blocked to see if there
11417                          * is anything that can run.
11418                          */
11419                         for (io = (union ctl_io *)TAILQ_FIRST(
11420                              &lun->blocked_queue); io != NULL; io = next_io) {
11421
11422                                 next_io = (union ctl_io *)TAILQ_NEXT(
11423                                     &io->io_hdr, blocked_links);
11424
11425                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11426                                         TAILQ_REMOVE(&lun->blocked_queue,
11427                                                      &io->io_hdr,blocked_links);
11428                                         io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11429                                         TAILQ_REMOVE(&lun->ooa_queue,
11430                                                      &io->io_hdr, ooa_links);
11431
11432                                         ctl_free_io(io);
11433                                 }
11434                         }
11435
11436                         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11437                              io != NULL; io = next_io) {
11438
11439                                 next_io = (union ctl_io *)TAILQ_NEXT(
11440                                     &io->io_hdr, ooa_links);
11441
11442                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11443
11444                                         TAILQ_REMOVE(&lun->ooa_queue,
11445                                                 &io->io_hdr,
11446                                                 ooa_links);
11447
11448                                         ctl_free_io(io);
11449                                 }
11450                         }
11451                         ctl_check_blocked(lun);
11452                 } else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11453                         && (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11454
11455                         printf("FAILOVER: primary lun %d\n", lun_idx);
11456                         /*
11457                          * Abort all commands from the other SC.  We can't
11458                          * send status back for them now.  These should get
11459                          * cleaned up when they are completed or come out
11460                          * for a datamove operation.
11461                          */
11462                         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11463                              io != NULL; io = next_io) {
11464                                 next_io = (union ctl_io *)TAILQ_NEXT(
11465                                         &io->io_hdr, ooa_links);
11466
11467                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11468                                         io->io_hdr.flags |= CTL_FLAG_ABORT;
11469                         }
11470                 } else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11471                         && (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11472
11473                         printf("FAILOVER: secondary lun %d\n", lun_idx);
11474
11475                         lun->flags |= CTL_LUN_PRIMARY_SC;
11476
11477                         /*
11478                          * We send all I/O that was sent to this controller
11479                          * and redirected to the other side back with
11480                          * busy status, and have the initiator retry it.
11481                          * Figuring out how much data has been transferred,
11482                          * etc. and picking up where we left off would be
11483                          * very tricky.
11484                          *
11485                          * XXX KDM need to remove I/O from the blocked
11486                          * queue as well!
11487                          */
11488                         for (pending_io = (union ctl_io *)TAILQ_FIRST(
11489                              &lun->ooa_queue); pending_io != NULL;
11490                              pending_io = next_io) {
11491
11492                                 next_io =  (union ctl_io *)TAILQ_NEXT(
11493                                         &pending_io->io_hdr, ooa_links);
11494
11495                                 pending_io->io_hdr.flags &=
11496                                         ~CTL_FLAG_SENT_2OTHER_SC;
11497
11498                                 if (pending_io->io_hdr.flags &
11499                                     CTL_FLAG_IO_ACTIVE) {
11500                                         pending_io->io_hdr.flags |=
11501                                                 CTL_FLAG_FAILOVER;
11502                                 } else {
11503                                         ctl_set_busy(&pending_io->scsiio);
11504                                         ctl_done(pending_io);
11505                                 }
11506                         }
11507
11508                         /*
11509                          * Build Unit Attention
11510                          */
11511                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11512                                 lun->pending_ua[i] |=
11513                                                      CTL_UA_ASYM_ACC_CHANGE;
11514                         }
11515                 } else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11516                         && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11517                         printf("FAILOVER: secondary lun %d\n", lun_idx);
11518                         /*
11519                          * if the first io on the OOA is not on the RtR queue
11520                          * add it.
11521                          */
11522                         lun->flags |= CTL_LUN_PRIMARY_SC;
11523
11524                         pending_io = (union ctl_io *)TAILQ_FIRST(
11525                             &lun->ooa_queue);
11526                         if (pending_io==NULL) {
11527                                 printf("Nothing on OOA queue\n");
11528                                 continue;
11529                         }
11530
11531                         pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11532                         if ((pending_io->io_hdr.flags &
11533                              CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11534                                 pending_io->io_hdr.flags |=
11535                                     CTL_FLAG_IS_WAS_ON_RTR;
11536                                 ctl_enqueue_rtr(pending_io);
11537                         }
11538 #if 0
11539                         else
11540                         {
11541                                 printf("Tag 0x%04x is running\n",
11542                                       pending_io->scsiio.tag_num);
11543                         }
11544 #endif
11545
11546                         next_io = (union ctl_io *)TAILQ_NEXT(
11547                             &pending_io->io_hdr, ooa_links);
11548                         for (pending_io=next_io; pending_io != NULL;
11549                              pending_io = next_io) {
11550                                 pending_io->io_hdr.flags &=
11551                                     ~CTL_FLAG_SENT_2OTHER_SC;
11552                                 next_io = (union ctl_io *)TAILQ_NEXT(
11553                                         &pending_io->io_hdr, ooa_links);
11554                                 if (pending_io->io_hdr.flags &
11555                                     CTL_FLAG_IS_WAS_ON_RTR) {
11556 #if 0
11557                                         printf("Tag 0x%04x is running\n",
11558                                                 pending_io->scsiio.tag_num);
11559 #endif
11560                                         continue;
11561                                 }
11562
11563                                 switch (ctl_check_ooa(lun, pending_io,
11564                                     (union ctl_io *)TAILQ_PREV(
11565                                     &pending_io->io_hdr, ctl_ooaq,
11566                                     ooa_links))) {
11567
11568                                 case CTL_ACTION_BLOCK:
11569                                         TAILQ_INSERT_TAIL(&lun->blocked_queue,
11570                                                           &pending_io->io_hdr,
11571                                                           blocked_links);
11572                                         pending_io->io_hdr.flags |=
11573                                             CTL_FLAG_BLOCKED;
11574                                         break;
11575                                 case CTL_ACTION_PASS:
11576                                 case CTL_ACTION_SKIP:
11577                                         pending_io->io_hdr.flags |=
11578                                             CTL_FLAG_IS_WAS_ON_RTR;
11579                                         ctl_enqueue_rtr(pending_io);
11580                                         break;
11581                                 case CTL_ACTION_OVERLAP:
11582                                         ctl_set_overlapped_cmd(
11583                                             (struct ctl_scsiio *)pending_io);
11584                                         ctl_done(pending_io);
11585                                         break;
11586                                 case CTL_ACTION_OVERLAP_TAG:
11587                                         ctl_set_overlapped_tag(
11588                                             (struct ctl_scsiio *)pending_io,
11589                                             pending_io->scsiio.tag_num & 0xff);
11590                                         ctl_done(pending_io);
11591                                         break;
11592                                 case CTL_ACTION_ERROR:
11593                                 default:
11594                                         ctl_set_internal_failure(
11595                                                 (struct ctl_scsiio *)pending_io,
11596                                                 0,  // sks_valid
11597                                                 0); //retry count
11598                                         ctl_done(pending_io);
11599                                         break;
11600                                 }
11601                         }
11602
11603                         /*
11604                          * Build Unit Attention
11605                          */
11606                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11607                                 lun->pending_ua[i] |=
11608                                                      CTL_UA_ASYM_ACC_CHANGE;
11609                         }
11610                 } else {
11611                         panic("Unhandled HA mode failover, LUN flags = %#x, "
11612                               "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11613                 }
11614         }
11615         ctl_pause_rtr = 0;
11616         mtx_unlock(&ctl_softc->ctl_lock);
11617 }
11618
11619 static int
11620 ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11621 {
11622         struct ctl_lun *lun;
11623         const struct ctl_cmd_entry *entry;
11624         uint32_t initidx, targ_lun;
11625         int retval;
11626
11627         retval = 0;
11628
11629         lun = NULL;
11630
11631         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11632         if ((targ_lun < CTL_MAX_LUNS)
11633          && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11634                 lun = ctl_softc->ctl_luns[targ_lun];
11635                 /*
11636                  * If the LUN is invalid, pretend that it doesn't exist.
11637                  * It will go away as soon as all pending I/O has been
11638                  * completed.
11639                  */
11640                 if (lun->flags & CTL_LUN_DISABLED) {
11641                         lun = NULL;
11642                 } else {
11643                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11644                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11645                                 lun->be_lun;
11646                         if (lun->be_lun->lun_type == T_PROCESSOR) {
11647                                 ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11648                         }
11649
11650                         /*
11651                          * Every I/O goes into the OOA queue for a
11652                          * particular LUN, and stays there until completion.
11653                          */
11654                         mtx_lock(&lun->lun_lock);
11655                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11656                             ooa_links);
11657                 }
11658         } else {
11659                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11660                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11661         }
11662
11663         /* Get command entry and return error if it is unsuppotyed. */
11664         entry = ctl_validate_command(ctsio);
11665         if (entry == NULL) {
11666                 if (lun)
11667                         mtx_unlock(&lun->lun_lock);
11668                 return (retval);
11669         }
11670
11671         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11672         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11673
11674         /*
11675          * Check to see whether we can send this command to LUNs that don't
11676          * exist.  This should pretty much only be the case for inquiry
11677          * and request sense.  Further checks, below, really require having
11678          * a LUN, so we can't really check the command anymore.  Just put
11679          * it on the rtr queue.
11680          */
11681         if (lun == NULL) {
11682                 if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11683                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11684                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11685                         return (retval);
11686                 }
11687
11688                 ctl_set_unsupported_lun(ctsio);
11689                 ctl_done((union ctl_io *)ctsio);
11690                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11691                 return (retval);
11692         } else {
11693                 /*
11694                  * Make sure we support this particular command on this LUN.
11695                  * e.g., we don't support writes to the control LUN.
11696                  */
11697                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11698                         mtx_unlock(&lun->lun_lock);
11699                         ctl_set_invalid_opcode(ctsio);
11700                         ctl_done((union ctl_io *)ctsio);
11701                         return (retval);
11702                 }
11703         }
11704
11705         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11706
11707 #ifdef CTL_WITH_CA
11708         /*
11709          * If we've got a request sense, it'll clear the contingent
11710          * allegiance condition.  Otherwise, if we have a CA condition for
11711          * this initiator, clear it, because it sent down a command other
11712          * than request sense.
11713          */
11714         if ((ctsio->cdb[0] != REQUEST_SENSE)
11715          && (ctl_is_set(lun->have_ca, initidx)))
11716                 ctl_clear_mask(lun->have_ca, initidx);
11717 #endif
11718
11719         /*
11720          * If the command has this flag set, it handles its own unit
11721          * attention reporting, we shouldn't do anything.  Otherwise we
11722          * check for any pending unit attentions, and send them back to the
11723          * initiator.  We only do this when a command initially comes in,
11724          * not when we pull it off the blocked queue.
11725          *
11726          * According to SAM-3, section 5.3.2, the order that things get
11727          * presented back to the host is basically unit attentions caused
11728          * by some sort of reset event, busy status, reservation conflicts
11729          * or task set full, and finally any other status.
11730          *
11731          * One issue here is that some of the unit attentions we report
11732          * don't fall into the "reset" category (e.g. "reported luns data
11733          * has changed").  So reporting it here, before the reservation
11734          * check, may be technically wrong.  I guess the only thing to do
11735          * would be to check for and report the reset events here, and then
11736          * check for the other unit attention types after we check for a
11737          * reservation conflict.
11738          *
11739          * XXX KDM need to fix this
11740          */
11741         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11742                 ctl_ua_type ua_type;
11743
11744                 if (lun->pending_ua[initidx] != CTL_UA_NONE) {
11745                         scsi_sense_data_type sense_format;
11746
11747                         if (lun != NULL)
11748                                 sense_format = (lun->flags &
11749                                     CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11750                                     SSD_TYPE_FIXED;
11751                         else
11752                                 sense_format = SSD_TYPE_FIXED;
11753
11754                         ua_type = ctl_build_ua(&lun->pending_ua[initidx],
11755                             &ctsio->sense_data, sense_format);
11756                         if (ua_type != CTL_UA_NONE) {
11757                                 ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11758                                 ctsio->io_hdr.status = CTL_SCSI_ERROR |
11759                                                        CTL_AUTOSENSE;
11760                                 ctsio->sense_len = SSD_FULL_SIZE;
11761                                 mtx_unlock(&lun->lun_lock);
11762                                 ctl_done((union ctl_io *)ctsio);
11763                                 return (retval);
11764                         }
11765                 }
11766         }
11767
11768
11769         if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11770                 mtx_unlock(&lun->lun_lock);
11771                 ctl_done((union ctl_io *)ctsio);
11772                 return (retval);
11773         }
11774
11775         /*
11776          * XXX CHD this is where we want to send IO to other side if
11777          * this LUN is secondary on this SC. We will need to make a copy
11778          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11779          * the copy we send as FROM_OTHER.
11780          * We also need to stuff the address of the original IO so we can
11781          * find it easily. Something similar will need be done on the other
11782          * side so when we are done we can find the copy.
11783          */
11784         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11785                 union ctl_ha_msg msg_info;
11786                 int isc_retval;
11787
11788                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11789
11790                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11791                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11792 #if 0
11793                 printf("1. ctsio %p\n", ctsio);
11794 #endif
11795                 msg_info.hdr.serializing_sc = NULL;
11796                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11797                 msg_info.scsi.tag_num = ctsio->tag_num;
11798                 msg_info.scsi.tag_type = ctsio->tag_type;
11799                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11800
11801                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11802
11803                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11804                     (void *)&msg_info, sizeof(msg_info), 0)) >
11805                     CTL_HA_STATUS_SUCCESS) {
11806                         printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11807                                isc_retval);
11808                         printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11809                 } else {
11810 #if 0
11811                         printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11812 #endif
11813                 }
11814
11815                 /*
11816                  * XXX KDM this I/O is off the incoming queue, but hasn't
11817                  * been inserted on any other queue.  We may need to come
11818                  * up with a holding queue while we wait for serialization
11819                  * so that we have an idea of what we're waiting for from
11820                  * the other side.
11821                  */
11822                 mtx_unlock(&lun->lun_lock);
11823                 return (retval);
11824         }
11825
11826         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11827                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11828                               ctl_ooaq, ooa_links))) {
11829         case CTL_ACTION_BLOCK:
11830                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11831                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11832                                   blocked_links);
11833                 mtx_unlock(&lun->lun_lock);
11834                 return (retval);
11835         case CTL_ACTION_PASS:
11836         case CTL_ACTION_SKIP:
11837                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11838                 mtx_unlock(&lun->lun_lock);
11839                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11840                 break;
11841         case CTL_ACTION_OVERLAP:
11842                 mtx_unlock(&lun->lun_lock);
11843                 ctl_set_overlapped_cmd(ctsio);
11844                 ctl_done((union ctl_io *)ctsio);
11845                 break;
11846         case CTL_ACTION_OVERLAP_TAG:
11847                 mtx_unlock(&lun->lun_lock);
11848                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11849                 ctl_done((union ctl_io *)ctsio);
11850                 break;
11851         case CTL_ACTION_ERROR:
11852         default:
11853                 mtx_unlock(&lun->lun_lock);
11854                 ctl_set_internal_failure(ctsio,
11855                                          /*sks_valid*/ 0,
11856                                          /*retry_count*/ 0);
11857                 ctl_done((union ctl_io *)ctsio);
11858                 break;
11859         }
11860         return (retval);
11861 }
11862
11863 const struct ctl_cmd_entry *
11864 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11865 {
11866         const struct ctl_cmd_entry *entry;
11867         int service_action;
11868
11869         entry = &ctl_cmd_table[ctsio->cdb[0]];
11870         if (sa)
11871                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11872         if (entry->flags & CTL_CMD_FLAG_SA5) {
11873                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11874                 entry = &((const struct ctl_cmd_entry *)
11875                     entry->execute)[service_action];
11876         }
11877         return (entry);
11878 }
11879
11880 const struct ctl_cmd_entry *
11881 ctl_validate_command(struct ctl_scsiio *ctsio)
11882 {
11883         const struct ctl_cmd_entry *entry;
11884         int i, sa;
11885         uint8_t diff;
11886
11887         entry = ctl_get_cmd_entry(ctsio, &sa);
11888         if (entry->execute == NULL) {
11889                 if (sa)
11890                         ctl_set_invalid_field(ctsio,
11891                                               /*sks_valid*/ 1,
11892                                               /*command*/ 1,
11893                                               /*field*/ 1,
11894                                               /*bit_valid*/ 1,
11895                                               /*bit*/ 4);
11896                 else
11897                         ctl_set_invalid_opcode(ctsio);
11898                 ctl_done((union ctl_io *)ctsio);
11899                 return (NULL);
11900         }
11901         KASSERT(entry->length > 0,
11902             ("Not defined length for command 0x%02x/0x%02x",
11903              ctsio->cdb[0], ctsio->cdb[1]));
11904         for (i = 1; i < entry->length; i++) {
11905                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11906                 if (diff == 0)
11907                         continue;
11908                 ctl_set_invalid_field(ctsio,
11909                                       /*sks_valid*/ 1,
11910                                       /*command*/ 1,
11911                                       /*field*/ i,
11912                                       /*bit_valid*/ 1,
11913                                       /*bit*/ fls(diff) - 1);
11914                 ctl_done((union ctl_io *)ctsio);
11915                 return (NULL);
11916         }
11917         return (entry);
11918 }
11919
11920 static int
11921 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11922 {
11923
11924         switch (lun_type) {
11925         case T_PROCESSOR:
11926                 if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11927                     ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11928                         return (0);
11929                 break;
11930         case T_DIRECT:
11931                 if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11932                     ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11933                         return (0);
11934                 break;
11935         default:
11936                 return (0);
11937         }
11938         return (1);
11939 }
11940
11941 static int
11942 ctl_scsiio(struct ctl_scsiio *ctsio)
11943 {
11944         int retval;
11945         const struct ctl_cmd_entry *entry;
11946
11947         retval = CTL_RETVAL_COMPLETE;
11948
11949         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11950
11951         entry = ctl_get_cmd_entry(ctsio, NULL);
11952
11953         /*
11954          * If this I/O has been aborted, just send it straight to
11955          * ctl_done() without executing it.
11956          */
11957         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11958                 ctl_done((union ctl_io *)ctsio);
11959                 goto bailout;
11960         }
11961
11962         /*
11963          * All the checks should have been handled by ctl_scsiio_precheck().
11964          * We should be clear now to just execute the I/O.
11965          */
11966         retval = entry->execute(ctsio);
11967
11968 bailout:
11969         return (retval);
11970 }
11971
11972 /*
11973  * Since we only implement one target right now, a bus reset simply resets
11974  * our single target.
11975  */
11976 static int
11977 ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11978 {
11979         return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11980 }
11981
11982 static int
11983 ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11984                  ctl_ua_type ua_type)
11985 {
11986         struct ctl_lun *lun;
11987         int retval;
11988
11989         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11990                 union ctl_ha_msg msg_info;
11991
11992                 io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11993                 msg_info.hdr.nexus = io->io_hdr.nexus;
11994                 if (ua_type==CTL_UA_TARG_RESET)
11995                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11996                 else
11997                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11998                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11999                 msg_info.hdr.original_sc = NULL;
12000                 msg_info.hdr.serializing_sc = NULL;
12001                 if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12002                     (void *)&msg_info, sizeof(msg_info), 0)) {
12003                 }
12004         }
12005         retval = 0;
12006
12007         mtx_lock(&ctl_softc->ctl_lock);
12008         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
12009                 retval += ctl_lun_reset(lun, io, ua_type);
12010         mtx_unlock(&ctl_softc->ctl_lock);
12011
12012         return (retval);
12013 }
12014
12015 /*
12016  * The LUN should always be set.  The I/O is optional, and is used to
12017  * distinguish between I/Os sent by this initiator, and by other
12018  * initiators.  We set unit attention for initiators other than this one.
12019  * SAM-3 is vague on this point.  It does say that a unit attention should
12020  * be established for other initiators when a LUN is reset (see section
12021  * 5.7.3), but it doesn't specifically say that the unit attention should
12022  * be established for this particular initiator when a LUN is reset.  Here
12023  * is the relevant text, from SAM-3 rev 8:
12024  *
12025  * 5.7.2 When a SCSI initiator port aborts its own tasks
12026  *
12027  * When a SCSI initiator port causes its own task(s) to be aborted, no
12028  * notification that the task(s) have been aborted shall be returned to
12029  * the SCSI initiator port other than the completion response for the
12030  * command or task management function action that caused the task(s) to
12031  * be aborted and notification(s) associated with related effects of the
12032  * action (e.g., a reset unit attention condition).
12033  *
12034  * XXX KDM for now, we're setting unit attention for all initiators.
12035  */
12036 static int
12037 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
12038 {
12039         union ctl_io *xio;
12040 #if 0
12041         uint32_t initindex;
12042 #endif
12043         int i;
12044
12045         mtx_lock(&lun->lun_lock);
12046         /*
12047          * Run through the OOA queue and abort each I/O.
12048          */
12049 #if 0
12050         TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12051 #endif
12052         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12053              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12054                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
12055         }
12056
12057         /*
12058          * This version sets unit attention for every
12059          */
12060 #if 0
12061         initindex = ctl_get_initindex(&io->io_hdr.nexus);
12062         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12063                 if (initindex == i)
12064                         continue;
12065                 lun->pending_ua[i] |= ua_type;
12066         }
12067 #endif
12068
12069         /*
12070          * A reset (any kind, really) clears reservations established with
12071          * RESERVE/RELEASE.  It does not clear reservations established
12072          * with PERSISTENT RESERVE OUT, but we don't support that at the
12073          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
12074          * reservations made with the RESERVE/RELEASE commands, because
12075          * those commands are obsolete in SPC-3.
12076          */
12077         lun->flags &= ~CTL_LUN_RESERVED;
12078
12079         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12080 #ifdef CTL_WITH_CA
12081                 ctl_clear_mask(lun->have_ca, i);
12082 #endif
12083                 lun->pending_ua[i] |= ua_type;
12084         }
12085         mtx_unlock(&lun->lun_lock);
12086
12087         return (0);
12088 }
12089
12090 static void
12091 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
12092     int other_sc)
12093 {
12094         union ctl_io *xio;
12095
12096         mtx_assert(&lun->lun_lock, MA_OWNED);
12097
12098         /*
12099          * Run through the OOA queue and attempt to find the given I/O.
12100          * The target port, initiator ID, tag type and tag number have to
12101          * match the values that we got from the initiator.  If we have an
12102          * untagged command to abort, simply abort the first untagged command
12103          * we come to.  We only allow one untagged command at a time of course.
12104          */
12105         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12106              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12107
12108                 if ((targ_port == UINT32_MAX ||
12109                      targ_port == xio->io_hdr.nexus.targ_port) &&
12110                     (init_id == UINT32_MAX ||
12111                      init_id == xio->io_hdr.nexus.initid.id)) {
12112                         if (targ_port != xio->io_hdr.nexus.targ_port ||
12113                             init_id != xio->io_hdr.nexus.initid.id)
12114                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12115                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12116                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12117                                 union ctl_ha_msg msg_info;
12118
12119                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
12120                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12121                                 msg_info.task.tag_num = xio->scsiio.tag_num;
12122                                 msg_info.task.tag_type = xio->scsiio.tag_type;
12123                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12124                                 msg_info.hdr.original_sc = NULL;
12125                                 msg_info.hdr.serializing_sc = NULL;
12126                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12127                                     (void *)&msg_info, sizeof(msg_info), 0);
12128                         }
12129                 }
12130         }
12131 }
12132
12133 static int
12134 ctl_abort_task_set(union ctl_io *io)
12135 {
12136         struct ctl_softc *softc = control_softc;
12137         struct ctl_lun *lun;
12138         uint32_t targ_lun;
12139
12140         /*
12141          * Look up the LUN.
12142          */
12143         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12144         mtx_lock(&softc->ctl_lock);
12145         if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12146                 lun = softc->ctl_luns[targ_lun];
12147         else {
12148                 mtx_unlock(&softc->ctl_lock);
12149                 return (1);
12150         }
12151
12152         mtx_lock(&lun->lun_lock);
12153         mtx_unlock(&softc->ctl_lock);
12154         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12155                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12156                     io->io_hdr.nexus.initid.id,
12157                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12158         } else { /* CTL_TASK_CLEAR_TASK_SET */
12159                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12160                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12161         }
12162         mtx_unlock(&lun->lun_lock);
12163         return (0);
12164 }
12165
12166 static int
12167 ctl_i_t_nexus_reset(union ctl_io *io)
12168 {
12169         struct ctl_softc *softc = control_softc;
12170         struct ctl_lun *lun;
12171         uint32_t initindex, residx;
12172
12173         initindex = ctl_get_initindex(&io->io_hdr.nexus);
12174         residx = ctl_get_resindex(&io->io_hdr.nexus);
12175         mtx_lock(&softc->ctl_lock);
12176         STAILQ_FOREACH(lun, &softc->lun_list, links) {
12177                 mtx_lock(&lun->lun_lock);
12178                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12179                     io->io_hdr.nexus.initid.id,
12180                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12181 #ifdef CTL_WITH_CA
12182                 ctl_clear_mask(lun->have_ca, initindex);
12183 #endif
12184                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
12185                         lun->flags &= ~CTL_LUN_RESERVED;
12186                 lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12187                 mtx_unlock(&lun->lun_lock);
12188         }
12189         mtx_unlock(&softc->ctl_lock);
12190         return (0);
12191 }
12192
12193 static int
12194 ctl_abort_task(union ctl_io *io)
12195 {
12196         union ctl_io *xio;
12197         struct ctl_lun *lun;
12198         struct ctl_softc *ctl_softc;
12199 #if 0
12200         struct sbuf sb;
12201         char printbuf[128];
12202 #endif
12203         int found;
12204         uint32_t targ_lun;
12205
12206         ctl_softc = control_softc;
12207         found = 0;
12208
12209         /*
12210          * Look up the LUN.
12211          */
12212         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12213         mtx_lock(&ctl_softc->ctl_lock);
12214         if ((targ_lun < CTL_MAX_LUNS)
12215          && (ctl_softc->ctl_luns[targ_lun] != NULL))
12216                 lun = ctl_softc->ctl_luns[targ_lun];
12217         else {
12218                 mtx_unlock(&ctl_softc->ctl_lock);
12219                 return (1);
12220         }
12221
12222 #if 0
12223         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12224                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12225 #endif
12226
12227         mtx_lock(&lun->lun_lock);
12228         mtx_unlock(&ctl_softc->ctl_lock);
12229         /*
12230          * Run through the OOA queue and attempt to find the given I/O.
12231          * The target port, initiator ID, tag type and tag number have to
12232          * match the values that we got from the initiator.  If we have an
12233          * untagged command to abort, simply abort the first untagged command
12234          * we come to.  We only allow one untagged command at a time of course.
12235          */
12236 #if 0
12237         TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12238 #endif
12239         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12240              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12241 #if 0
12242                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12243
12244                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12245                             lun->lun, xio->scsiio.tag_num,
12246                             xio->scsiio.tag_type,
12247                             (xio->io_hdr.blocked_links.tqe_prev
12248                             == NULL) ? "" : " BLOCKED",
12249                             (xio->io_hdr.flags &
12250                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12251                             (xio->io_hdr.flags &
12252                             CTL_FLAG_ABORT) ? " ABORT" : "",
12253                             (xio->io_hdr.flags &
12254                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12255                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12256                 sbuf_finish(&sb);
12257                 printf("%s\n", sbuf_data(&sb));
12258 #endif
12259
12260                 if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12261                  && (xio->io_hdr.nexus.initid.id ==
12262                      io->io_hdr.nexus.initid.id)) {
12263                         /*
12264                          * If the abort says that the task is untagged, the
12265                          * task in the queue must be untagged.  Otherwise,
12266                          * we just check to see whether the tag numbers
12267                          * match.  This is because the QLogic firmware
12268                          * doesn't pass back the tag type in an abort
12269                          * request.
12270                          */
12271 #if 0
12272                         if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12273                           && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12274                          || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12275 #endif
12276                         /*
12277                          * XXX KDM we've got problems with FC, because it
12278                          * doesn't send down a tag type with aborts.  So we
12279                          * can only really go by the tag number...
12280                          * This may cause problems with parallel SCSI.
12281                          * Need to figure that out!!
12282                          */
12283                         if (xio->scsiio.tag_num == io->taskio.tag_num) {
12284                                 xio->io_hdr.flags |= CTL_FLAG_ABORT;
12285                                 found = 1;
12286                                 if ((io->io_hdr.flags &
12287                                      CTL_FLAG_FROM_OTHER_SC) == 0 &&
12288                                     !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12289                                         union ctl_ha_msg msg_info;
12290
12291                                         io->io_hdr.flags |=
12292                                                         CTL_FLAG_SENT_2OTHER_SC;
12293                                         msg_info.hdr.nexus = io->io_hdr.nexus;
12294                                         msg_info.task.task_action =
12295                                                 CTL_TASK_ABORT_TASK;
12296                                         msg_info.task.tag_num =
12297                                                 io->taskio.tag_num;
12298                                         msg_info.task.tag_type =
12299                                                 io->taskio.tag_type;
12300                                         msg_info.hdr.msg_type =
12301                                                 CTL_MSG_MANAGE_TASKS;
12302                                         msg_info.hdr.original_sc = NULL;
12303                                         msg_info.hdr.serializing_sc = NULL;
12304 #if 0
12305                                         printf("Sent Abort to other side\n");
12306 #endif
12307                                         if (CTL_HA_STATUS_SUCCESS !=
12308                                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12309                                                 (void *)&msg_info,
12310                                                 sizeof(msg_info), 0)) {
12311                                         }
12312                                 }
12313 #if 0
12314                                 printf("ctl_abort_task: found I/O to abort\n");
12315 #endif
12316                                 break;
12317                         }
12318                 }
12319         }
12320         mtx_unlock(&lun->lun_lock);
12321
12322         if (found == 0) {
12323                 /*
12324                  * This isn't really an error.  It's entirely possible for
12325                  * the abort and command completion to cross on the wire.
12326                  * This is more of an informative/diagnostic error.
12327                  */
12328 #if 0
12329                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12330                        "%d:%d:%d:%d tag %d type %d\n",
12331                        io->io_hdr.nexus.initid.id,
12332                        io->io_hdr.nexus.targ_port,
12333                        io->io_hdr.nexus.targ_target.id,
12334                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12335                        io->taskio.tag_type);
12336 #endif
12337         }
12338         return (0);
12339 }
12340
12341 static void
12342 ctl_run_task(union ctl_io *io)
12343 {
12344         struct ctl_softc *ctl_softc = control_softc;
12345         int retval = 1;
12346         const char *task_desc;
12347
12348         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12349
12350         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12351             ("ctl_run_task: Unextected io_type %d\n",
12352              io->io_hdr.io_type));
12353
12354         task_desc = ctl_scsi_task_string(&io->taskio);
12355         if (task_desc != NULL) {
12356 #ifdef NEEDTOPORT
12357                 csevent_log(CSC_CTL | CSC_SHELF_SW |
12358                             CTL_TASK_REPORT,
12359                             csevent_LogType_Trace,
12360                             csevent_Severity_Information,
12361                             csevent_AlertLevel_Green,
12362                             csevent_FRU_Firmware,
12363                             csevent_FRU_Unknown,
12364                             "CTL: received task: %s",task_desc);
12365 #endif
12366         } else {
12367 #ifdef NEEDTOPORT
12368                 csevent_log(CSC_CTL | CSC_SHELF_SW |
12369                             CTL_TASK_REPORT,
12370                             csevent_LogType_Trace,
12371                             csevent_Severity_Information,
12372                             csevent_AlertLevel_Green,
12373                             csevent_FRU_Firmware,
12374                             csevent_FRU_Unknown,
12375                             "CTL: received unknown task "
12376                             "type: %d (%#x)",
12377                             io->taskio.task_action,
12378                             io->taskio.task_action);
12379 #endif
12380         }
12381         switch (io->taskio.task_action) {
12382         case CTL_TASK_ABORT_TASK:
12383                 retval = ctl_abort_task(io);
12384                 break;
12385         case CTL_TASK_ABORT_TASK_SET:
12386         case CTL_TASK_CLEAR_TASK_SET:
12387                 retval = ctl_abort_task_set(io);
12388                 break;
12389         case CTL_TASK_CLEAR_ACA:
12390                 break;
12391         case CTL_TASK_I_T_NEXUS_RESET:
12392                 retval = ctl_i_t_nexus_reset(io);
12393                 break;
12394         case CTL_TASK_LUN_RESET: {
12395                 struct ctl_lun *lun;
12396                 uint32_t targ_lun;
12397
12398                 targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12399                 mtx_lock(&ctl_softc->ctl_lock);
12400                 if ((targ_lun < CTL_MAX_LUNS)
12401                  && (ctl_softc->ctl_luns[targ_lun] != NULL))
12402                         lun = ctl_softc->ctl_luns[targ_lun];
12403                 else {
12404                         mtx_unlock(&ctl_softc->ctl_lock);
12405                         retval = 1;
12406                         break;
12407                 }
12408
12409                 if (!(io->io_hdr.flags &
12410                     CTL_FLAG_FROM_OTHER_SC)) {
12411                         union ctl_ha_msg msg_info;
12412
12413                         io->io_hdr.flags |=
12414                                 CTL_FLAG_SENT_2OTHER_SC;
12415                         msg_info.hdr.msg_type =
12416                                 CTL_MSG_MANAGE_TASKS;
12417                         msg_info.hdr.nexus = io->io_hdr.nexus;
12418                         msg_info.task.task_action =
12419                                 CTL_TASK_LUN_RESET;
12420                         msg_info.hdr.original_sc = NULL;
12421                         msg_info.hdr.serializing_sc = NULL;
12422                         if (CTL_HA_STATUS_SUCCESS !=
12423                             ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12424                             (void *)&msg_info,
12425                             sizeof(msg_info), 0)) {
12426                         }
12427                 }
12428
12429                 retval = ctl_lun_reset(lun, io,
12430                                        CTL_UA_LUN_RESET);
12431                 mtx_unlock(&ctl_softc->ctl_lock);
12432                 break;
12433         }
12434         case CTL_TASK_TARGET_RESET:
12435                 retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12436                 break;
12437         case CTL_TASK_BUS_RESET:
12438                 retval = ctl_bus_reset(ctl_softc, io);
12439                 break;
12440         case CTL_TASK_PORT_LOGIN:
12441                 break;
12442         case CTL_TASK_PORT_LOGOUT:
12443                 break;
12444         default:
12445                 printf("ctl_run_task: got unknown task management event %d\n",
12446                        io->taskio.task_action);
12447                 break;
12448         }
12449         if (retval == 0)
12450                 io->io_hdr.status = CTL_SUCCESS;
12451         else
12452                 io->io_hdr.status = CTL_ERROR;
12453         ctl_done(io);
12454 }
12455
12456 /*
12457  * For HA operation.  Handle commands that come in from the other
12458  * controller.
12459  */
12460 static void
12461 ctl_handle_isc(union ctl_io *io)
12462 {
12463         int free_io;
12464         struct ctl_lun *lun;
12465         struct ctl_softc *ctl_softc;
12466         uint32_t targ_lun;
12467
12468         ctl_softc = control_softc;
12469
12470         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12471         lun = ctl_softc->ctl_luns[targ_lun];
12472
12473         switch (io->io_hdr.msg_type) {
12474         case CTL_MSG_SERIALIZE:
12475                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12476                 break;
12477         case CTL_MSG_R2R: {
12478                 const struct ctl_cmd_entry *entry;
12479
12480                 /*
12481                  * This is only used in SER_ONLY mode.
12482                  */
12483                 free_io = 0;
12484                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12485                 mtx_lock(&lun->lun_lock);
12486                 if (ctl_scsiio_lun_check(ctl_softc, lun,
12487                     entry, (struct ctl_scsiio *)io) != 0) {
12488                         mtx_unlock(&lun->lun_lock);
12489                         ctl_done(io);
12490                         break;
12491                 }
12492                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12493                 mtx_unlock(&lun->lun_lock);
12494                 ctl_enqueue_rtr(io);
12495                 break;
12496         }
12497         case CTL_MSG_FINISH_IO:
12498                 if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12499                         free_io = 0;
12500                         ctl_done(io);
12501                 } else {
12502                         free_io = 1;
12503                         mtx_lock(&lun->lun_lock);
12504                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12505                                      ooa_links);
12506                         ctl_check_blocked(lun);
12507                         mtx_unlock(&lun->lun_lock);
12508                 }
12509                 break;
12510         case CTL_MSG_PERS_ACTION:
12511                 ctl_hndl_per_res_out_on_other_sc(
12512                         (union ctl_ha_msg *)&io->presio.pr_msg);
12513                 free_io = 1;
12514                 break;
12515         case CTL_MSG_BAD_JUJU:
12516                 free_io = 0;
12517                 ctl_done(io);
12518                 break;
12519         case CTL_MSG_DATAMOVE:
12520                 /* Only used in XFER mode */
12521                 free_io = 0;
12522                 ctl_datamove_remote(io);
12523                 break;
12524         case CTL_MSG_DATAMOVE_DONE:
12525                 /* Only used in XFER mode */
12526                 free_io = 0;
12527                 io->scsiio.be_move_done(io);
12528                 break;
12529         default:
12530                 free_io = 1;
12531                 printf("%s: Invalid message type %d\n",
12532                        __func__, io->io_hdr.msg_type);
12533                 break;
12534         }
12535         if (free_io)
12536                 ctl_free_io(io);
12537
12538 }
12539
12540
12541 /*
12542  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12543  * there is no match.
12544  */
12545 static ctl_lun_error_pattern
12546 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12547 {
12548         const struct ctl_cmd_entry *entry;
12549         ctl_lun_error_pattern filtered_pattern, pattern;
12550
12551         pattern = desc->error_pattern;
12552
12553         /*
12554          * XXX KDM we need more data passed into this function to match a
12555          * custom pattern, and we actually need to implement custom pattern
12556          * matching.
12557          */
12558         if (pattern & CTL_LUN_PAT_CMD)
12559                 return (CTL_LUN_PAT_CMD);
12560
12561         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12562                 return (CTL_LUN_PAT_ANY);
12563
12564         entry = ctl_get_cmd_entry(ctsio, NULL);
12565
12566         filtered_pattern = entry->pattern & pattern;
12567
12568         /*
12569          * If the user requested specific flags in the pattern (e.g.
12570          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12571          * flags.
12572          *
12573          * If the user did not specify any flags, it doesn't matter whether
12574          * or not the command supports the flags.
12575          */
12576         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12577              (pattern & ~CTL_LUN_PAT_MASK))
12578                 return (CTL_LUN_PAT_NONE);
12579
12580         /*
12581          * If the user asked for a range check, see if the requested LBA
12582          * range overlaps with this command's LBA range.
12583          */
12584         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12585                 uint64_t lba1;
12586                 uint64_t len1;
12587                 ctl_action action;
12588                 int retval;
12589
12590                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12591                 if (retval != 0)
12592                         return (CTL_LUN_PAT_NONE);
12593
12594                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12595                                               desc->lba_range.len);
12596                 /*
12597                  * A "pass" means that the LBA ranges don't overlap, so
12598                  * this doesn't match the user's range criteria.
12599                  */
12600                 if (action == CTL_ACTION_PASS)
12601                         return (CTL_LUN_PAT_NONE);
12602         }
12603
12604         return (filtered_pattern);
12605 }
12606
12607 static void
12608 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12609 {
12610         struct ctl_error_desc *desc, *desc2;
12611
12612         mtx_assert(&lun->lun_lock, MA_OWNED);
12613
12614         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12615                 ctl_lun_error_pattern pattern;
12616                 /*
12617                  * Check to see whether this particular command matches
12618                  * the pattern in the descriptor.
12619                  */
12620                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12621                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12622                         continue;
12623
12624                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12625                 case CTL_LUN_INJ_ABORTED:
12626                         ctl_set_aborted(&io->scsiio);
12627                         break;
12628                 case CTL_LUN_INJ_MEDIUM_ERR:
12629                         ctl_set_medium_error(&io->scsiio);
12630                         break;
12631                 case CTL_LUN_INJ_UA:
12632                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12633                          * OCCURRED */
12634                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12635                         break;
12636                 case CTL_LUN_INJ_CUSTOM:
12637                         /*
12638                          * We're assuming the user knows what he is doing.
12639                          * Just copy the sense information without doing
12640                          * checks.
12641                          */
12642                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12643                               ctl_min(sizeof(desc->custom_sense),
12644                                       sizeof(io->scsiio.sense_data)));
12645                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12646                         io->scsiio.sense_len = SSD_FULL_SIZE;
12647                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12648                         break;
12649                 case CTL_LUN_INJ_NONE:
12650                 default:
12651                         /*
12652                          * If this is an error injection type we don't know
12653                          * about, clear the continuous flag (if it is set)
12654                          * so it will get deleted below.
12655                          */
12656                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12657                         break;
12658                 }
12659                 /*
12660                  * By default, each error injection action is a one-shot
12661                  */
12662                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12663                         continue;
12664
12665                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12666
12667                 free(desc, M_CTL);
12668         }
12669 }
12670
12671 #ifdef CTL_IO_DELAY
12672 static void
12673 ctl_datamove_timer_wakeup(void *arg)
12674 {
12675         union ctl_io *io;
12676
12677         io = (union ctl_io *)arg;
12678
12679         ctl_datamove(io);
12680 }
12681 #endif /* CTL_IO_DELAY */
12682
12683 void
12684 ctl_datamove(union ctl_io *io)
12685 {
12686         void (*fe_datamove)(union ctl_io *io);
12687
12688         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12689
12690         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12691
12692 #ifdef CTL_TIME_IO
12693         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12694                 char str[256];
12695                 char path_str[64];
12696                 struct sbuf sb;
12697
12698                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12699                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12700
12701                 sbuf_cat(&sb, path_str);
12702                 switch (io->io_hdr.io_type) {
12703                 case CTL_IO_SCSI:
12704                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12705                         sbuf_printf(&sb, "\n");
12706                         sbuf_cat(&sb, path_str);
12707                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12708                                     io->scsiio.tag_num, io->scsiio.tag_type);
12709                         break;
12710                 case CTL_IO_TASK:
12711                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12712                                     "Tag Type: %d\n", io->taskio.task_action,
12713                                     io->taskio.tag_num, io->taskio.tag_type);
12714                         break;
12715                 default:
12716                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12717                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12718                         break;
12719                 }
12720                 sbuf_cat(&sb, path_str);
12721                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12722                             (intmax_t)time_uptime - io->io_hdr.start_time);
12723                 sbuf_finish(&sb);
12724                 printf("%s", sbuf_data(&sb));
12725         }
12726 #endif /* CTL_TIME_IO */
12727
12728 #ifdef CTL_IO_DELAY
12729         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12730                 struct ctl_lun *lun;
12731
12732                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12733
12734                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12735         } else {
12736                 struct ctl_lun *lun;
12737
12738                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12739                 if ((lun != NULL)
12740                  && (lun->delay_info.datamove_delay > 0)) {
12741                         struct callout *callout;
12742
12743                         callout = (struct callout *)&io->io_hdr.timer_bytes;
12744                         callout_init(callout, /*mpsafe*/ 1);
12745                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12746                         callout_reset(callout,
12747                                       lun->delay_info.datamove_delay * hz,
12748                                       ctl_datamove_timer_wakeup, io);
12749                         if (lun->delay_info.datamove_type ==
12750                             CTL_DELAY_TYPE_ONESHOT)
12751                                 lun->delay_info.datamove_delay = 0;
12752                         return;
12753                 }
12754         }
12755 #endif
12756
12757         /*
12758          * This command has been aborted.  Set the port status, so we fail
12759          * the data move.
12760          */
12761         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12762                 printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12763                        io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12764                        io->io_hdr.nexus.targ_port,
12765                        (uintmax_t)io->io_hdr.nexus.targ_target.id,
12766                        io->io_hdr.nexus.targ_lun);
12767                 io->io_hdr.port_status = 31337;
12768                 /*
12769                  * Note that the backend, in this case, will get the
12770                  * callback in its context.  In other cases it may get
12771                  * called in the frontend's interrupt thread context.
12772                  */
12773                 io->scsiio.be_move_done(io);
12774                 return;
12775         }
12776
12777         /*
12778          * If we're in XFER mode and this I/O is from the other shelf
12779          * controller, we need to send the DMA to the other side to
12780          * actually transfer the data to/from the host.  In serialize only
12781          * mode the transfer happens below CTL and ctl_datamove() is only
12782          * called on the machine that originally received the I/O.
12783          */
12784         if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12785          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12786                 union ctl_ha_msg msg;
12787                 uint32_t sg_entries_sent;
12788                 int do_sg_copy;
12789                 int i;
12790
12791                 memset(&msg, 0, sizeof(msg));
12792                 msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12793                 msg.hdr.original_sc = io->io_hdr.original_sc;
12794                 msg.hdr.serializing_sc = io;
12795                 msg.hdr.nexus = io->io_hdr.nexus;
12796                 msg.dt.flags = io->io_hdr.flags;
12797                 /*
12798                  * We convert everything into a S/G list here.  We can't
12799                  * pass by reference, only by value between controllers.
12800                  * So we can't pass a pointer to the S/G list, only as many
12801                  * S/G entries as we can fit in here.  If it's possible for
12802                  * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12803                  * then we need to break this up into multiple transfers.
12804                  */
12805                 if (io->scsiio.kern_sg_entries == 0) {
12806                         msg.dt.kern_sg_entries = 1;
12807                         /*
12808                          * If this is in cached memory, flush the cache
12809                          * before we send the DMA request to the other
12810                          * controller.  We want to do this in either the
12811                          * read or the write case.  The read case is
12812                          * straightforward.  In the write case, we want to
12813                          * make sure nothing is in the local cache that
12814                          * could overwrite the DMAed data.
12815                          */
12816                         if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12817                                 /*
12818                                  * XXX KDM use bus_dmamap_sync() here.
12819                                  */
12820                         }
12821
12822                         /*
12823                          * Convert to a physical address if this is a
12824                          * virtual address.
12825                          */
12826                         if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12827                                 msg.dt.sg_list[0].addr =
12828                                         io->scsiio.kern_data_ptr;
12829                         } else {
12830                                 /*
12831                                  * XXX KDM use busdma here!
12832                                  */
12833 #if 0
12834                                 msg.dt.sg_list[0].addr = (void *)
12835                                         vtophys(io->scsiio.kern_data_ptr);
12836 #endif
12837                         }
12838
12839                         msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12840                         do_sg_copy = 0;
12841                 } else {
12842                         struct ctl_sg_entry *sgl;
12843
12844                         do_sg_copy = 1;
12845                         msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12846                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12847                         if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12848                                 /*
12849                                  * XXX KDM use bus_dmamap_sync() here.
12850                                  */
12851                         }
12852                 }
12853
12854                 msg.dt.kern_data_len = io->scsiio.kern_data_len;
12855                 msg.dt.kern_total_len = io->scsiio.kern_total_len;
12856                 msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12857                 msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12858                 msg.dt.sg_sequence = 0;
12859
12860                 /*
12861                  * Loop until we've sent all of the S/G entries.  On the
12862                  * other end, we'll recompose these S/G entries into one
12863                  * contiguous list before passing it to the
12864                  */
12865                 for (sg_entries_sent = 0; sg_entries_sent <
12866                      msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12867                         msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12868                                 sizeof(msg.dt.sg_list[0])),
12869                                 msg.dt.kern_sg_entries - sg_entries_sent);
12870
12871                         if (do_sg_copy != 0) {
12872                                 struct ctl_sg_entry *sgl;
12873                                 int j;
12874
12875                                 sgl = (struct ctl_sg_entry *)
12876                                         io->scsiio.kern_data_ptr;
12877                                 /*
12878                                  * If this is in cached memory, flush the cache
12879                                  * before we send the DMA request to the other
12880                                  * controller.  We want to do this in either
12881                                  * the * read or the write case.  The read
12882                                  * case is straightforward.  In the write
12883                                  * case, we want to make sure nothing is
12884                                  * in the local cache that could overwrite
12885                                  * the DMAed data.
12886                                  */
12887
12888                                 for (i = sg_entries_sent, j = 0;
12889                                      i < msg.dt.cur_sg_entries; i++, j++) {
12890                                         if ((io->io_hdr.flags &
12891                                              CTL_FLAG_NO_DATASYNC) == 0) {
12892                                                 /*
12893                                                  * XXX KDM use bus_dmamap_sync()
12894                                                  */
12895                                         }
12896                                         if ((io->io_hdr.flags &
12897                                              CTL_FLAG_BUS_ADDR) == 0) {
12898                                                 /*
12899                                                  * XXX KDM use busdma.
12900                                                  */
12901 #if 0
12902                                                 msg.dt.sg_list[j].addr =(void *)
12903                                                        vtophys(sgl[i].addr);
12904 #endif
12905                                         } else {
12906                                                 msg.dt.sg_list[j].addr =
12907                                                         sgl[i].addr;
12908                                         }
12909                                         msg.dt.sg_list[j].len = sgl[i].len;
12910                                 }
12911                         }
12912
12913                         sg_entries_sent += msg.dt.cur_sg_entries;
12914                         if (sg_entries_sent >= msg.dt.kern_sg_entries)
12915                                 msg.dt.sg_last = 1;
12916                         else
12917                                 msg.dt.sg_last = 0;
12918
12919                         /*
12920                          * XXX KDM drop and reacquire the lock here?
12921                          */
12922                         if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12923                             sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12924                                 /*
12925                                  * XXX do something here.
12926                                  */
12927                         }
12928
12929                         msg.dt.sent_sg_entries = sg_entries_sent;
12930                 }
12931                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12932                 if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12933                         ctl_failover_io(io, /*have_lock*/ 0);
12934
12935         } else {
12936
12937                 /*
12938                  * Lookup the fe_datamove() function for this particular
12939                  * front end.
12940                  */
12941                 fe_datamove =
12942                     control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12943
12944                 fe_datamove(io);
12945         }
12946 }
12947
12948 static void
12949 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12950 {
12951         union ctl_ha_msg msg;
12952         int isc_status;
12953
12954         memset(&msg, 0, sizeof(msg));
12955
12956         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12957         msg.hdr.original_sc = io;
12958         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12959         msg.hdr.nexus = io->io_hdr.nexus;
12960         msg.hdr.status = io->io_hdr.status;
12961         msg.scsi.tag_num = io->scsiio.tag_num;
12962         msg.scsi.tag_type = io->scsiio.tag_type;
12963         msg.scsi.scsi_status = io->scsiio.scsi_status;
12964         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12965                sizeof(io->scsiio.sense_data));
12966         msg.scsi.sense_len = io->scsiio.sense_len;
12967         msg.scsi.sense_residual = io->scsiio.sense_residual;
12968         msg.scsi.fetd_status = io->io_hdr.port_status;
12969         msg.scsi.residual = io->scsiio.residual;
12970         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12971
12972         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12973                 ctl_failover_io(io, /*have_lock*/ have_lock);
12974                 return;
12975         }
12976
12977         isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12978         if (isc_status > CTL_HA_STATUS_SUCCESS) {
12979                 /* XXX do something if this fails */
12980         }
12981
12982 }
12983
12984 /*
12985  * The DMA to the remote side is done, now we need to tell the other side
12986  * we're done so it can continue with its data movement.
12987  */
12988 static void
12989 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12990 {
12991         union ctl_io *io;
12992
12993         io = rq->context;
12994
12995         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12996                 printf("%s: ISC DMA write failed with error %d", __func__,
12997                        rq->ret);
12998                 ctl_set_internal_failure(&io->scsiio,
12999                                          /*sks_valid*/ 1,
13000                                          /*retry_count*/ rq->ret);
13001         }
13002
13003         ctl_dt_req_free(rq);
13004
13005         /*
13006          * In this case, we had to malloc the memory locally.  Free it.
13007          */
13008         if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13009                 int i;
13010                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13011                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13012         }
13013         /*
13014          * The data is in local and remote memory, so now we need to send
13015          * status (good or back) back to the other side.
13016          */
13017         ctl_send_datamove_done(io, /*have_lock*/ 0);
13018 }
13019
13020 /*
13021  * We've moved the data from the host/controller into local memory.  Now we
13022  * need to push it over to the remote controller's memory.
13023  */
13024 static int
13025 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
13026 {
13027         int retval;
13028
13029         retval = 0;
13030
13031         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
13032                                           ctl_datamove_remote_write_cb);
13033
13034         return (retval);
13035 }
13036
13037 static void
13038 ctl_datamove_remote_write(union ctl_io *io)
13039 {
13040         int retval;
13041         void (*fe_datamove)(union ctl_io *io);
13042
13043         /*
13044          * - Get the data from the host/HBA into local memory.
13045          * - DMA memory from the local controller to the remote controller.
13046          * - Send status back to the remote controller.
13047          */
13048
13049         retval = ctl_datamove_remote_sgl_setup(io);
13050         if (retval != 0)
13051                 return;
13052
13053         /* Switch the pointer over so the FETD knows what to do */
13054         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13055
13056         /*
13057          * Use a custom move done callback, since we need to send completion
13058          * back to the other controller, not to the backend on this side.
13059          */
13060         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
13061
13062         fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13063
13064         fe_datamove(io);
13065
13066         return;
13067
13068 }
13069
13070 static int
13071 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
13072 {
13073 #if 0
13074         char str[256];
13075         char path_str[64];
13076         struct sbuf sb;
13077 #endif
13078
13079         /*
13080          * In this case, we had to malloc the memory locally.  Free it.
13081          */
13082         if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13083                 int i;
13084                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13085                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13086         }
13087
13088 #if 0
13089         scsi_path_string(io, path_str, sizeof(path_str));
13090         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13091         sbuf_cat(&sb, path_str);
13092         scsi_command_string(&io->scsiio, NULL, &sb);
13093         sbuf_printf(&sb, "\n");
13094         sbuf_cat(&sb, path_str);
13095         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13096                     io->scsiio.tag_num, io->scsiio.tag_type);
13097         sbuf_cat(&sb, path_str);
13098         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
13099                     io->io_hdr.flags, io->io_hdr.status);
13100         sbuf_finish(&sb);
13101         printk("%s", sbuf_data(&sb));
13102 #endif
13103
13104
13105         /*
13106          * The read is done, now we need to send status (good or bad) back
13107          * to the other side.
13108          */
13109         ctl_send_datamove_done(io, /*have_lock*/ 0);
13110
13111         return (0);
13112 }
13113
13114 static void
13115 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
13116 {
13117         union ctl_io *io;
13118         void (*fe_datamove)(union ctl_io *io);
13119
13120         io = rq->context;
13121
13122         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13123                 printf("%s: ISC DMA read failed with error %d", __func__,
13124                        rq->ret);
13125                 ctl_set_internal_failure(&io->scsiio,
13126                                          /*sks_valid*/ 1,
13127                                          /*retry_count*/ rq->ret);
13128         }
13129
13130         ctl_dt_req_free(rq);
13131
13132         /* Switch the pointer over so the FETD knows what to do */
13133         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13134
13135         /*
13136          * Use a custom move done callback, since we need to send completion
13137          * back to the other controller, not to the backend on this side.
13138          */
13139         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13140
13141         /* XXX KDM add checks like the ones in ctl_datamove? */
13142
13143         fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13144
13145         fe_datamove(io);
13146 }
13147
13148 static int
13149 ctl_datamove_remote_sgl_setup(union ctl_io *io)
13150 {
13151         struct ctl_sg_entry *local_sglist, *remote_sglist;
13152         struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13153         struct ctl_softc *softc;
13154         int retval;
13155         int i;
13156
13157         retval = 0;
13158         softc = control_softc;
13159
13160         local_sglist = io->io_hdr.local_sglist;
13161         local_dma_sglist = io->io_hdr.local_dma_sglist;
13162         remote_sglist = io->io_hdr.remote_sglist;
13163         remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13164
13165         if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13166                 for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13167                         local_sglist[i].len = remote_sglist[i].len;
13168
13169                         /*
13170                          * XXX Detect the situation where the RS-level I/O
13171                          * redirector on the other side has already read the
13172                          * data off of the AOR RS on this side, and
13173                          * transferred it to remote (mirror) memory on the
13174                          * other side.  Since we already have the data in
13175                          * memory here, we just need to use it.
13176                          *
13177                          * XXX KDM this can probably be removed once we
13178                          * get the cache device code in and take the
13179                          * current AOR implementation out.
13180                          */
13181 #ifdef NEEDTOPORT
13182                         if ((remote_sglist[i].addr >=
13183                              (void *)vtophys(softc->mirr->addr))
13184                          && (remote_sglist[i].addr <
13185                              ((void *)vtophys(softc->mirr->addr) +
13186                              CacheMirrorOffset))) {
13187                                 local_sglist[i].addr = remote_sglist[i].addr -
13188                                         CacheMirrorOffset;
13189                                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13190                                      CTL_FLAG_DATA_IN)
13191                                         io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13192                         } else {
13193                                 local_sglist[i].addr = remote_sglist[i].addr +
13194                                         CacheMirrorOffset;
13195                         }
13196 #endif
13197 #if 0
13198                         printf("%s: local %p, remote %p, len %d\n",
13199                                __func__, local_sglist[i].addr,
13200                                remote_sglist[i].addr, local_sglist[i].len);
13201 #endif
13202                 }
13203         } else {
13204                 uint32_t len_to_go;
13205
13206                 /*
13207                  * In this case, we don't have automatically allocated
13208                  * memory for this I/O on this controller.  This typically
13209                  * happens with internal CTL I/O -- e.g. inquiry, mode
13210                  * sense, etc.  Anything coming from RAIDCore will have
13211                  * a mirror area available.
13212                  */
13213                 len_to_go = io->scsiio.kern_data_len;
13214
13215                 /*
13216                  * Clear the no datasync flag, we have to use malloced
13217                  * buffers.
13218                  */
13219                 io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13220
13221                 /*
13222                  * The difficult thing here is that the size of the various
13223                  * S/G segments may be different than the size from the
13224                  * remote controller.  That'll make it harder when DMAing
13225                  * the data back to the other side.
13226                  */
13227                 for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13228                      sizeof(io->io_hdr.remote_sglist[0])) &&
13229                      (len_to_go > 0); i++) {
13230                         local_sglist[i].len = ctl_min(len_to_go, 131072);
13231                         CTL_SIZE_8B(local_dma_sglist[i].len,
13232                                     local_sglist[i].len);
13233                         local_sglist[i].addr =
13234                                 malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13235
13236                         local_dma_sglist[i].addr = local_sglist[i].addr;
13237
13238                         if (local_sglist[i].addr == NULL) {
13239                                 int j;
13240
13241                                 printf("malloc failed for %zd bytes!",
13242                                        local_dma_sglist[i].len);
13243                                 for (j = 0; j < i; j++) {
13244                                         free(local_sglist[j].addr, M_CTL);
13245                                 }
13246                                 ctl_set_internal_failure(&io->scsiio,
13247                                                          /*sks_valid*/ 1,
13248                                                          /*retry_count*/ 4857);
13249                                 retval = 1;
13250                                 goto bailout_error;
13251
13252                         }
13253                         /* XXX KDM do we need a sync here? */
13254
13255                         len_to_go -= local_sglist[i].len;
13256                 }
13257                 /*
13258                  * Reset the number of S/G entries accordingly.  The
13259                  * original number of S/G entries is available in
13260                  * rem_sg_entries.
13261                  */
13262                 io->scsiio.kern_sg_entries = i;
13263
13264 #if 0
13265                 printf("%s: kern_sg_entries = %d\n", __func__,
13266                        io->scsiio.kern_sg_entries);
13267                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13268                         printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13269                                local_sglist[i].addr, local_sglist[i].len,
13270                                local_dma_sglist[i].len);
13271 #endif
13272         }
13273
13274
13275         return (retval);
13276
13277 bailout_error:
13278
13279         ctl_send_datamove_done(io, /*have_lock*/ 0);
13280
13281         return (retval);
13282 }
13283
13284 static int
13285 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13286                          ctl_ha_dt_cb callback)
13287 {
13288         struct ctl_ha_dt_req *rq;
13289         struct ctl_sg_entry *remote_sglist, *local_sglist;
13290         struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13291         uint32_t local_used, remote_used, total_used;
13292         int retval;
13293         int i, j;
13294
13295         retval = 0;
13296
13297         rq = ctl_dt_req_alloc();
13298
13299         /*
13300          * If we failed to allocate the request, and if the DMA didn't fail
13301          * anyway, set busy status.  This is just a resource allocation
13302          * failure.
13303          */
13304         if ((rq == NULL)
13305          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13306                 ctl_set_busy(&io->scsiio);
13307
13308         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13309
13310                 if (rq != NULL)
13311                         ctl_dt_req_free(rq);
13312
13313                 /*
13314                  * The data move failed.  We need to return status back
13315                  * to the other controller.  No point in trying to DMA
13316                  * data to the remote controller.
13317                  */
13318
13319                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13320
13321                 retval = 1;
13322
13323                 goto bailout;
13324         }
13325
13326         local_sglist = io->io_hdr.local_sglist;
13327         local_dma_sglist = io->io_hdr.local_dma_sglist;
13328         remote_sglist = io->io_hdr.remote_sglist;
13329         remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13330         local_used = 0;
13331         remote_used = 0;
13332         total_used = 0;
13333
13334         if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13335                 rq->ret = CTL_HA_STATUS_SUCCESS;
13336                 rq->context = io;
13337                 callback(rq);
13338                 goto bailout;
13339         }
13340
13341         /*
13342          * Pull/push the data over the wire from/to the other controller.
13343          * This takes into account the possibility that the local and
13344          * remote sglists may not be identical in terms of the size of
13345          * the elements and the number of elements.
13346          *
13347          * One fundamental assumption here is that the length allocated for
13348          * both the local and remote sglists is identical.  Otherwise, we've
13349          * essentially got a coding error of some sort.
13350          */
13351         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13352                 int isc_ret;
13353                 uint32_t cur_len, dma_length;
13354                 uint8_t *tmp_ptr;
13355
13356                 rq->id = CTL_HA_DATA_CTL;
13357                 rq->command = command;
13358                 rq->context = io;
13359
13360                 /*
13361                  * Both pointers should be aligned.  But it is possible
13362                  * that the allocation length is not.  They should both
13363                  * also have enough slack left over at the end, though,
13364                  * to round up to the next 8 byte boundary.
13365                  */
13366                 cur_len = ctl_min(local_sglist[i].len - local_used,
13367                                   remote_sglist[j].len - remote_used);
13368
13369                 /*
13370                  * In this case, we have a size issue and need to decrease
13371                  * the size, except in the case where we actually have less
13372                  * than 8 bytes left.  In that case, we need to increase
13373                  * the DMA length to get the last bit.
13374                  */
13375                 if ((cur_len & 0x7) != 0) {
13376                         if (cur_len > 0x7) {
13377                                 cur_len = cur_len - (cur_len & 0x7);
13378                                 dma_length = cur_len;
13379                         } else {
13380                                 CTL_SIZE_8B(dma_length, cur_len);
13381                         }
13382
13383                 } else
13384                         dma_length = cur_len;
13385
13386                 /*
13387                  * If we had to allocate memory for this I/O, instead of using
13388                  * the non-cached mirror memory, we'll need to flush the cache
13389                  * before trying to DMA to the other controller.
13390                  *
13391                  * We could end up doing this multiple times for the same
13392                  * segment if we have a larger local segment than remote
13393                  * segment.  That shouldn't be an issue.
13394                  */
13395                 if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13396                         /*
13397                          * XXX KDM use bus_dmamap_sync() here.
13398                          */
13399                 }
13400
13401                 rq->size = dma_length;
13402
13403                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
13404                 tmp_ptr += local_used;
13405
13406                 /* Use physical addresses when talking to ISC hardware */
13407                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13408                         /* XXX KDM use busdma */
13409 #if 0
13410                         rq->local = vtophys(tmp_ptr);
13411 #endif
13412                 } else
13413                         rq->local = tmp_ptr;
13414
13415                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13416                 tmp_ptr += remote_used;
13417                 rq->remote = tmp_ptr;
13418
13419                 rq->callback = NULL;
13420
13421                 local_used += cur_len;
13422                 if (local_used >= local_sglist[i].len) {
13423                         i++;
13424                         local_used = 0;
13425                 }
13426
13427                 remote_used += cur_len;
13428                 if (remote_used >= remote_sglist[j].len) {
13429                         j++;
13430                         remote_used = 0;
13431                 }
13432                 total_used += cur_len;
13433
13434                 if (total_used >= io->scsiio.kern_data_len)
13435                         rq->callback = callback;
13436
13437                 if ((rq->size & 0x7) != 0) {
13438                         printf("%s: warning: size %d is not on 8b boundary\n",
13439                                __func__, rq->size);
13440                 }
13441                 if (((uintptr_t)rq->local & 0x7) != 0) {
13442                         printf("%s: warning: local %p not on 8b boundary\n",
13443                                __func__, rq->local);
13444                 }
13445                 if (((uintptr_t)rq->remote & 0x7) != 0) {
13446                         printf("%s: warning: remote %p not on 8b boundary\n",
13447                                __func__, rq->local);
13448                 }
13449 #if 0
13450                 printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13451                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13452                        rq->local, rq->remote, rq->size);
13453 #endif
13454
13455                 isc_ret = ctl_dt_single(rq);
13456                 if (isc_ret == CTL_HA_STATUS_WAIT)
13457                         continue;
13458
13459                 if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13460                         rq->ret = CTL_HA_STATUS_SUCCESS;
13461                 } else {
13462                         rq->ret = isc_ret;
13463                 }
13464                 callback(rq);
13465                 goto bailout;
13466         }
13467
13468 bailout:
13469         return (retval);
13470
13471 }
13472
13473 static void
13474 ctl_datamove_remote_read(union ctl_io *io)
13475 {
13476         int retval;
13477         int i;
13478
13479         /*
13480          * This will send an error to the other controller in the case of a
13481          * failure.
13482          */
13483         retval = ctl_datamove_remote_sgl_setup(io);
13484         if (retval != 0)
13485                 return;
13486
13487         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13488                                           ctl_datamove_remote_read_cb);
13489         if ((retval != 0)
13490          && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13491                 /*
13492                  * Make sure we free memory if there was an error..  The
13493                  * ctl_datamove_remote_xfer() function will send the
13494                  * datamove done message, or call the callback with an
13495                  * error if there is a problem.
13496                  */
13497                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13498                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13499         }
13500
13501         return;
13502 }
13503
13504 /*
13505  * Process a datamove request from the other controller.  This is used for
13506  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13507  * first.  Once that is complete, the data gets DMAed into the remote
13508  * controller's memory.  For reads, we DMA from the remote controller's
13509  * memory into our memory first, and then move it out to the FETD.
13510  */
13511 static void
13512 ctl_datamove_remote(union ctl_io *io)
13513 {
13514         struct ctl_softc *softc;
13515
13516         softc = control_softc;
13517
13518         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13519
13520         /*
13521          * Note that we look for an aborted I/O here, but don't do some of
13522          * the other checks that ctl_datamove() normally does.
13523          * We don't need to run the datamove delay code, since that should
13524          * have been done if need be on the other controller.
13525          */
13526         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13527                 printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13528                        io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13529                        io->io_hdr.nexus.targ_port,
13530                        io->io_hdr.nexus.targ_target.id,
13531                        io->io_hdr.nexus.targ_lun);
13532                 io->io_hdr.port_status = 31338;
13533                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13534                 return;
13535         }
13536
13537         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13538                 ctl_datamove_remote_write(io);
13539         } else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13540                 ctl_datamove_remote_read(io);
13541         } else {
13542                 union ctl_ha_msg msg;
13543                 struct scsi_sense_data *sense;
13544                 uint8_t sks[3];
13545                 int retry_count;
13546
13547                 memset(&msg, 0, sizeof(msg));
13548
13549                 msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13550                 msg.hdr.status = CTL_SCSI_ERROR;
13551                 msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13552
13553                 retry_count = 4243;
13554
13555                 sense = &msg.scsi.sense_data;
13556                 sks[0] = SSD_SCS_VALID;
13557                 sks[1] = (retry_count >> 8) & 0xff;
13558                 sks[2] = retry_count & 0xff;
13559
13560                 /* "Internal target failure" */
13561                 scsi_set_sense_data(sense,
13562                                     /*sense_format*/ SSD_TYPE_NONE,
13563                                     /*current_error*/ 1,
13564                                     /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13565                                     /*asc*/ 0x44,
13566                                     /*ascq*/ 0x00,
13567                                     /*type*/ SSD_ELEM_SKS,
13568                                     /*size*/ sizeof(sks),
13569                                     /*data*/ sks,
13570                                     SSD_ELEM_NONE);
13571
13572                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13573                 if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13574                         ctl_failover_io(io, /*have_lock*/ 1);
13575                         return;
13576                 }
13577
13578                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13579                     CTL_HA_STATUS_SUCCESS) {
13580                         /* XXX KDM what to do if this fails? */
13581                 }
13582                 return;
13583         }
13584
13585 }
13586
13587 static int
13588 ctl_process_done(union ctl_io *io)
13589 {
13590         struct ctl_lun *lun;
13591         struct ctl_softc *ctl_softc;
13592         void (*fe_done)(union ctl_io *io);
13593         uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13594
13595         CTL_DEBUG_PRINT(("ctl_process_done\n"));
13596
13597         fe_done =
13598             control_softc->ctl_ports[targ_port]->fe_done;
13599
13600 #ifdef CTL_TIME_IO
13601         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13602                 char str[256];
13603                 char path_str[64];
13604                 struct sbuf sb;
13605
13606                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
13607                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13608
13609                 sbuf_cat(&sb, path_str);
13610                 switch (io->io_hdr.io_type) {
13611                 case CTL_IO_SCSI:
13612                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13613                         sbuf_printf(&sb, "\n");
13614                         sbuf_cat(&sb, path_str);
13615                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13616                                     io->scsiio.tag_num, io->scsiio.tag_type);
13617                         break;
13618                 case CTL_IO_TASK:
13619                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13620                                     "Tag Type: %d\n", io->taskio.task_action,
13621                                     io->taskio.tag_num, io->taskio.tag_type);
13622                         break;
13623                 default:
13624                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13625                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13626                         break;
13627                 }
13628                 sbuf_cat(&sb, path_str);
13629                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13630                             (intmax_t)time_uptime - io->io_hdr.start_time);
13631                 sbuf_finish(&sb);
13632                 printf("%s", sbuf_data(&sb));
13633         }
13634 #endif /* CTL_TIME_IO */
13635
13636         switch (io->io_hdr.io_type) {
13637         case CTL_IO_SCSI:
13638                 break;
13639         case CTL_IO_TASK:
13640                 if (bootverbose || verbose > 0)
13641                         ctl_io_error_print(io, NULL);
13642                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13643                         ctl_free_io(io);
13644                 else
13645                         fe_done(io);
13646                 return (CTL_RETVAL_COMPLETE);
13647                 break;
13648         default:
13649                 printf("ctl_process_done: invalid io type %d\n",
13650                        io->io_hdr.io_type);
13651                 panic("ctl_process_done: invalid io type %d\n",
13652                       io->io_hdr.io_type);
13653                 break; /* NOTREACHED */
13654         }
13655
13656         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13657         if (lun == NULL) {
13658                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13659                                  io->io_hdr.nexus.targ_mapped_lun));
13660                 fe_done(io);
13661                 goto bailout;
13662         }
13663         ctl_softc = lun->ctl_softc;
13664
13665         mtx_lock(&lun->lun_lock);
13666
13667         /*
13668          * Check to see if we have any errors to inject here.  We only
13669          * inject errors for commands that don't already have errors set.
13670          */
13671         if ((STAILQ_FIRST(&lun->error_list) != NULL)
13672          && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13673                 ctl_inject_error(lun, io);
13674
13675         /*
13676          * XXX KDM how do we treat commands that aren't completed
13677          * successfully?
13678          *
13679          * XXX KDM should we also track I/O latency?
13680          */
13681         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13682             io->io_hdr.io_type == CTL_IO_SCSI) {
13683 #ifdef CTL_TIME_IO
13684                 struct bintime cur_bt;
13685 #endif
13686                 int type;
13687
13688                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13689                     CTL_FLAG_DATA_IN)
13690                         type = CTL_STATS_READ;
13691                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13692                     CTL_FLAG_DATA_OUT)
13693                         type = CTL_STATS_WRITE;
13694                 else
13695                         type = CTL_STATS_NO_IO;
13696
13697                 lun->stats.ports[targ_port].bytes[type] +=
13698                     io->scsiio.kern_total_len;
13699                 lun->stats.ports[targ_port].operations[type]++;
13700 #ifdef CTL_TIME_IO
13701                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13702                    &io->io_hdr.dma_bt);
13703                 lun->stats.ports[targ_port].num_dmas[type] +=
13704                     io->io_hdr.num_dmas;
13705                 getbintime(&cur_bt);
13706                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13707                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13708 #endif
13709         }
13710
13711         /*
13712          * Remove this from the OOA queue.
13713          */
13714         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13715
13716         /*
13717          * Run through the blocked queue on this LUN and see if anything
13718          * has become unblocked, now that this transaction is done.
13719          */
13720         ctl_check_blocked(lun);
13721
13722         /*
13723          * If the LUN has been invalidated, free it if there is nothing
13724          * left on its OOA queue.
13725          */
13726         if ((lun->flags & CTL_LUN_INVALID)
13727          && TAILQ_EMPTY(&lun->ooa_queue)) {
13728                 mtx_unlock(&lun->lun_lock);
13729                 mtx_lock(&ctl_softc->ctl_lock);
13730                 ctl_free_lun(lun);
13731                 mtx_unlock(&ctl_softc->ctl_lock);
13732         } else
13733                 mtx_unlock(&lun->lun_lock);
13734
13735         /*
13736          * If this command has been aborted, make sure we set the status
13737          * properly.  The FETD is responsible for freeing the I/O and doing
13738          * whatever it needs to do to clean up its state.
13739          */
13740         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13741                 ctl_set_task_aborted(&io->scsiio);
13742
13743         /*
13744          * We print out status for every task management command.  For SCSI
13745          * commands, we filter out any unit attention errors; they happen
13746          * on every boot, and would clutter up the log.  Note:  task
13747          * management commands aren't printed here, they are printed above,
13748          * since they should never even make it down here.
13749          */
13750         switch (io->io_hdr.io_type) {
13751         case CTL_IO_SCSI: {
13752                 int error_code, sense_key, asc, ascq;
13753
13754                 sense_key = 0;
13755
13756                 if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13757                  && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13758                         /*
13759                          * Since this is just for printing, no need to
13760                          * show errors here.
13761                          */
13762                         scsi_extract_sense_len(&io->scsiio.sense_data,
13763                                                io->scsiio.sense_len,
13764                                                &error_code,
13765                                                &sense_key,
13766                                                &asc,
13767                                                &ascq,
13768                                                /*show_errors*/ 0);
13769                 }
13770
13771                 if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13772                  && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13773                   || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13774                   || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13775
13776                         if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13777                                 ctl_softc->skipped_prints++;
13778                         } else {
13779                                 uint32_t skipped_prints;
13780
13781                                 skipped_prints = ctl_softc->skipped_prints;
13782
13783                                 ctl_softc->skipped_prints = 0;
13784                                 ctl_softc->last_print_jiffies = time_uptime;
13785
13786                                 if (skipped_prints > 0) {
13787 #ifdef NEEDTOPORT
13788                                         csevent_log(CSC_CTL | CSC_SHELF_SW |
13789                                             CTL_ERROR_REPORT,
13790                                             csevent_LogType_Trace,
13791                                             csevent_Severity_Information,
13792                                             csevent_AlertLevel_Green,
13793                                             csevent_FRU_Firmware,
13794                                             csevent_FRU_Unknown,
13795                                             "High CTL error volume, %d prints "
13796                                             "skipped", skipped_prints);
13797 #endif
13798                                 }
13799                                 if (bootverbose || verbose > 0)
13800                                         ctl_io_error_print(io, NULL);
13801                         }
13802                 }
13803                 break;
13804         }
13805         case CTL_IO_TASK:
13806                 if (bootverbose || verbose > 0)
13807                         ctl_io_error_print(io, NULL);
13808                 break;
13809         default:
13810                 break;
13811         }
13812
13813         /*
13814          * Tell the FETD or the other shelf controller we're done with this
13815          * command.  Note that only SCSI commands get to this point.  Task
13816          * management commands are completed above.
13817          *
13818          * We only send status to the other controller if we're in XFER
13819          * mode.  In SER_ONLY mode, the I/O is done on the controller that
13820          * received the I/O (from CTL's perspective), and so the status is
13821          * generated there.
13822          *
13823          * XXX KDM if we hold the lock here, we could cause a deadlock
13824          * if the frontend comes back in in this context to queue
13825          * something.
13826          */
13827         if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13828          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13829                 union ctl_ha_msg msg;
13830
13831                 memset(&msg, 0, sizeof(msg));
13832                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13833                 msg.hdr.original_sc = io->io_hdr.original_sc;
13834                 msg.hdr.nexus = io->io_hdr.nexus;
13835                 msg.hdr.status = io->io_hdr.status;
13836                 msg.scsi.scsi_status = io->scsiio.scsi_status;
13837                 msg.scsi.tag_num = io->scsiio.tag_num;
13838                 msg.scsi.tag_type = io->scsiio.tag_type;
13839                 msg.scsi.sense_len = io->scsiio.sense_len;
13840                 msg.scsi.sense_residual = io->scsiio.sense_residual;
13841                 msg.scsi.residual = io->scsiio.residual;
13842                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13843                        sizeof(io->scsiio.sense_data));
13844                 /*
13845                  * We copy this whether or not this is an I/O-related
13846                  * command.  Otherwise, we'd have to go and check to see
13847                  * whether it's a read/write command, and it really isn't
13848                  * worth it.
13849                  */
13850                 memcpy(&msg.scsi.lbalen,
13851                        &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13852                        sizeof(msg.scsi.lbalen));
13853
13854                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13855                                 sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13856                         /* XXX do something here */
13857                 }
13858
13859                 ctl_free_io(io);
13860         } else
13861                 fe_done(io);
13862
13863 bailout:
13864
13865         return (CTL_RETVAL_COMPLETE);
13866 }
13867
13868 #ifdef CTL_WITH_CA
13869 /*
13870  * Front end should call this if it doesn't do autosense.  When the request
13871  * sense comes back in from the initiator, we'll dequeue this and send it.
13872  */
13873 int
13874 ctl_queue_sense(union ctl_io *io)
13875 {
13876         struct ctl_lun *lun;
13877         struct ctl_softc *ctl_softc;
13878         uint32_t initidx, targ_lun;
13879
13880         ctl_softc = control_softc;
13881
13882         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13883
13884         /*
13885          * LUN lookup will likely move to the ctl_work_thread() once we
13886          * have our new queueing infrastructure (that doesn't put things on
13887          * a per-LUN queue initially).  That is so that we can handle
13888          * things like an INQUIRY to a LUN that we don't have enabled.  We
13889          * can't deal with that right now.
13890          */
13891         mtx_lock(&ctl_softc->ctl_lock);
13892
13893         /*
13894          * If we don't have a LUN for this, just toss the sense
13895          * information.
13896          */
13897         targ_lun = io->io_hdr.nexus.targ_lun;
13898         targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13899         if ((targ_lun < CTL_MAX_LUNS)
13900          && (ctl_softc->ctl_luns[targ_lun] != NULL))
13901                 lun = ctl_softc->ctl_luns[targ_lun];
13902         else
13903                 goto bailout;
13904
13905         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13906
13907         mtx_lock(&lun->lun_lock);
13908         /*
13909          * Already have CA set for this LUN...toss the sense information.
13910          */
13911         if (ctl_is_set(lun->have_ca, initidx)) {
13912                 mtx_unlock(&lun->lun_lock);
13913                 goto bailout;
13914         }
13915
13916         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13917                ctl_min(sizeof(lun->pending_sense[initidx]),
13918                sizeof(io->scsiio.sense_data)));
13919         ctl_set_mask(lun->have_ca, initidx);
13920         mtx_unlock(&lun->lun_lock);
13921
13922 bailout:
13923         mtx_unlock(&ctl_softc->ctl_lock);
13924
13925         ctl_free_io(io);
13926
13927         return (CTL_RETVAL_COMPLETE);
13928 }
13929 #endif
13930
13931 /*
13932  * Primary command inlet from frontend ports.  All SCSI and task I/O
13933  * requests must go through this function.
13934  */
13935 int
13936 ctl_queue(union ctl_io *io)
13937 {
13938         struct ctl_softc *ctl_softc;
13939
13940         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13941
13942         ctl_softc = control_softc;
13943
13944 #ifdef CTL_TIME_IO
13945         io->io_hdr.start_time = time_uptime;
13946         getbintime(&io->io_hdr.start_bt);
13947 #endif /* CTL_TIME_IO */
13948
13949         /* Map FE-specific LUN ID into global one. */
13950         io->io_hdr.nexus.targ_mapped_lun =
13951             ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13952
13953         switch (io->io_hdr.io_type) {
13954         case CTL_IO_SCSI:
13955         case CTL_IO_TASK:
13956                 ctl_enqueue_incoming(io);
13957                 break;
13958         default:
13959                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13960                 return (EINVAL);
13961         }
13962
13963         return (CTL_RETVAL_COMPLETE);
13964 }
13965
13966 #ifdef CTL_IO_DELAY
13967 static void
13968 ctl_done_timer_wakeup(void *arg)
13969 {
13970         union ctl_io *io;
13971
13972         io = (union ctl_io *)arg;
13973         ctl_done(io);
13974 }
13975 #endif /* CTL_IO_DELAY */
13976
13977 void
13978 ctl_done(union ctl_io *io)
13979 {
13980         struct ctl_softc *ctl_softc;
13981
13982         ctl_softc = control_softc;
13983
13984         /*
13985          * Enable this to catch duplicate completion issues.
13986          */
13987 #if 0
13988         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13989                 printf("%s: type %d msg %d cdb %x iptl: "
13990                        "%d:%d:%d:%d tag 0x%04x "
13991                        "flag %#x status %x\n",
13992                         __func__,
13993                         io->io_hdr.io_type,
13994                         io->io_hdr.msg_type,
13995                         io->scsiio.cdb[0],
13996                         io->io_hdr.nexus.initid.id,
13997                         io->io_hdr.nexus.targ_port,
13998                         io->io_hdr.nexus.targ_target.id,
13999                         io->io_hdr.nexus.targ_lun,
14000                         (io->io_hdr.io_type ==
14001                         CTL_IO_TASK) ?
14002                         io->taskio.tag_num :
14003                         io->scsiio.tag_num,
14004                         io->io_hdr.flags,
14005                         io->io_hdr.status);
14006         } else
14007                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
14008 #endif
14009
14010         /*
14011          * This is an internal copy of an I/O, and should not go through
14012          * the normal done processing logic.
14013          */
14014         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
14015                 return;
14016
14017         /*
14018          * We need to send a msg to the serializing shelf to finish the IO
14019          * as well.  We don't send a finish message to the other shelf if
14020          * this is a task management command.  Task management commands
14021          * aren't serialized in the OOA queue, but rather just executed on
14022          * both shelf controllers for commands that originated on that
14023          * controller.
14024          */
14025         if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
14026          && (io->io_hdr.io_type != CTL_IO_TASK)) {
14027                 union ctl_ha_msg msg_io;
14028
14029                 msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
14030                 msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
14031                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
14032                     sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
14033                 }
14034                 /* continue on to finish IO */
14035         }
14036 #ifdef CTL_IO_DELAY
14037         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
14038                 struct ctl_lun *lun;
14039
14040                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14041
14042                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
14043         } else {
14044                 struct ctl_lun *lun;
14045
14046                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14047
14048                 if ((lun != NULL)
14049                  && (lun->delay_info.done_delay > 0)) {
14050                         struct callout *callout;
14051
14052                         callout = (struct callout *)&io->io_hdr.timer_bytes;
14053                         callout_init(callout, /*mpsafe*/ 1);
14054                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
14055                         callout_reset(callout,
14056                                       lun->delay_info.done_delay * hz,
14057                                       ctl_done_timer_wakeup, io);
14058                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
14059                                 lun->delay_info.done_delay = 0;
14060                         return;
14061                 }
14062         }
14063 #endif /* CTL_IO_DELAY */
14064
14065         ctl_enqueue_done(io);
14066 }
14067
14068 int
14069 ctl_isc(struct ctl_scsiio *ctsio)
14070 {
14071         struct ctl_lun *lun;
14072         int retval;
14073
14074         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14075
14076         CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
14077
14078         CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
14079
14080         retval = lun->backend->data_submit((union ctl_io *)ctsio);
14081
14082         return (retval);
14083 }
14084
14085
14086 static void
14087 ctl_work_thread(void *arg)
14088 {
14089         struct ctl_thread *thr = (struct ctl_thread *)arg;
14090         struct ctl_softc *softc = thr->ctl_softc;
14091         union ctl_io *io;
14092         int retval;
14093
14094         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
14095
14096         for (;;) {
14097                 retval = 0;
14098
14099                 /*
14100                  * We handle the queues in this order:
14101                  * - ISC
14102                  * - done queue (to free up resources, unblock other commands)
14103                  * - RtR queue
14104                  * - incoming queue
14105                  *
14106                  * If those queues are empty, we break out of the loop and
14107                  * go to sleep.
14108                  */
14109                 mtx_lock(&thr->queue_lock);
14110                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
14111                 if (io != NULL) {
14112                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
14113                         mtx_unlock(&thr->queue_lock);
14114                         ctl_handle_isc(io);
14115                         continue;
14116                 }
14117                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
14118                 if (io != NULL) {
14119                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
14120                         /* clear any blocked commands, call fe_done */
14121                         mtx_unlock(&thr->queue_lock);
14122                         retval = ctl_process_done(io);
14123                         continue;
14124                 }
14125                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
14126                 if (io != NULL) {
14127                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
14128                         mtx_unlock(&thr->queue_lock);
14129                         if (io->io_hdr.io_type == CTL_IO_TASK)
14130                                 ctl_run_task(io);
14131                         else
14132                                 ctl_scsiio_precheck(softc, &io->scsiio);
14133                         continue;
14134                 }
14135                 if (!ctl_pause_rtr) {
14136                         io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
14137                         if (io != NULL) {
14138                                 STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
14139                                 mtx_unlock(&thr->queue_lock);
14140                                 retval = ctl_scsiio(&io->scsiio);
14141                                 if (retval != CTL_RETVAL_COMPLETE)
14142                                         CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
14143                                 continue;
14144                         }
14145                 }
14146
14147                 /* Sleep until we have something to do. */
14148                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
14149         }
14150 }
14151
14152 static void
14153 ctl_lun_thread(void *arg)
14154 {
14155         struct ctl_softc *softc = (struct ctl_softc *)arg;
14156         struct ctl_be_lun *be_lun;
14157         int retval;
14158
14159         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14160
14161         for (;;) {
14162                 retval = 0;
14163                 mtx_lock(&softc->ctl_lock);
14164                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14165                 if (be_lun != NULL) {
14166                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14167                         mtx_unlock(&softc->ctl_lock);
14168                         ctl_create_lun(be_lun);
14169                         continue;
14170                 }
14171
14172                 /* Sleep until we have something to do. */
14173                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14174                     PDROP | PRIBIO, "-", 0);
14175         }
14176 }
14177
14178 static void
14179 ctl_enqueue_incoming(union ctl_io *io)
14180 {
14181         struct ctl_softc *softc = control_softc;
14182         struct ctl_thread *thr;
14183         u_int idx;
14184
14185         idx = (io->io_hdr.nexus.targ_port * 127 +
14186                io->io_hdr.nexus.initid.id) % worker_threads;
14187         thr = &softc->threads[idx];
14188         mtx_lock(&thr->queue_lock);
14189         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14190         mtx_unlock(&thr->queue_lock);
14191         wakeup(thr);
14192 }
14193
14194 static void
14195 ctl_enqueue_rtr(union ctl_io *io)
14196 {
14197         struct ctl_softc *softc = control_softc;
14198         struct ctl_thread *thr;
14199
14200         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14201         mtx_lock(&thr->queue_lock);
14202         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14203         mtx_unlock(&thr->queue_lock);
14204         wakeup(thr);
14205 }
14206
14207 static void
14208 ctl_enqueue_done(union ctl_io *io)
14209 {
14210         struct ctl_softc *softc = control_softc;
14211         struct ctl_thread *thr;
14212
14213         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14214         mtx_lock(&thr->queue_lock);
14215         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14216         mtx_unlock(&thr->queue_lock);
14217         wakeup(thr);
14218 }
14219
14220 static void
14221 ctl_enqueue_isc(union ctl_io *io)
14222 {
14223         struct ctl_softc *softc = control_softc;
14224         struct ctl_thread *thr;
14225
14226         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14227         mtx_lock(&thr->queue_lock);
14228         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14229         mtx_unlock(&thr->queue_lock);
14230         wakeup(thr);
14231 }
14232
14233 /* Initialization and failover */
14234
14235 void
14236 ctl_init_isc_msg(void)
14237 {
14238         printf("CTL: Still calling this thing\n");
14239 }
14240
14241 /*
14242  * Init component
14243  *      Initializes component into configuration defined by bootMode
14244  *      (see hasc-sv.c)
14245  *      returns hasc_Status:
14246  *              OK
14247  *              ERROR - fatal error
14248  */
14249 static ctl_ha_comp_status
14250 ctl_isc_init(struct ctl_ha_component *c)
14251 {
14252         ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14253
14254         c->status = ret;
14255         return ret;
14256 }
14257
14258 /* Start component
14259  *      Starts component in state requested. If component starts successfully,
14260  *      it must set its own state to the requestrd state
14261  *      When requested state is HASC_STATE_HA, the component may refine it
14262  *      by adding _SLAVE or _MASTER flags.
14263  *      Currently allowed state transitions are:
14264  *      UNKNOWN->HA             - initial startup
14265  *      UNKNOWN->SINGLE - initial startup when no parter detected
14266  *      HA->SINGLE              - failover
14267  * returns ctl_ha_comp_status:
14268  *              OK      - component successfully started in requested state
14269  *              FAILED  - could not start the requested state, failover may
14270  *                        be possible
14271  *              ERROR   - fatal error detected, no future startup possible
14272  */
14273 static ctl_ha_comp_status
14274 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14275 {
14276         ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14277
14278         printf("%s: go\n", __func__);
14279
14280         // UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14281         if (c->state == CTL_HA_STATE_UNKNOWN ) {
14282                 ctl_is_single = 0;
14283                 if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14284                     != CTL_HA_STATUS_SUCCESS) {
14285                         printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14286                         ret = CTL_HA_COMP_STATUS_ERROR;
14287                 }
14288         } else if (CTL_HA_STATE_IS_HA(c->state)
14289                 && CTL_HA_STATE_IS_SINGLE(state)){
14290                 // HA->SINGLE transition
14291                 ctl_failover();
14292                 ctl_is_single = 1;
14293         } else {
14294                 printf("ctl_isc_start:Invalid state transition %X->%X\n",
14295                        c->state, state);
14296                 ret = CTL_HA_COMP_STATUS_ERROR;
14297         }
14298         if (CTL_HA_STATE_IS_SINGLE(state))
14299                 ctl_is_single = 1;
14300
14301         c->state = state;
14302         c->status = ret;
14303         return ret;
14304 }
14305
14306 /*
14307  * Quiesce component
14308  * The component must clear any error conditions (set status to OK) and
14309  * prepare itself to another Start call
14310  * returns ctl_ha_comp_status:
14311  *      OK
14312  *      ERROR
14313  */
14314 static ctl_ha_comp_status
14315 ctl_isc_quiesce(struct ctl_ha_component *c)
14316 {
14317         int ret = CTL_HA_COMP_STATUS_OK;
14318
14319         ctl_pause_rtr = 1;
14320         c->status = ret;
14321         return ret;
14322 }
14323
14324 struct ctl_ha_component ctl_ha_component_ctlisc =
14325 {
14326         .name = "CTL ISC",
14327         .state = CTL_HA_STATE_UNKNOWN,
14328         .init = ctl_isc_init,
14329         .start = ctl_isc_start,
14330         .quiesce = ctl_isc_quiesce
14331 };
14332
14333 /*
14334  *  vim: ts=8
14335  */