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*/SCP_SWP,
 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->pr_keys[idx] == 0)
2996                                                 continue;
2997                                         printf("  LUN %d port %d iid %d key "
2998                                                "%#jx\n", i, j, k,
2999                                                (uintmax_t)lun->pr_keys[idx]);
3000                                 }
3001                         }
3002                 }
3003                 printf("CTL Persistent Reservation information end\n");
3004                 printf("CTL Ports:\n");
3005                 STAILQ_FOREACH(port, &softc->port_list, links) {
3006                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3007                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3008                                port->frontend->name, port->port_type,
3009                                port->physical_port, port->virtual_port,
3010                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3011                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3012                                 if (port->wwpn_iid[j].in_use == 0 &&
3013                                     port->wwpn_iid[j].wwpn == 0 &&
3014                                     port->wwpn_iid[j].name == NULL)
3015                                         continue;
3016
3017                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
3018                                     j, port->wwpn_iid[j].in_use,
3019                                     (uintmax_t)port->wwpn_iid[j].wwpn,
3020                                     port->wwpn_iid[j].name);
3021                         }
3022                 }
3023                 printf("CTL Port information end\n");
3024                 mtx_unlock(&softc->ctl_lock);
3025                 /*
3026                  * XXX KDM calling this without a lock.  We'd likely want
3027                  * to drop the lock before calling the frontend's dump
3028                  * routine anyway.
3029                  */
3030                 printf("CTL Frontends:\n");
3031                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
3032                         printf("  Frontend '%s'\n", fe->name);
3033                         if (fe->fe_dump != NULL)
3034                                 fe->fe_dump();
3035                 }
3036                 printf("CTL Frontend information end\n");
3037                 break;
3038         }
3039         case CTL_LUN_REQ: {
3040                 struct ctl_lun_req *lun_req;
3041                 struct ctl_backend_driver *backend;
3042
3043                 lun_req = (struct ctl_lun_req *)addr;
3044
3045                 backend = ctl_backend_find(lun_req->backend);
3046                 if (backend == NULL) {
3047                         lun_req->status = CTL_LUN_ERROR;
3048                         snprintf(lun_req->error_str,
3049                                  sizeof(lun_req->error_str),
3050                                  "Backend \"%s\" not found.",
3051                                  lun_req->backend);
3052                         break;
3053                 }
3054                 if (lun_req->num_be_args > 0) {
3055                         lun_req->kern_be_args = ctl_copyin_args(
3056                                 lun_req->num_be_args,
3057                                 lun_req->be_args,
3058                                 lun_req->error_str,
3059                                 sizeof(lun_req->error_str));
3060                         if (lun_req->kern_be_args == NULL) {
3061                                 lun_req->status = CTL_LUN_ERROR;
3062                                 break;
3063                         }
3064                 }
3065
3066                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3067
3068                 if (lun_req->num_be_args > 0) {
3069                         ctl_copyout_args(lun_req->num_be_args,
3070                                       lun_req->kern_be_args);
3071                         ctl_free_args(lun_req->num_be_args,
3072                                       lun_req->kern_be_args);
3073                 }
3074                 break;
3075         }
3076         case CTL_LUN_LIST: {
3077                 struct sbuf *sb;
3078                 struct ctl_lun *lun;
3079                 struct ctl_lun_list *list;
3080                 struct ctl_option *opt;
3081
3082                 list = (struct ctl_lun_list *)addr;
3083
3084                 /*
3085                  * Allocate a fixed length sbuf here, based on the length
3086                  * of the user's buffer.  We could allocate an auto-extending
3087                  * buffer, and then tell the user how much larger our
3088                  * amount of data is than his buffer, but that presents
3089                  * some problems:
3090                  *
3091                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3092                  *     we can't hold a lock while calling them with an
3093                  *     auto-extending buffer.
3094                  *
3095                  * 2.  There is not currently a LUN reference counting
3096                  *     mechanism, outside of outstanding transactions on
3097                  *     the LUN's OOA queue.  So a LUN could go away on us
3098                  *     while we're getting the LUN number, backend-specific
3099                  *     information, etc.  Thus, given the way things
3100                  *     currently work, we need to hold the CTL lock while
3101                  *     grabbing LUN information.
3102                  *
3103                  * So, from the user's standpoint, the best thing to do is
3104                  * allocate what he thinks is a reasonable buffer length,
3105                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3106                  * double the buffer length and try again.  (And repeat
3107                  * that until he succeeds.)
3108                  */
3109                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3110                 if (sb == NULL) {
3111                         list->status = CTL_LUN_LIST_ERROR;
3112                         snprintf(list->error_str, sizeof(list->error_str),
3113                                  "Unable to allocate %d bytes for LUN list",
3114                                  list->alloc_len);
3115                         break;
3116                 }
3117
3118                 sbuf_printf(sb, "<ctllunlist>\n");
3119
3120                 mtx_lock(&softc->ctl_lock);
3121                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3122                         mtx_lock(&lun->lun_lock);
3123                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3124                                              (uintmax_t)lun->lun);
3125
3126                         /*
3127                          * Bail out as soon as we see that we've overfilled
3128                          * the buffer.
3129                          */
3130                         if (retval != 0)
3131                                 break;
3132
3133                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3134                                              "</backend_type>\n",
3135                                              (lun->backend == NULL) ?  "none" :
3136                                              lun->backend->name);
3137
3138                         if (retval != 0)
3139                                 break;
3140
3141                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3142                                              lun->be_lun->lun_type);
3143
3144                         if (retval != 0)
3145                                 break;
3146
3147                         if (lun->backend == NULL) {
3148                                 retval = sbuf_printf(sb, "</lun>\n");
3149                                 if (retval != 0)
3150                                         break;
3151                                 continue;
3152                         }
3153
3154                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3155                                              (lun->be_lun->maxlba > 0) ?
3156                                              lun->be_lun->maxlba + 1 : 0);
3157
3158                         if (retval != 0)
3159                                 break;
3160
3161                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3162                                              lun->be_lun->blocksize);
3163
3164                         if (retval != 0)
3165                                 break;
3166
3167                         retval = sbuf_printf(sb, "\t<serial_number>");
3168
3169                         if (retval != 0)
3170                                 break;
3171
3172                         retval = ctl_sbuf_printf_esc(sb,
3173                                                      lun->be_lun->serial_num);
3174
3175                         if (retval != 0)
3176                                 break;
3177
3178                         retval = sbuf_printf(sb, "</serial_number>\n");
3179
3180                         if (retval != 0)
3181                                 break;
3182
3183                         retval = sbuf_printf(sb, "\t<device_id>");
3184
3185                         if (retval != 0)
3186                                 break;
3187
3188                         retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3189
3190                         if (retval != 0)
3191                                 break;
3192
3193                         retval = sbuf_printf(sb, "</device_id>\n");
3194
3195                         if (retval != 0)
3196                                 break;
3197
3198                         if (lun->backend->lun_info != NULL) {
3199                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3200                                 if (retval != 0)
3201                                         break;
3202                         }
3203                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3204                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3205                                     opt->name, opt->value, opt->name);
3206                                 if (retval != 0)
3207                                         break;
3208                         }
3209
3210                         retval = sbuf_printf(sb, "</lun>\n");
3211
3212                         if (retval != 0)
3213                                 break;
3214                         mtx_unlock(&lun->lun_lock);
3215                 }
3216                 if (lun != NULL)
3217                         mtx_unlock(&lun->lun_lock);
3218                 mtx_unlock(&softc->ctl_lock);
3219
3220                 if ((retval != 0)
3221                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3222                         retval = 0;
3223                         sbuf_delete(sb);
3224                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3225                         snprintf(list->error_str, sizeof(list->error_str),
3226                                  "Out of space, %d bytes is too small",
3227                                  list->alloc_len);
3228                         break;
3229                 }
3230
3231                 sbuf_finish(sb);
3232
3233                 retval = copyout(sbuf_data(sb), list->lun_xml,
3234                                  sbuf_len(sb) + 1);
3235
3236                 list->fill_len = sbuf_len(sb) + 1;
3237                 list->status = CTL_LUN_LIST_OK;
3238                 sbuf_delete(sb);
3239                 break;
3240         }
3241         case CTL_ISCSI: {
3242                 struct ctl_iscsi *ci;
3243                 struct ctl_frontend *fe;
3244
3245                 ci = (struct ctl_iscsi *)addr;
3246
3247                 fe = ctl_frontend_find("iscsi");
3248                 if (fe == NULL) {
3249                         ci->status = CTL_ISCSI_ERROR;
3250                         snprintf(ci->error_str, sizeof(ci->error_str),
3251                             "Frontend \"iscsi\" not found.");
3252                         break;
3253                 }
3254
3255                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3256                 break;
3257         }
3258         case CTL_PORT_REQ: {
3259                 struct ctl_req *req;
3260                 struct ctl_frontend *fe;
3261
3262                 req = (struct ctl_req *)addr;
3263
3264                 fe = ctl_frontend_find(req->driver);
3265                 if (fe == NULL) {
3266                         req->status = CTL_LUN_ERROR;
3267                         snprintf(req->error_str, sizeof(req->error_str),
3268                             "Frontend \"%s\" not found.", req->driver);
3269                         break;
3270                 }
3271                 if (req->num_args > 0) {
3272                         req->kern_args = ctl_copyin_args(req->num_args,
3273                             req->args, req->error_str, sizeof(req->error_str));
3274                         if (req->kern_args == NULL) {
3275                                 req->status = CTL_LUN_ERROR;
3276                                 break;
3277                         }
3278                 }
3279
3280                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3281
3282                 if (req->num_args > 0) {
3283                         ctl_copyout_args(req->num_args, req->kern_args);
3284                         ctl_free_args(req->num_args, req->kern_args);
3285                 }
3286                 break;
3287         }
3288         case CTL_PORT_LIST: {
3289                 struct sbuf *sb;
3290                 struct ctl_port *port;
3291                 struct ctl_lun_list *list;
3292                 struct ctl_option *opt;
3293
3294                 list = (struct ctl_lun_list *)addr;
3295
3296                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3297                 if (sb == NULL) {
3298                         list->status = CTL_LUN_LIST_ERROR;
3299                         snprintf(list->error_str, sizeof(list->error_str),
3300                                  "Unable to allocate %d bytes for LUN list",
3301                                  list->alloc_len);
3302                         break;
3303                 }
3304
3305                 sbuf_printf(sb, "<ctlportlist>\n");
3306
3307                 mtx_lock(&softc->ctl_lock);
3308                 STAILQ_FOREACH(port, &softc->port_list, links) {
3309                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3310                                              (uintmax_t)port->targ_port);
3311
3312                         /*
3313                          * Bail out as soon as we see that we've overfilled
3314                          * the buffer.
3315                          */
3316                         if (retval != 0)
3317                                 break;
3318
3319                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3320                             "</frontend_type>\n", port->frontend->name);
3321                         if (retval != 0)
3322                                 break;
3323
3324                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3325                                              port->port_type);
3326                         if (retval != 0)
3327                                 break;
3328
3329                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3330                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3331                         if (retval != 0)
3332                                 break;
3333
3334                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3335                             port->port_name);
3336                         if (retval != 0)
3337                                 break;
3338
3339                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3340                             port->physical_port);
3341                         if (retval != 0)
3342                                 break;
3343
3344                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3345                             port->virtual_port);
3346                         if (retval != 0)
3347                                 break;
3348
3349                         retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3350                             (uintmax_t)port->wwnn);
3351                         if (retval != 0)
3352                                 break;
3353
3354                         retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3355                             (uintmax_t)port->wwpn);
3356                         if (retval != 0)
3357                                 break;
3358
3359                         if (port->port_info != NULL) {
3360                                 retval = port->port_info(port->onoff_arg, sb);
3361                                 if (retval != 0)
3362                                         break;
3363                         }
3364                         STAILQ_FOREACH(opt, &port->options, links) {
3365                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3366                                     opt->name, opt->value, opt->name);
3367                                 if (retval != 0)
3368                                         break;
3369                         }
3370
3371                         retval = sbuf_printf(sb, "</targ_port>\n");
3372                         if (retval != 0)
3373                                 break;
3374                 }
3375                 mtx_unlock(&softc->ctl_lock);
3376
3377                 if ((retval != 0)
3378                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3379                         retval = 0;
3380                         sbuf_delete(sb);
3381                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3382                         snprintf(list->error_str, sizeof(list->error_str),
3383                                  "Out of space, %d bytes is too small",
3384                                  list->alloc_len);
3385                         break;
3386                 }
3387
3388                 sbuf_finish(sb);
3389
3390                 retval = copyout(sbuf_data(sb), list->lun_xml,
3391                                  sbuf_len(sb) + 1);
3392
3393                 list->fill_len = sbuf_len(sb) + 1;
3394                 list->status = CTL_LUN_LIST_OK;
3395                 sbuf_delete(sb);
3396                 break;
3397         }
3398         default: {
3399                 /* XXX KDM should we fix this? */
3400 #if 0
3401                 struct ctl_backend_driver *backend;
3402                 unsigned int type;
3403                 int found;
3404
3405                 found = 0;
3406
3407                 /*
3408                  * We encode the backend type as the ioctl type for backend
3409                  * ioctls.  So parse it out here, and then search for a
3410                  * backend of this type.
3411                  */
3412                 type = _IOC_TYPE(cmd);
3413
3414                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3415                         if (backend->type == type) {
3416                                 found = 1;
3417                                 break;
3418                         }
3419                 }
3420                 if (found == 0) {
3421                         printf("ctl: unknown ioctl command %#lx or backend "
3422                                "%d\n", cmd, type);
3423                         retval = EINVAL;
3424                         break;
3425                 }
3426                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3427 #endif
3428                 retval = ENOTTY;
3429                 break;
3430         }
3431         }
3432         return (retval);
3433 }
3434
3435 uint32_t
3436 ctl_get_initindex(struct ctl_nexus *nexus)
3437 {
3438         if (nexus->targ_port < CTL_MAX_PORTS)
3439                 return (nexus->initid.id +
3440                         (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3441         else
3442                 return (nexus->initid.id +
3443                        ((nexus->targ_port - CTL_MAX_PORTS) *
3444                         CTL_MAX_INIT_PER_PORT));
3445 }
3446
3447 uint32_t
3448 ctl_get_resindex(struct ctl_nexus *nexus)
3449 {
3450         return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3451 }
3452
3453 uint32_t
3454 ctl_port_idx(int port_num)
3455 {
3456         if (port_num < CTL_MAX_PORTS)
3457                 return(port_num);
3458         else
3459                 return(port_num - CTL_MAX_PORTS);
3460 }
3461
3462 static uint32_t
3463 ctl_map_lun(int port_num, uint32_t lun_id)
3464 {
3465         struct ctl_port *port;
3466
3467         port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3468         if (port == NULL)
3469                 return (UINT32_MAX);
3470         if (port->lun_map == NULL)
3471                 return (lun_id);
3472         return (port->lun_map(port->targ_lun_arg, lun_id));
3473 }
3474
3475 static uint32_t
3476 ctl_map_lun_back(int port_num, uint32_t lun_id)
3477 {
3478         struct ctl_port *port;
3479         uint32_t i;
3480
3481         port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3482         if (port->lun_map == NULL)
3483                 return (lun_id);
3484         for (i = 0; i < CTL_MAX_LUNS; i++) {
3485                 if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3486                         return (i);
3487         }
3488         return (UINT32_MAX);
3489 }
3490
3491 /*
3492  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3493  * that are a power of 2.
3494  */
3495 int
3496 ctl_ffz(uint32_t *mask, uint32_t size)
3497 {
3498         uint32_t num_chunks, num_pieces;
3499         int i, j;
3500
3501         num_chunks = (size >> 5);
3502         if (num_chunks == 0)
3503                 num_chunks++;
3504         num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3505
3506         for (i = 0; i < num_chunks; i++) {
3507                 for (j = 0; j < num_pieces; j++) {
3508                         if ((mask[i] & (1 << j)) == 0)
3509                                 return ((i << 5) + j);
3510                 }
3511         }
3512
3513         return (-1);
3514 }
3515
3516 int
3517 ctl_set_mask(uint32_t *mask, uint32_t bit)
3518 {
3519         uint32_t chunk, piece;
3520
3521         chunk = bit >> 5;
3522         piece = bit % (sizeof(uint32_t) * 8);
3523
3524         if ((mask[chunk] & (1 << piece)) != 0)
3525                 return (-1);
3526         else
3527                 mask[chunk] |= (1 << piece);
3528
3529         return (0);
3530 }
3531
3532 int
3533 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3534 {
3535         uint32_t chunk, piece;
3536
3537         chunk = bit >> 5;
3538         piece = bit % (sizeof(uint32_t) * 8);
3539
3540         if ((mask[chunk] & (1 << piece)) == 0)
3541                 return (-1);
3542         else
3543                 mask[chunk] &= ~(1 << piece);
3544
3545         return (0);
3546 }
3547
3548 int
3549 ctl_is_set(uint32_t *mask, uint32_t bit)
3550 {
3551         uint32_t chunk, piece;
3552
3553         chunk = bit >> 5;
3554         piece = bit % (sizeof(uint32_t) * 8);
3555
3556         if ((mask[chunk] & (1 << piece)) == 0)
3557                 return (0);
3558         else
3559                 return (1);
3560 }
3561
3562 #ifdef unused
3563 /*
3564  * The bus, target and lun are optional, they can be filled in later.
3565  * can_wait is used to determine whether we can wait on the malloc or not.
3566  */
3567 union ctl_io*
3568 ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3569               uint32_t targ_lun, int can_wait)
3570 {
3571         union ctl_io *io;
3572
3573         if (can_wait)
3574                 io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3575         else
3576                 io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3577
3578         if (io != NULL) {
3579                 io->io_hdr.io_type = io_type;
3580                 io->io_hdr.targ_port = targ_port;
3581                 /*
3582                  * XXX KDM this needs to change/go away.  We need to move
3583                  * to a preallocated pool of ctl_scsiio structures.
3584                  */
3585                 io->io_hdr.nexus.targ_target.id = targ_target;
3586                 io->io_hdr.nexus.targ_lun = targ_lun;
3587         }
3588
3589         return (io);
3590 }
3591
3592 void
3593 ctl_kfree_io(union ctl_io *io)
3594 {
3595         free(io, M_CTL);
3596 }
3597 #endif /* unused */
3598
3599 /*
3600  * ctl_softc, pool_type, total_ctl_io are passed in.
3601  * npool is passed out.
3602  */
3603 int
3604 ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3605                 uint32_t total_ctl_io, struct ctl_io_pool **npool)
3606 {
3607         uint32_t i;
3608         union ctl_io *cur_io, *next_io;
3609         struct ctl_io_pool *pool;
3610         int retval;
3611
3612         retval = 0;
3613
3614         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3615                                             M_NOWAIT | M_ZERO);
3616         if (pool == NULL) {
3617                 retval = ENOMEM;
3618                 goto bailout;
3619         }
3620
3621         pool->type = pool_type;
3622         pool->ctl_softc = ctl_softc;
3623
3624         mtx_lock(&ctl_softc->pool_lock);
3625         pool->id = ctl_softc->cur_pool_id++;
3626         mtx_unlock(&ctl_softc->pool_lock);
3627
3628         pool->flags = CTL_POOL_FLAG_NONE;
3629         pool->refcount = 1;             /* Reference for validity. */
3630         STAILQ_INIT(&pool->free_queue);
3631
3632         /*
3633          * XXX KDM other options here:
3634          * - allocate a page at a time
3635          * - allocate one big chunk of memory.
3636          * Page allocation might work well, but would take a little more
3637          * tracking.
3638          */
3639         for (i = 0; i < total_ctl_io; i++) {
3640                 cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3641                                                 M_NOWAIT);
3642                 if (cur_io == NULL) {
3643                         retval = ENOMEM;
3644                         break;
3645                 }
3646                 cur_io->io_hdr.pool = pool;
3647                 STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3648                 pool->total_ctl_io++;
3649                 pool->free_ctl_io++;
3650         }
3651
3652         if (retval != 0) {
3653                 for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3654                      cur_io != NULL; cur_io = next_io) {
3655                         next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3656                                                               links);
3657                         STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3658                                       ctl_io_hdr, links);
3659                         free(cur_io, M_CTLIO);
3660                 }
3661
3662                 free(pool, M_CTL);
3663                 goto bailout;
3664         }
3665         mtx_lock(&ctl_softc->pool_lock);
3666         ctl_softc->num_pools++;
3667         STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3668         /*
3669          * Increment our usage count if this is an external consumer, so we
3670          * can't get unloaded until the external consumer (most likely a
3671          * FETD) unloads and frees his pool.
3672          *
3673          * XXX KDM will this increment the caller's module use count, or
3674          * mine?
3675          */
3676 #if 0
3677         if ((pool_type != CTL_POOL_EMERGENCY)
3678          && (pool_type != CTL_POOL_INTERNAL)
3679          && (pool_type != CTL_POOL_4OTHERSC))
3680                 MOD_INC_USE_COUNT;
3681 #endif
3682
3683         mtx_unlock(&ctl_softc->pool_lock);
3684
3685         *npool = pool;
3686
3687 bailout:
3688
3689         return (retval);
3690 }
3691
3692 static int
3693 ctl_pool_acquire(struct ctl_io_pool *pool)
3694 {
3695
3696         mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3697
3698         if (pool->flags & CTL_POOL_FLAG_INVALID)
3699                 return (EINVAL);
3700
3701         pool->refcount++;
3702
3703         return (0);
3704 }
3705
3706 static void
3707 ctl_pool_release(struct ctl_io_pool *pool)
3708 {
3709         struct ctl_softc *ctl_softc = pool->ctl_softc;
3710         union ctl_io *io;
3711
3712         mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3713
3714         if (--pool->refcount != 0)
3715                 return;
3716
3717         while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3718                 STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3719                               links);
3720                 free(io, M_CTLIO);
3721         }
3722
3723         STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3724         ctl_softc->num_pools--;
3725
3726         /*
3727          * XXX KDM will this decrement the caller's usage count or mine?
3728          */
3729 #if 0
3730         if ((pool->type != CTL_POOL_EMERGENCY)
3731          && (pool->type != CTL_POOL_INTERNAL)
3732          && (pool->type != CTL_POOL_4OTHERSC))
3733                 MOD_DEC_USE_COUNT;
3734 #endif
3735
3736         free(pool, M_CTL);
3737 }
3738
3739 void
3740 ctl_pool_free(struct ctl_io_pool *pool)
3741 {
3742         struct ctl_softc *ctl_softc;
3743
3744         if (pool == NULL)
3745                 return;
3746
3747         ctl_softc = pool->ctl_softc;
3748         mtx_lock(&ctl_softc->pool_lock);
3749         pool->flags |= CTL_POOL_FLAG_INVALID;
3750         ctl_pool_release(pool);
3751         mtx_unlock(&ctl_softc->pool_lock);
3752 }
3753
3754 /*
3755  * This routine does not block (except for spinlocks of course).
3756  * It tries to allocate a ctl_io union from the caller's pool as quickly as
3757  * possible.
3758  */
3759 union ctl_io *
3760 ctl_alloc_io(void *pool_ref)
3761 {
3762         union ctl_io *io;
3763         struct ctl_softc *ctl_softc;
3764         struct ctl_io_pool *pool, *npool;
3765         struct ctl_io_pool *emergency_pool;
3766
3767         pool = (struct ctl_io_pool *)pool_ref;
3768
3769         if (pool == NULL) {
3770                 printf("%s: pool is NULL\n", __func__);
3771                 return (NULL);
3772         }
3773
3774         emergency_pool = NULL;
3775
3776         ctl_softc = pool->ctl_softc;
3777
3778         mtx_lock(&ctl_softc->pool_lock);
3779         /*
3780          * First, try to get the io structure from the user's pool.
3781          */
3782         if (ctl_pool_acquire(pool) == 0) {
3783                 io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3784                 if (io != NULL) {
3785                         STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3786                         pool->total_allocated++;
3787                         pool->free_ctl_io--;
3788                         mtx_unlock(&ctl_softc->pool_lock);
3789                         return (io);
3790                 } else
3791                         ctl_pool_release(pool);
3792         }
3793         /*
3794          * If he doesn't have any io structures left, search for an
3795          * emergency pool and grab one from there.
3796          */
3797         STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3798                 if (npool->type != CTL_POOL_EMERGENCY)
3799                         continue;
3800
3801                 if (ctl_pool_acquire(npool) != 0)
3802                         continue;
3803
3804                 emergency_pool = npool;
3805
3806                 io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3807                 if (io != NULL) {
3808                         STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3809                         npool->total_allocated++;
3810                         npool->free_ctl_io--;
3811                         mtx_unlock(&ctl_softc->pool_lock);
3812                         return (io);
3813                 } else
3814                         ctl_pool_release(npool);
3815         }
3816
3817         /* Drop the spinlock before we malloc */
3818         mtx_unlock(&ctl_softc->pool_lock);
3819
3820         /*
3821          * The emergency pool (if it exists) didn't have one, so try an
3822          * atomic (i.e. nonblocking) malloc and see if we get lucky.
3823          */
3824         io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3825         if (io != NULL) {
3826                 /*
3827                  * If the emergency pool exists but is empty, add this
3828                  * ctl_io to its list when it gets freed.
3829                  */
3830                 if (emergency_pool != NULL) {
3831                         mtx_lock(&ctl_softc->pool_lock);
3832                         if (ctl_pool_acquire(emergency_pool) == 0) {
3833                                 io->io_hdr.pool = emergency_pool;
3834                                 emergency_pool->total_ctl_io++;
3835                                 /*
3836                                  * Need to bump this, otherwise
3837                                  * total_allocated and total_freed won't
3838                                  * match when we no longer have anything
3839                                  * outstanding.
3840                                  */
3841                                 emergency_pool->total_allocated++;
3842                         }
3843                         mtx_unlock(&ctl_softc->pool_lock);
3844                 } else
3845                         io->io_hdr.pool = NULL;
3846         }
3847
3848         return (io);
3849 }
3850
3851 void
3852 ctl_free_io(union ctl_io *io)
3853 {
3854         if (io == NULL)
3855                 return;
3856
3857         /*
3858          * If this ctl_io has a pool, return it to that pool.
3859          */
3860         if (io->io_hdr.pool != NULL) {
3861                 struct ctl_io_pool *pool;
3862
3863                 pool = (struct ctl_io_pool *)io->io_hdr.pool;
3864                 mtx_lock(&pool->ctl_softc->pool_lock);
3865                 io->io_hdr.io_type = 0xff;
3866                 STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3867                 pool->total_freed++;
3868                 pool->free_ctl_io++;
3869                 ctl_pool_release(pool);
3870                 mtx_unlock(&pool->ctl_softc->pool_lock);
3871         } else {
3872                 /*
3873                  * Otherwise, just free it.  We probably malloced it and
3874                  * the emergency pool wasn't available.
3875                  */
3876                 free(io, M_CTLIO);
3877         }
3878
3879 }
3880
3881 void
3882 ctl_zero_io(union ctl_io *io)
3883 {
3884         void *pool_ref;
3885
3886         if (io == NULL)
3887                 return;
3888
3889         /*
3890          * May need to preserve linked list pointers at some point too.
3891          */
3892         pool_ref = io->io_hdr.pool;
3893
3894         memset(io, 0, sizeof(*io));
3895
3896         io->io_hdr.pool = pool_ref;
3897 }
3898
3899 /*
3900  * This routine is currently used for internal copies of ctl_ios that need
3901  * to persist for some reason after we've already returned status to the
3902  * FETD.  (Thus the flag set.)
3903  *
3904  * XXX XXX
3905  * Note that this makes a blind copy of all fields in the ctl_io, except
3906  * for the pool reference.  This includes any memory that has been
3907  * allocated!  That memory will no longer be valid after done has been
3908  * called, so this would be VERY DANGEROUS for command that actually does
3909  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3910  * start and stop commands, which don't transfer any data, so this is not a
3911  * problem.  If it is used for anything else, the caller would also need to
3912  * allocate data buffer space and this routine would need to be modified to
3913  * copy the data buffer(s) as well.
3914  */
3915 void
3916 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3917 {
3918         void *pool_ref;
3919
3920         if ((src == NULL)
3921          || (dest == NULL))
3922                 return;
3923
3924         /*
3925          * May need to preserve linked list pointers at some point too.
3926          */
3927         pool_ref = dest->io_hdr.pool;
3928
3929         memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3930
3931         dest->io_hdr.pool = pool_ref;
3932         /*
3933          * We need to know that this is an internal copy, and doesn't need
3934          * to get passed back to the FETD that allocated it.
3935          */
3936         dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3937 }
3938
3939 #ifdef NEEDTOPORT
3940 static void
3941 ctl_update_power_subpage(struct copan_power_subpage *page)
3942 {
3943         int num_luns, num_partitions, config_type;
3944         struct ctl_softc *softc;
3945         cs_BOOL_t aor_present, shelf_50pct_power;
3946         cs_raidset_personality_t rs_type;
3947         int max_active_luns;
3948
3949         softc = control_softc;
3950
3951         /* subtract out the processor LUN */
3952         num_luns = softc->num_luns - 1;
3953         /*
3954          * Default to 7 LUNs active, which was the only number we allowed
3955          * in the past.
3956          */
3957         max_active_luns = 7;
3958
3959         num_partitions = config_GetRsPartitionInfo();
3960         config_type = config_GetConfigType();
3961         shelf_50pct_power = config_GetShelfPowerMode();
3962         aor_present = config_IsAorRsPresent();
3963
3964         rs_type = ddb_GetRsRaidType(1);
3965         if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3966          && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3967                 EPRINT(0, "Unsupported RS type %d!", rs_type);
3968         }
3969
3970
3971         page->total_luns = num_luns;
3972
3973         switch (config_type) {
3974         case 40:
3975                 /*
3976                  * In a 40 drive configuration, it doesn't matter what DC
3977                  * cards we have, whether we have AOR enabled or not,
3978                  * partitioning or not, or what type of RAIDset we have.
3979                  * In that scenario, we can power up every LUN we present
3980                  * to the user.
3981                  */
3982                 max_active_luns = num_luns;
3983
3984                 break;
3985         case 64:
3986                 if (shelf_50pct_power == CS_FALSE) {
3987                         /* 25% power */
3988                         if (aor_present == CS_TRUE) {
3989                                 if (rs_type ==
3990                                      CS_RAIDSET_PERSONALITY_RAID5) {
3991                                         max_active_luns = 7;
3992                                 } else if (rs_type ==
3993                                          CS_RAIDSET_PERSONALITY_RAID1){
3994                                         max_active_luns = 14;
3995                                 } else {
3996                                         /* XXX KDM now what?? */
3997                                 }
3998                         } else {
3999                                 if (rs_type ==
4000                                      CS_RAIDSET_PERSONALITY_RAID5) {
4001                                         max_active_luns = 8;
4002                                 } else if (rs_type ==
4003                                          CS_RAIDSET_PERSONALITY_RAID1){
4004                                         max_active_luns = 16;
4005                                 } else {
4006                                         /* XXX KDM now what?? */
4007                                 }
4008                         }
4009                 } else {
4010                         /* 50% power */
4011                         /*
4012                          * With 50% power in a 64 drive configuration, we
4013                          * can power all LUNs we present.
4014                          */
4015                         max_active_luns = num_luns;
4016                 }
4017                 break;
4018         case 112:
4019                 if (shelf_50pct_power == CS_FALSE) {
4020                         /* 25% power */
4021                         if (aor_present == CS_TRUE) {
4022                                 if (rs_type ==
4023                                      CS_RAIDSET_PERSONALITY_RAID5) {
4024                                         max_active_luns = 7;
4025                                 } else if (rs_type ==
4026                                          CS_RAIDSET_PERSONALITY_RAID1){
4027                                         max_active_luns = 14;
4028                                 } else {
4029                                         /* XXX KDM now what?? */
4030                                 }
4031                         } else {
4032                                 if (rs_type ==
4033                                      CS_RAIDSET_PERSONALITY_RAID5) {
4034                                         max_active_luns = 8;
4035                                 } else if (rs_type ==
4036                                          CS_RAIDSET_PERSONALITY_RAID1){
4037                                         max_active_luns = 16;
4038                                 } else {
4039                                         /* XXX KDM now what?? */
4040                                 }
4041                         }
4042                 } else {
4043                         /* 50% power */
4044                         if (aor_present == CS_TRUE) {
4045                                 if (rs_type ==
4046                                      CS_RAIDSET_PERSONALITY_RAID5) {
4047                                         max_active_luns = 14;
4048                                 } else if (rs_type ==
4049                                          CS_RAIDSET_PERSONALITY_RAID1){
4050                                         /*
4051                                          * We're assuming here that disk
4052                                          * caching is enabled, and so we're
4053                                          * able to power up half of each
4054                                          * LUN, and cache all writes.
4055                                          */
4056                                         max_active_luns = num_luns;
4057                                 } else {
4058                                         /* XXX KDM now what?? */
4059                                 }
4060                         } else {
4061                                 if (rs_type ==
4062                                      CS_RAIDSET_PERSONALITY_RAID5) {
4063                                         max_active_luns = 15;
4064                                 } else if (rs_type ==
4065                                          CS_RAIDSET_PERSONALITY_RAID1){
4066                                         max_active_luns = 30;
4067                                 } else {
4068                                         /* XXX KDM now what?? */
4069                                 }
4070                         }
4071                 }
4072                 break;
4073         default:
4074                 /*
4075                  * In this case, we have an unknown configuration, so we
4076                  * just use the default from above.
4077                  */
4078                 break;
4079         }
4080
4081         page->max_active_luns = max_active_luns;
4082 #if 0
4083         printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4084                page->total_luns, page->max_active_luns);
4085 #endif
4086 }
4087 #endif /* NEEDTOPORT */
4088
4089 /*
4090  * This routine could be used in the future to load default and/or saved
4091  * mode page parameters for a particuar lun.
4092  */
4093 static int
4094 ctl_init_page_index(struct ctl_lun *lun)
4095 {
4096         int i;
4097         struct ctl_page_index *page_index;
4098         struct ctl_softc *softc;
4099         const char *value;
4100
4101         memcpy(&lun->mode_pages.index, page_index_template,
4102                sizeof(page_index_template));
4103
4104         softc = lun->ctl_softc;
4105
4106         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4107
4108                 page_index = &lun->mode_pages.index[i];
4109                 /*
4110                  * If this is a disk-only mode page, there's no point in
4111                  * setting it up.  For some pages, we have to have some
4112                  * basic information about the disk in order to calculate the
4113                  * mode page data.
4114                  */
4115                 if ((lun->be_lun->lun_type != T_DIRECT)
4116                  && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4117                         continue;
4118
4119                 switch (page_index->page_code & SMPH_PC_MASK) {
4120                 case SMS_FORMAT_DEVICE_PAGE: {
4121                         struct scsi_format_page *format_page;
4122
4123                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4124                                 panic("subpage is incorrect!");
4125
4126                         /*
4127                          * Sectors per track are set above.  Bytes per
4128                          * sector need to be set here on a per-LUN basis.
4129                          */
4130                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4131                                &format_page_default,
4132                                sizeof(format_page_default));
4133                         memcpy(&lun->mode_pages.format_page[
4134                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4135                                sizeof(format_page_changeable));
4136                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4137                                &format_page_default,
4138                                sizeof(format_page_default));
4139                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4140                                &format_page_default,
4141                                sizeof(format_page_default));
4142
4143                         format_page = &lun->mode_pages.format_page[
4144                                 CTL_PAGE_CURRENT];
4145                         scsi_ulto2b(lun->be_lun->blocksize,
4146                                     format_page->bytes_per_sector);
4147
4148                         format_page = &lun->mode_pages.format_page[
4149                                 CTL_PAGE_DEFAULT];
4150                         scsi_ulto2b(lun->be_lun->blocksize,
4151                                     format_page->bytes_per_sector);
4152
4153                         format_page = &lun->mode_pages.format_page[
4154                                 CTL_PAGE_SAVED];
4155                         scsi_ulto2b(lun->be_lun->blocksize,
4156                                     format_page->bytes_per_sector);
4157
4158                         page_index->page_data =
4159                                 (uint8_t *)lun->mode_pages.format_page;
4160                         break;
4161                 }
4162                 case SMS_RIGID_DISK_PAGE: {
4163                         struct scsi_rigid_disk_page *rigid_disk_page;
4164                         uint32_t sectors_per_cylinder;
4165                         uint64_t cylinders;
4166 #ifndef __XSCALE__
4167                         int shift;
4168 #endif /* !__XSCALE__ */
4169
4170                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4171                                 panic("invalid subpage value %d",
4172                                       page_index->subpage);
4173
4174                         /*
4175                          * Rotation rate and sectors per track are set
4176                          * above.  We calculate the cylinders here based on
4177                          * capacity.  Due to the number of heads and
4178                          * sectors per track we're using, smaller arrays
4179                          * may turn out to have 0 cylinders.  Linux and
4180                          * FreeBSD don't pay attention to these mode pages
4181                          * to figure out capacity, but Solaris does.  It
4182                          * seems to deal with 0 cylinders just fine, and
4183                          * works out a fake geometry based on the capacity.
4184                          */
4185                         memcpy(&lun->mode_pages.rigid_disk_page[
4186                                CTL_PAGE_CURRENT], &rigid_disk_page_default,
4187                                sizeof(rigid_disk_page_default));
4188                         memcpy(&lun->mode_pages.rigid_disk_page[
4189                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4190                                sizeof(rigid_disk_page_changeable));
4191                         memcpy(&lun->mode_pages.rigid_disk_page[
4192                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4193                                sizeof(rigid_disk_page_default));
4194                         memcpy(&lun->mode_pages.rigid_disk_page[
4195                                CTL_PAGE_SAVED], &rigid_disk_page_default,
4196                                sizeof(rigid_disk_page_default));
4197
4198                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4199                                 CTL_DEFAULT_HEADS;
4200
4201                         /*
4202                          * The divide method here will be more accurate,
4203                          * probably, but results in floating point being
4204                          * used in the kernel on i386 (__udivdi3()).  On the
4205                          * XScale, though, __udivdi3() is implemented in
4206                          * software.
4207                          *
4208                          * The shift method for cylinder calculation is
4209                          * accurate if sectors_per_cylinder is a power of
4210                          * 2.  Otherwise it might be slightly off -- you
4211                          * might have a bit of a truncation problem.
4212                          */
4213 #ifdef  __XSCALE__
4214                         cylinders = (lun->be_lun->maxlba + 1) /
4215                                 sectors_per_cylinder;
4216 #else
4217                         for (shift = 31; shift > 0; shift--) {
4218                                 if (sectors_per_cylinder & (1 << shift))
4219                                         break;
4220                         }
4221                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4222 #endif
4223
4224                         /*
4225                          * We've basically got 3 bytes, or 24 bits for the
4226                          * cylinder size in the mode page.  If we're over,
4227                          * just round down to 2^24.
4228                          */
4229                         if (cylinders > 0xffffff)
4230                                 cylinders = 0xffffff;
4231
4232                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4233                                 CTL_PAGE_CURRENT];
4234                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4235
4236                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4237                                 CTL_PAGE_DEFAULT];
4238                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4239
4240                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4241                                 CTL_PAGE_SAVED];
4242                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4243
4244                         page_index->page_data =
4245                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4246                         break;
4247                 }
4248                 case SMS_CACHING_PAGE: {
4249                         struct scsi_caching_page *caching_page;
4250
4251                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4252                                 panic("invalid subpage value %d",
4253                                       page_index->subpage);
4254                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4255                                &caching_page_default,
4256                                sizeof(caching_page_default));
4257                         memcpy(&lun->mode_pages.caching_page[
4258                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4259                                sizeof(caching_page_changeable));
4260                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4261                                &caching_page_default,
4262                                sizeof(caching_page_default));
4263                         caching_page = &lun->mode_pages.caching_page[
4264                             CTL_PAGE_SAVED];
4265                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
4266                         if (value != NULL && strcmp(value, "off") == 0)
4267                                 caching_page->flags1 &= ~SCP_WCE;
4268                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4269                         if (value != NULL && strcmp(value, "off") == 0)
4270                                 caching_page->flags1 |= SCP_RCD;
4271                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4272                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4273                                sizeof(caching_page_default));
4274                         page_index->page_data =
4275                                 (uint8_t *)lun->mode_pages.caching_page;
4276                         break;
4277                 }
4278                 case SMS_CONTROL_MODE_PAGE: {
4279                         struct scsi_control_page *control_page;
4280
4281                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4282                                 panic("invalid subpage value %d",
4283                                       page_index->subpage);
4284
4285                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4286                                &control_page_default,
4287                                sizeof(control_page_default));
4288                         memcpy(&lun->mode_pages.control_page[
4289                                CTL_PAGE_CHANGEABLE], &control_page_changeable,
4290                                sizeof(control_page_changeable));
4291                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4292                                &control_page_default,
4293                                sizeof(control_page_default));
4294                         control_page = &lun->mode_pages.control_page[
4295                             CTL_PAGE_SAVED];
4296                         value = ctl_get_opt(&lun->be_lun->options, "reordering");
4297                         if (value != NULL && strcmp(value, "unrestricted") == 0) {
4298                                 control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4299                                 control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4300                         }
4301                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4302                                &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4303                                sizeof(control_page_default));
4304                         page_index->page_data =
4305                                 (uint8_t *)lun->mode_pages.control_page;
4306                         break;
4307
4308                 }
4309                 case SMS_VENDOR_SPECIFIC_PAGE:{
4310                         switch (page_index->subpage) {
4311                         case PWR_SUBPAGE_CODE: {
4312                                 struct copan_power_subpage *current_page,
4313                                                            *saved_page;
4314
4315                                 memcpy(&lun->mode_pages.power_subpage[
4316                                        CTL_PAGE_CURRENT],
4317                                        &power_page_default,
4318                                        sizeof(power_page_default));
4319                                 memcpy(&lun->mode_pages.power_subpage[
4320                                        CTL_PAGE_CHANGEABLE],
4321                                        &power_page_changeable,
4322                                        sizeof(power_page_changeable));
4323                                 memcpy(&lun->mode_pages.power_subpage[
4324                                        CTL_PAGE_DEFAULT],
4325                                        &power_page_default,
4326                                        sizeof(power_page_default));
4327                                 memcpy(&lun->mode_pages.power_subpage[
4328                                        CTL_PAGE_SAVED],
4329                                        &power_page_default,
4330                                        sizeof(power_page_default));
4331                                 page_index->page_data =
4332                                     (uint8_t *)lun->mode_pages.power_subpage;
4333
4334                                 current_page = (struct copan_power_subpage *)
4335                                         (page_index->page_data +
4336                                          (page_index->page_len *
4337                                           CTL_PAGE_CURRENT));
4338                                 saved_page = (struct copan_power_subpage *)
4339                                         (page_index->page_data +
4340                                          (page_index->page_len *
4341                                           CTL_PAGE_SAVED));
4342                                 break;
4343                         }
4344                         case APS_SUBPAGE_CODE: {
4345                                 struct copan_aps_subpage *current_page,
4346                                                          *saved_page;
4347
4348                                 // This gets set multiple times but
4349                                 // it should always be the same. It's
4350                                 // only done during init so who cares.
4351                                 index_to_aps_page = i;
4352
4353                                 memcpy(&lun->mode_pages.aps_subpage[
4354                                        CTL_PAGE_CURRENT],
4355                                        &aps_page_default,
4356                                        sizeof(aps_page_default));
4357                                 memcpy(&lun->mode_pages.aps_subpage[
4358                                        CTL_PAGE_CHANGEABLE],
4359                                        &aps_page_changeable,
4360                                        sizeof(aps_page_changeable));
4361                                 memcpy(&lun->mode_pages.aps_subpage[
4362                                        CTL_PAGE_DEFAULT],
4363                                        &aps_page_default,
4364                                        sizeof(aps_page_default));
4365                                 memcpy(&lun->mode_pages.aps_subpage[
4366                                        CTL_PAGE_SAVED],
4367                                        &aps_page_default,
4368                                        sizeof(aps_page_default));
4369                                 page_index->page_data =
4370                                         (uint8_t *)lun->mode_pages.aps_subpage;
4371
4372                                 current_page = (struct copan_aps_subpage *)
4373                                         (page_index->page_data +
4374                                          (page_index->page_len *
4375                                           CTL_PAGE_CURRENT));
4376                                 saved_page = (struct copan_aps_subpage *)
4377                                         (page_index->page_data +
4378                                          (page_index->page_len *
4379                                           CTL_PAGE_SAVED));
4380                                 break;
4381                         }
4382                         case DBGCNF_SUBPAGE_CODE: {
4383                                 struct copan_debugconf_subpage *current_page,
4384                                                                *saved_page;
4385
4386                                 memcpy(&lun->mode_pages.debugconf_subpage[
4387                                        CTL_PAGE_CURRENT],
4388                                        &debugconf_page_default,
4389                                        sizeof(debugconf_page_default));
4390                                 memcpy(&lun->mode_pages.debugconf_subpage[
4391                                        CTL_PAGE_CHANGEABLE],
4392                                        &debugconf_page_changeable,
4393                                        sizeof(debugconf_page_changeable));
4394                                 memcpy(&lun->mode_pages.debugconf_subpage[
4395                                        CTL_PAGE_DEFAULT],
4396                                        &debugconf_page_default,
4397                                        sizeof(debugconf_page_default));
4398                                 memcpy(&lun->mode_pages.debugconf_subpage[
4399                                        CTL_PAGE_SAVED],
4400                                        &debugconf_page_default,
4401                                        sizeof(debugconf_page_default));
4402                                 page_index->page_data =
4403                                         (uint8_t *)lun->mode_pages.debugconf_subpage;
4404
4405                                 current_page = (struct copan_debugconf_subpage *)
4406                                         (page_index->page_data +
4407                                          (page_index->page_len *
4408                                           CTL_PAGE_CURRENT));
4409                                 saved_page = (struct copan_debugconf_subpage *)
4410                                         (page_index->page_data +
4411                                          (page_index->page_len *
4412                                           CTL_PAGE_SAVED));
4413                                 break;
4414                         }
4415                         default:
4416                                 panic("invalid subpage value %d",
4417                                       page_index->subpage);
4418                                 break;
4419                         }
4420                         break;
4421                 }
4422                 default:
4423                         panic("invalid page value %d",
4424                               page_index->page_code & SMPH_PC_MASK);
4425                         break;
4426         }
4427         }
4428
4429         return (CTL_RETVAL_COMPLETE);
4430 }
4431
4432 /*
4433  * LUN allocation.
4434  *
4435  * Requirements:
4436  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4437  *   wants us to allocate the LUN and he can block.
4438  * - ctl_softc is always set
4439  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4440  *
4441  * Returns 0 for success, non-zero (errno) for failure.
4442  */
4443 static int
4444 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4445               struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4446 {
4447         struct ctl_lun *nlun, *lun;
4448         struct ctl_port *port;
4449         struct scsi_vpd_id_descriptor *desc;
4450         struct scsi_vpd_id_t10 *t10id;
4451         const char *eui, *naa, *scsiname, *vendor, *value;
4452         int lun_number, i, lun_malloced;
4453         int devidlen, idlen1, idlen2 = 0, len;
4454
4455         if (be_lun == NULL)
4456                 return (EINVAL);
4457
4458         /*
4459          * We currently only support Direct Access or Processor LUN types.
4460          */
4461         switch (be_lun->lun_type) {
4462         case T_DIRECT:
4463                 break;
4464         case T_PROCESSOR:
4465                 break;
4466         case T_SEQUENTIAL:
4467         case T_CHANGER:
4468         default:
4469                 be_lun->lun_config_status(be_lun->be_lun,
4470                                           CTL_LUN_CONFIG_FAILURE);
4471                 break;
4472         }
4473         if (ctl_lun == NULL) {
4474                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4475                 lun_malloced = 1;
4476         } else {
4477                 lun_malloced = 0;
4478                 lun = ctl_lun;
4479         }
4480
4481         memset(lun, 0, sizeof(*lun));
4482         if (lun_malloced)
4483                 lun->flags = CTL_LUN_MALLOCED;
4484
4485         /* Generate LUN ID. */
4486         devidlen = max(CTL_DEVID_MIN_LEN,
4487             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4488         idlen1 = sizeof(*t10id) + devidlen;
4489         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4490         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4491         if (scsiname != NULL) {
4492                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4493                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4494         }
4495         eui = ctl_get_opt(&be_lun->options, "eui");
4496         if (eui != NULL) {
4497                 len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4498         }
4499         naa = ctl_get_opt(&be_lun->options, "naa");
4500         if (naa != NULL) {
4501                 len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4502         }
4503         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4504             M_CTL, M_WAITOK | M_ZERO);
4505         lun->lun_devid->len = len;
4506         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4507         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4508         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4509         desc->length = idlen1;
4510         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4511         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4512         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4513                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4514         } else {
4515                 strncpy(t10id->vendor, vendor,
4516                     min(sizeof(t10id->vendor), strlen(vendor)));
4517         }
4518         strncpy((char *)t10id->vendor_spec_id,
4519             (char *)be_lun->device_id, devidlen);
4520         if (scsiname != NULL) {
4521                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4522                     desc->length);
4523                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4524                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4525                     SVPD_ID_TYPE_SCSI_NAME;
4526                 desc->length = idlen2;
4527                 strlcpy(desc->identifier, scsiname, idlen2);
4528         }
4529         if (eui != NULL) {
4530                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4531                     desc->length);
4532                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4533                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4534                     SVPD_ID_TYPE_EUI64;
4535                 desc->length = 8;
4536                 scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4537         }
4538         if (naa != NULL) {
4539                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4540                     desc->length);
4541                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4542                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4543                     SVPD_ID_TYPE_NAA;
4544                 desc->length = 8;
4545                 scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4546         }
4547
4548         mtx_lock(&ctl_softc->ctl_lock);
4549         /*
4550          * See if the caller requested a particular LUN number.  If so, see
4551          * if it is available.  Otherwise, allocate the first available LUN.
4552          */
4553         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4554                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4555                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4556                         mtx_unlock(&ctl_softc->ctl_lock);
4557                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4558                                 printf("ctl: requested LUN ID %d is higher "
4559                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4560                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4561                         } else {
4562                                 /*
4563                                  * XXX KDM return an error, or just assign
4564                                  * another LUN ID in this case??
4565                                  */
4566                                 printf("ctl: requested LUN ID %d is already "
4567                                        "in use\n", be_lun->req_lun_id);
4568                         }
4569                         if (lun->flags & CTL_LUN_MALLOCED)
4570                                 free(lun, M_CTL);
4571                         be_lun->lun_config_status(be_lun->be_lun,
4572                                                   CTL_LUN_CONFIG_FAILURE);
4573                         return (ENOSPC);
4574                 }
4575                 lun_number = be_lun->req_lun_id;
4576         } else {
4577                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4578                 if (lun_number == -1) {
4579                         mtx_unlock(&ctl_softc->ctl_lock);
4580                         printf("ctl: can't allocate LUN on target %ju, out of "
4581                                "LUNs\n", (uintmax_t)target_id.id);
4582                         if (lun->flags & CTL_LUN_MALLOCED)
4583                                 free(lun, M_CTL);
4584                         be_lun->lun_config_status(be_lun->be_lun,
4585                                                   CTL_LUN_CONFIG_FAILURE);
4586                         return (ENOSPC);
4587                 }
4588         }
4589         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4590
4591         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4592         lun->target = target_id;
4593         lun->lun = lun_number;
4594         lun->be_lun = be_lun;
4595         /*
4596          * The processor LUN is always enabled.  Disk LUNs come on line
4597          * disabled, and must be enabled by the backend.
4598          */
4599         lun->flags |= CTL_LUN_DISABLED;
4600         lun->backend = be_lun->be;
4601         be_lun->ctl_lun = lun;
4602         be_lun->lun_id = lun_number;
4603         atomic_add_int(&be_lun->be->num_luns, 1);
4604         if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4605                 lun->flags |= CTL_LUN_STOPPED;
4606
4607         if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4608                 lun->flags |= CTL_LUN_INOPERABLE;
4609
4610         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4611                 lun->flags |= CTL_LUN_PRIMARY_SC;
4612
4613         value = ctl_get_opt(&be_lun->options, "readonly");
4614         if (value != NULL && strcmp(value, "on") == 0)
4615                 lun->flags |= CTL_LUN_READONLY;
4616
4617         lun->ctl_softc = ctl_softc;
4618         TAILQ_INIT(&lun->ooa_queue);
4619         TAILQ_INIT(&lun->blocked_queue);
4620         STAILQ_INIT(&lun->error_list);
4621         ctl_tpc_lun_init(lun);
4622
4623         /*
4624          * Initialize the mode page index.
4625          */
4626         ctl_init_page_index(lun);
4627
4628         /*
4629          * Set the poweron UA for all initiators on this LUN only.
4630          */
4631         for (i = 0; i < CTL_MAX_INITIATORS; i++)
4632                 lun->pending_ua[i] = CTL_UA_POWERON;
4633
4634         /*
4635          * Now, before we insert this lun on the lun list, set the lun
4636          * inventory changed UA for all other luns.
4637          */
4638         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4639                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4640                         nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4641                 }
4642         }
4643
4644         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4645
4646         ctl_softc->ctl_luns[lun_number] = lun;
4647
4648         ctl_softc->num_luns++;
4649
4650         /* Setup statistics gathering */
4651         lun->stats.device_type = be_lun->lun_type;
4652         lun->stats.lun_number = lun_number;
4653         if (lun->stats.device_type == T_DIRECT)
4654                 lun->stats.blocksize = be_lun->blocksize;
4655         else
4656                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4657         for (i = 0;i < CTL_MAX_PORTS;i++)
4658                 lun->stats.ports[i].targ_port = i;
4659
4660         mtx_unlock(&ctl_softc->ctl_lock);
4661
4662         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4663
4664         /*
4665          * Run through each registered FETD and bring it online if it isn't
4666          * already.  Enable the target ID if it hasn't been enabled, and
4667          * enable this particular LUN.
4668          */
4669         STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4670                 int retval;
4671
4672                 retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4673                 if (retval != 0) {
4674                         printf("ctl_alloc_lun: FETD %s port %d returned error "
4675                                "%d for lun_enable on target %ju lun %d\n",
4676                                port->port_name, port->targ_port, retval,
4677                                (uintmax_t)target_id.id, lun_number);
4678                 } else
4679                         port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4680         }
4681         return (0);
4682 }
4683
4684 /*
4685  * Delete a LUN.
4686  * Assumptions:
4687  * - LUN has already been marked invalid and any pending I/O has been taken
4688  *   care of.
4689  */
4690 static int
4691 ctl_free_lun(struct ctl_lun *lun)
4692 {
4693         struct ctl_softc *softc;
4694 #if 0
4695         struct ctl_port *port;
4696 #endif
4697         struct ctl_lun *nlun;
4698         int i;
4699
4700         softc = lun->ctl_softc;
4701
4702         mtx_assert(&softc->ctl_lock, MA_OWNED);
4703
4704         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4705
4706         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4707
4708         softc->ctl_luns[lun->lun] = NULL;
4709
4710         if (!TAILQ_EMPTY(&lun->ooa_queue))
4711                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4712
4713         softc->num_luns--;
4714
4715         /*
4716          * XXX KDM this scheme only works for a single target/multiple LUN
4717          * setup.  It needs to be revamped for a multiple target scheme.
4718          *
4719          * XXX KDM this results in port->lun_disable() getting called twice,
4720          * once when ctl_disable_lun() is called, and a second time here.
4721          * We really need to re-think the LUN disable semantics.  There
4722          * should probably be several steps/levels to LUN removal:
4723          *  - disable
4724          *  - invalidate
4725          *  - free
4726          *
4727          * Right now we only have a disable method when communicating to
4728          * the front end ports, at least for individual LUNs.
4729          */
4730 #if 0
4731         STAILQ_FOREACH(port, &softc->port_list, links) {
4732                 int retval;
4733
4734                 retval = port->lun_disable(port->targ_lun_arg, lun->target,
4735                                          lun->lun);
4736                 if (retval != 0) {
4737                         printf("ctl_free_lun: FETD %s port %d returned error "
4738                                "%d for lun_disable on target %ju lun %jd\n",
4739                                port->port_name, port->targ_port, retval,
4740                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4741                 }
4742
4743                 if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4744                         port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4745
4746                         retval = port->targ_disable(port->targ_lun_arg,lun->target);
4747                         if (retval != 0) {
4748                                 printf("ctl_free_lun: FETD %s port %d "
4749                                        "returned error %d for targ_disable on "
4750                                        "target %ju\n", port->port_name,
4751                                        port->targ_port, retval,
4752                                        (uintmax_t)lun->target.id);
4753                         } else
4754                                 port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4755
4756                         if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4757                                 continue;
4758
4759 #if 0
4760                         port->port_offline(port->onoff_arg);
4761                         port->status &= ~CTL_PORT_STATUS_ONLINE;
4762 #endif
4763                 }
4764         }
4765 #endif
4766
4767         /*
4768          * Tell the backend to free resources, if this LUN has a backend.
4769          */
4770         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4771         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4772
4773         ctl_tpc_lun_shutdown(lun);
4774         mtx_destroy(&lun->lun_lock);
4775         free(lun->lun_devid, M_CTL);
4776         if (lun->flags & CTL_LUN_MALLOCED)
4777                 free(lun, M_CTL);
4778
4779         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4780                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4781                         nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4782                 }
4783         }
4784
4785         return (0);
4786 }
4787
4788 static void
4789 ctl_create_lun(struct ctl_be_lun *be_lun)
4790 {
4791         struct ctl_softc *ctl_softc;
4792
4793         ctl_softc = control_softc;
4794
4795         /*
4796          * ctl_alloc_lun() should handle all potential failure cases.
4797          */
4798         ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4799 }
4800
4801 int
4802 ctl_add_lun(struct ctl_be_lun *be_lun)
4803 {
4804         struct ctl_softc *ctl_softc = control_softc;
4805
4806         mtx_lock(&ctl_softc->ctl_lock);
4807         STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4808         mtx_unlock(&ctl_softc->ctl_lock);
4809         wakeup(&ctl_softc->pending_lun_queue);
4810
4811         return (0);
4812 }
4813
4814 int
4815 ctl_enable_lun(struct ctl_be_lun *be_lun)
4816 {
4817         struct ctl_softc *ctl_softc;
4818         struct ctl_port *port, *nport;
4819         struct ctl_lun *lun;
4820         int retval;
4821
4822         ctl_softc = control_softc;
4823
4824         lun = (struct ctl_lun *)be_lun->ctl_lun;
4825
4826         mtx_lock(&ctl_softc->ctl_lock);
4827         mtx_lock(&lun->lun_lock);
4828         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4829                 /*
4830                  * eh?  Why did we get called if the LUN is already
4831                  * enabled?
4832                  */
4833                 mtx_unlock(&lun->lun_lock);
4834                 mtx_unlock(&ctl_softc->ctl_lock);
4835                 return (0);
4836         }
4837         lun->flags &= ~CTL_LUN_DISABLED;
4838         mtx_unlock(&lun->lun_lock);
4839
4840         for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4841                 nport = STAILQ_NEXT(port, links);
4842
4843                 /*
4844                  * Drop the lock while we call the FETD's enable routine.
4845                  * This can lead to a callback into CTL (at least in the
4846                  * case of the internal initiator frontend.
4847                  */
4848                 mtx_unlock(&ctl_softc->ctl_lock);
4849                 retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4850                 mtx_lock(&ctl_softc->ctl_lock);
4851                 if (retval != 0) {
4852                         printf("%s: FETD %s port %d returned error "
4853                                "%d for lun_enable on target %ju lun %jd\n",
4854                                __func__, port->port_name, port->targ_port, retval,
4855                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4856                 }
4857 #if 0
4858                  else {
4859             /* NOTE:  TODO:  why does lun enable affect port status? */
4860                         port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4861                 }
4862 #endif
4863         }
4864
4865         mtx_unlock(&ctl_softc->ctl_lock);
4866
4867         return (0);
4868 }
4869
4870 int
4871 ctl_disable_lun(struct ctl_be_lun *be_lun)
4872 {
4873         struct ctl_softc *ctl_softc;
4874         struct ctl_port *port;
4875         struct ctl_lun *lun;
4876         int retval;
4877
4878         ctl_softc = control_softc;
4879
4880         lun = (struct ctl_lun *)be_lun->ctl_lun;
4881
4882         mtx_lock(&ctl_softc->ctl_lock);
4883         mtx_lock(&lun->lun_lock);
4884         if (lun->flags & CTL_LUN_DISABLED) {
4885                 mtx_unlock(&lun->lun_lock);
4886                 mtx_unlock(&ctl_softc->ctl_lock);
4887                 return (0);
4888         }
4889         lun->flags |= CTL_LUN_DISABLED;
4890         mtx_unlock(&lun->lun_lock);
4891
4892         STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4893                 mtx_unlock(&ctl_softc->ctl_lock);
4894                 /*
4895                  * Drop the lock before we call the frontend's disable
4896                  * routine, to avoid lock order reversals.
4897                  *
4898                  * XXX KDM what happens if the frontend list changes while
4899                  * we're traversing it?  It's unlikely, but should be handled.
4900                  */
4901                 retval = port->lun_disable(port->targ_lun_arg, lun->target,
4902                                          lun->lun);
4903                 mtx_lock(&ctl_softc->ctl_lock);
4904                 if (retval != 0) {
4905                         printf("ctl_alloc_lun: FETD %s port %d returned error "
4906                                "%d for lun_disable on target %ju lun %jd\n",
4907                                port->port_name, port->targ_port, retval,
4908                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4909                 }
4910         }
4911
4912         mtx_unlock(&ctl_softc->ctl_lock);
4913
4914         return (0);
4915 }
4916
4917 int
4918 ctl_start_lun(struct ctl_be_lun *be_lun)
4919 {
4920         struct ctl_softc *ctl_softc;
4921         struct ctl_lun *lun;
4922
4923         ctl_softc = control_softc;
4924
4925         lun = (struct ctl_lun *)be_lun->ctl_lun;
4926
4927         mtx_lock(&lun->lun_lock);
4928         lun->flags &= ~CTL_LUN_STOPPED;
4929         mtx_unlock(&lun->lun_lock);
4930
4931         return (0);
4932 }
4933
4934 int
4935 ctl_stop_lun(struct ctl_be_lun *be_lun)
4936 {
4937         struct ctl_softc *ctl_softc;
4938         struct ctl_lun *lun;
4939
4940         ctl_softc = control_softc;
4941
4942         lun = (struct ctl_lun *)be_lun->ctl_lun;
4943
4944         mtx_lock(&lun->lun_lock);
4945         lun->flags |= CTL_LUN_STOPPED;
4946         mtx_unlock(&lun->lun_lock);
4947
4948         return (0);
4949 }
4950
4951 int
4952 ctl_lun_offline(struct ctl_be_lun *be_lun)
4953 {
4954         struct ctl_softc *ctl_softc;
4955         struct ctl_lun *lun;
4956
4957         ctl_softc = control_softc;
4958
4959         lun = (struct ctl_lun *)be_lun->ctl_lun;
4960
4961         mtx_lock(&lun->lun_lock);
4962         lun->flags |= CTL_LUN_OFFLINE;
4963         mtx_unlock(&lun->lun_lock);
4964
4965         return (0);
4966 }
4967
4968 int
4969 ctl_lun_online(struct ctl_be_lun *be_lun)
4970 {
4971         struct ctl_softc *ctl_softc;
4972         struct ctl_lun *lun;
4973
4974         ctl_softc = control_softc;
4975
4976         lun = (struct ctl_lun *)be_lun->ctl_lun;
4977
4978         mtx_lock(&lun->lun_lock);
4979         lun->flags &= ~CTL_LUN_OFFLINE;
4980         mtx_unlock(&lun->lun_lock);
4981
4982         return (0);
4983 }
4984
4985 int
4986 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4987 {
4988         struct ctl_softc *ctl_softc;
4989         struct ctl_lun *lun;
4990
4991         ctl_softc = control_softc;
4992
4993         lun = (struct ctl_lun *)be_lun->ctl_lun;
4994
4995         mtx_lock(&lun->lun_lock);
4996
4997         /*
4998          * The LUN needs to be disabled before it can be marked invalid.
4999          */
5000         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
5001                 mtx_unlock(&lun->lun_lock);
5002                 return (-1);
5003         }
5004         /*
5005          * Mark the LUN invalid.
5006          */
5007         lun->flags |= CTL_LUN_INVALID;
5008
5009         /*
5010          * If there is nothing in the OOA queue, go ahead and free the LUN.
5011          * If we have something in the OOA queue, we'll free it when the
5012          * last I/O completes.
5013          */
5014         if (TAILQ_EMPTY(&lun->ooa_queue)) {
5015                 mtx_unlock(&lun->lun_lock);
5016                 mtx_lock(&ctl_softc->ctl_lock);
5017                 ctl_free_lun(lun);
5018                 mtx_unlock(&ctl_softc->ctl_lock);
5019         } else
5020                 mtx_unlock(&lun->lun_lock);
5021
5022         return (0);
5023 }
5024
5025 int
5026 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5027 {
5028         struct ctl_softc *ctl_softc;
5029         struct ctl_lun *lun;
5030
5031         ctl_softc = control_softc;
5032         lun = (struct ctl_lun *)be_lun->ctl_lun;
5033
5034         mtx_lock(&lun->lun_lock);
5035         lun->flags |= CTL_LUN_INOPERABLE;
5036         mtx_unlock(&lun->lun_lock);
5037
5038         return (0);
5039 }
5040
5041 int
5042 ctl_lun_operable(struct ctl_be_lun *be_lun)
5043 {
5044         struct ctl_softc *ctl_softc;
5045         struct ctl_lun *lun;
5046
5047         ctl_softc = control_softc;
5048         lun = (struct ctl_lun *)be_lun->ctl_lun;
5049
5050         mtx_lock(&lun->lun_lock);
5051         lun->flags &= ~CTL_LUN_INOPERABLE;
5052         mtx_unlock(&lun->lun_lock);
5053
5054         return (0);
5055 }
5056
5057 int
5058 ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5059                    int lock)
5060 {
5061         struct ctl_softc *softc;
5062         struct ctl_lun *lun;
5063         struct copan_aps_subpage *current_sp;
5064         struct ctl_page_index *page_index;
5065         int i;
5066
5067         softc = control_softc;
5068
5069         mtx_lock(&softc->ctl_lock);
5070
5071         lun = (struct ctl_lun *)be_lun->ctl_lun;
5072         mtx_lock(&lun->lun_lock);
5073
5074         page_index = NULL;
5075         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5076                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5077                      APS_PAGE_CODE)
5078                         continue;
5079
5080                 if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5081                         continue;
5082                 page_index = &lun->mode_pages.index[i];
5083         }
5084
5085         if (page_index == NULL) {
5086                 mtx_unlock(&lun->lun_lock);
5087                 mtx_unlock(&softc->ctl_lock);
5088                 printf("%s: APS subpage not found for lun %ju!\n", __func__,
5089                        (uintmax_t)lun->lun);
5090                 return (1);
5091         }
5092 #if 0
5093         if ((softc->aps_locked_lun != 0)
5094          && (softc->aps_locked_lun != lun->lun)) {
5095                 printf("%s: attempt to lock LUN %llu when %llu is already "
5096                        "locked\n");
5097                 mtx_unlock(&lun->lun_lock);
5098                 mtx_unlock(&softc->ctl_lock);
5099                 return (1);
5100         }
5101 #endif
5102
5103         current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5104                 (page_index->page_len * CTL_PAGE_CURRENT));
5105
5106         if (lock != 0) {
5107                 current_sp->lock_active = APS_LOCK_ACTIVE;
5108                 softc->aps_locked_lun = lun->lun;
5109         } else {
5110                 current_sp->lock_active = 0;
5111                 softc->aps_locked_lun = 0;
5112         }
5113
5114
5115         /*
5116          * If we're in HA mode, try to send the lock message to the other
5117          * side.
5118          */
5119         if (ctl_is_single == 0) {
5120                 int isc_retval;
5121                 union ctl_ha_msg lock_msg;
5122
5123                 lock_msg.hdr.nexus = *nexus;
5124                 lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5125                 if (lock != 0)
5126                         lock_msg.aps.lock_flag = 1;
5127                 else
5128                         lock_msg.aps.lock_flag = 0;
5129                 isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5130                                          sizeof(lock_msg), 0);
5131                 if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5132                         printf("%s: APS (lock=%d) error returned from "
5133                                "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5134                         mtx_unlock(&lun->lun_lock);
5135                         mtx_unlock(&softc->ctl_lock);
5136                         return (1);
5137                 }
5138         }
5139
5140         mtx_unlock(&lun->lun_lock);
5141         mtx_unlock(&softc->ctl_lock);
5142
5143         return (0);
5144 }
5145
5146 void
5147 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5148 {
5149         struct ctl_lun *lun;
5150         struct ctl_softc *softc;
5151         int i;
5152
5153         softc = control_softc;
5154
5155         lun = (struct ctl_lun *)be_lun->ctl_lun;
5156
5157         mtx_lock(&lun->lun_lock);
5158
5159         for (i = 0; i < CTL_MAX_INITIATORS; i++)
5160                 lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5161
5162         mtx_unlock(&lun->lun_lock);
5163 }
5164
5165 /*
5166  * Backend "memory move is complete" callback for requests that never
5167  * make it down to say RAIDCore's configuration code.
5168  */
5169 int
5170 ctl_config_move_done(union ctl_io *io)
5171 {
5172         int retval;
5173
5174         retval = CTL_RETVAL_COMPLETE;
5175
5176
5177         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5178         /*
5179          * XXX KDM this shouldn't happen, but what if it does?
5180          */
5181         if (io->io_hdr.io_type != CTL_IO_SCSI)
5182                 panic("I/O type isn't CTL_IO_SCSI!");
5183
5184         if ((io->io_hdr.port_status == 0)
5185          && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5186          && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5187                 io->io_hdr.status = CTL_SUCCESS;
5188         else if ((io->io_hdr.port_status != 0)
5189               && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5190               && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5191                 /*
5192                  * For hardware error sense keys, the sense key
5193                  * specific value is defined to be a retry count,
5194                  * but we use it to pass back an internal FETD
5195                  * error code.  XXX KDM  Hopefully the FETD is only
5196                  * using 16 bits for an error code, since that's
5197                  * all the space we have in the sks field.
5198                  */
5199                 ctl_set_internal_failure(&io->scsiio,
5200                                          /*sks_valid*/ 1,
5201                                          /*retry_count*/
5202                                          io->io_hdr.port_status);
5203                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5204                         free(io->scsiio.kern_data_ptr, M_CTL);
5205                 ctl_done(io);
5206                 goto bailout;
5207         }
5208
5209         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5210          || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5211          || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5212                 /*
5213                  * XXX KDM just assuming a single pointer here, and not a
5214                  * S/G list.  If we start using S/G lists for config data,
5215                  * we'll need to know how to clean them up here as well.
5216                  */
5217                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5218                         free(io->scsiio.kern_data_ptr, M_CTL);
5219                 /* Hopefully the user has already set the status... */
5220                 ctl_done(io);
5221         } else {
5222                 /*
5223                  * XXX KDM now we need to continue data movement.  Some
5224                  * options:
5225                  * - call ctl_scsiio() again?  We don't do this for data
5226                  *   writes, because for those at least we know ahead of
5227                  *   time where the write will go and how long it is.  For
5228                  *   config writes, though, that information is largely
5229                  *   contained within the write itself, thus we need to
5230                  *   parse out the data again.
5231                  *
5232                  * - Call some other function once the data is in?
5233                  */
5234
5235                 /*
5236                  * XXX KDM call ctl_scsiio() again for now, and check flag
5237                  * bits to see whether we're allocated or not.
5238                  */
5239                 retval = ctl_scsiio(&io->scsiio);
5240         }
5241 bailout:
5242         return (retval);
5243 }
5244
5245 /*
5246  * This gets called by a backend driver when it is done with a
5247  * data_submit method.
5248  */
5249 void
5250 ctl_data_submit_done(union ctl_io *io)
5251 {
5252         /*
5253          * If the IO_CONT flag is set, we need to call the supplied
5254          * function to continue processing the I/O, instead of completing
5255          * the I/O just yet.
5256          *
5257          * If there is an error, though, we don't want to keep processing.
5258          * Instead, just send status back to the initiator.
5259          */
5260         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5261             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5262             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5263              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5264                 io->scsiio.io_cont(io);
5265                 return;
5266         }
5267         ctl_done(io);
5268 }
5269
5270 /*
5271  * This gets called by a backend driver when it is done with a
5272  * configuration write.
5273  */
5274 void
5275 ctl_config_write_done(union ctl_io *io)
5276 {
5277         uint8_t *buf;
5278
5279         /*
5280          * If the IO_CONT flag is set, we need to call the supplied
5281          * function to continue processing the I/O, instead of completing
5282          * the I/O just yet.
5283          *
5284          * If there is an error, though, we don't want to keep processing.
5285          * Instead, just send status back to the initiator.
5286          */
5287         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5288             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5289             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5290              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5291                 io->scsiio.io_cont(io);
5292                 return;
5293         }
5294         /*
5295          * Since a configuration write can be done for commands that actually
5296          * have data allocated, like write buffer, and commands that have
5297          * no data, like start/stop unit, we need to check here.
5298          */
5299         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5300                 buf = io->scsiio.kern_data_ptr;
5301         else
5302                 buf = NULL;
5303         ctl_done(io);
5304         if (buf)
5305                 free(buf, M_CTL);
5306 }
5307
5308 /*
5309  * SCSI release command.
5310  */
5311 int
5312 ctl_scsi_release(struct ctl_scsiio *ctsio)
5313 {
5314         int length, longid, thirdparty_id, resv_id;
5315         struct ctl_softc *ctl_softc;
5316         struct ctl_lun *lun;
5317         uint32_t residx;
5318
5319         length = 0;
5320         resv_id = 0;
5321
5322         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5323
5324         residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5325         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5326         ctl_softc = control_softc;
5327
5328         switch (ctsio->cdb[0]) {
5329         case RELEASE_10: {
5330                 struct scsi_release_10 *cdb;
5331
5332                 cdb = (struct scsi_release_10 *)ctsio->cdb;
5333
5334                 if (cdb->byte2 & SR10_LONGID)
5335                         longid = 1;
5336                 else
5337                         thirdparty_id = cdb->thirdparty_id;
5338
5339                 resv_id = cdb->resv_id;
5340                 length = scsi_2btoul(cdb->length);
5341                 break;
5342         }
5343         }
5344
5345
5346         /*
5347          * XXX KDM right now, we only support LUN reservation.  We don't
5348          * support 3rd party reservations, or extent reservations, which
5349          * might actually need the parameter list.  If we've gotten this
5350          * far, we've got a LUN reservation.  Anything else got kicked out
5351          * above.  So, according to SPC, ignore the length.
5352          */
5353         length = 0;
5354
5355         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5356          && (length > 0)) {
5357                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5358                 ctsio->kern_data_len = length;
5359                 ctsio->kern_total_len = length;
5360                 ctsio->kern_data_resid = 0;
5361                 ctsio->kern_rel_offset = 0;
5362                 ctsio->kern_sg_entries = 0;
5363                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5364                 ctsio->be_move_done = ctl_config_move_done;
5365                 ctl_datamove((union ctl_io *)ctsio);
5366
5367                 return (CTL_RETVAL_COMPLETE);
5368         }
5369
5370         if (length > 0)
5371                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5372
5373         mtx_lock(&lun->lun_lock);
5374
5375         /*
5376          * According to SPC, it is not an error for an intiator to attempt
5377          * to release a reservation on a LUN that isn't reserved, or that
5378          * is reserved by another initiator.  The reservation can only be
5379          * released, though, by the initiator who made it or by one of
5380          * several reset type events.
5381          */
5382         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5383                         lun->flags &= ~CTL_LUN_RESERVED;
5384
5385         mtx_unlock(&lun->lun_lock);
5386
5387         ctsio->scsi_status = SCSI_STATUS_OK;
5388         ctsio->io_hdr.status = CTL_SUCCESS;
5389
5390         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5391                 free(ctsio->kern_data_ptr, M_CTL);
5392                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5393         }
5394
5395         ctl_done((union ctl_io *)ctsio);
5396         return (CTL_RETVAL_COMPLETE);
5397 }
5398
5399 int
5400 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5401 {
5402         int extent, thirdparty, longid;
5403         int resv_id, length;
5404         uint64_t thirdparty_id;
5405         struct ctl_softc *ctl_softc;
5406         struct ctl_lun *lun;
5407         uint32_t residx;
5408
5409         extent = 0;
5410         thirdparty = 0;
5411         longid = 0;
5412         resv_id = 0;
5413         length = 0;
5414         thirdparty_id = 0;
5415
5416         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5417
5418         residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5419         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5420         ctl_softc = control_softc;
5421
5422         switch (ctsio->cdb[0]) {
5423         case RESERVE_10: {
5424                 struct scsi_reserve_10 *cdb;
5425
5426                 cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5427
5428                 if (cdb->byte2 & SR10_LONGID)
5429                         longid = 1;
5430                 else
5431                         thirdparty_id = cdb->thirdparty_id;
5432
5433                 resv_id = cdb->resv_id;
5434                 length = scsi_2btoul(cdb->length);
5435                 break;
5436         }
5437         }
5438
5439         /*
5440          * XXX KDM right now, we only support LUN reservation.  We don't
5441          * support 3rd party reservations, or extent reservations, which
5442          * might actually need the parameter list.  If we've gotten this
5443          * far, we've got a LUN reservation.  Anything else got kicked out
5444          * above.  So, according to SPC, ignore the length.
5445          */
5446         length = 0;
5447
5448         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5449          && (length > 0)) {
5450                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5451                 ctsio->kern_data_len = length;
5452                 ctsio->kern_total_len = length;
5453                 ctsio->kern_data_resid = 0;
5454                 ctsio->kern_rel_offset = 0;
5455                 ctsio->kern_sg_entries = 0;
5456                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5457                 ctsio->be_move_done = ctl_config_move_done;
5458                 ctl_datamove((union ctl_io *)ctsio);
5459
5460                 return (CTL_RETVAL_COMPLETE);
5461         }
5462
5463         if (length > 0)
5464                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5465
5466         mtx_lock(&lun->lun_lock);
5467         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5468                 ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5469                 ctsio->io_hdr.status = CTL_SCSI_ERROR;
5470                 goto bailout;
5471         }
5472
5473         lun->flags |= CTL_LUN_RESERVED;
5474         lun->res_idx = residx;
5475
5476         ctsio->scsi_status = SCSI_STATUS_OK;
5477         ctsio->io_hdr.status = CTL_SUCCESS;
5478
5479 bailout:
5480         mtx_unlock(&lun->lun_lock);
5481
5482         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5483                 free(ctsio->kern_data_ptr, M_CTL);
5484                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5485         }
5486
5487         ctl_done((union ctl_io *)ctsio);
5488         return (CTL_RETVAL_COMPLETE);
5489 }
5490
5491 int
5492 ctl_start_stop(struct ctl_scsiio *ctsio)
5493 {
5494         struct scsi_start_stop_unit *cdb;
5495         struct ctl_lun *lun;
5496         struct ctl_softc *ctl_softc;
5497         int retval;
5498
5499         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5500
5501         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5502         ctl_softc = control_softc;
5503         retval = 0;
5504
5505         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5506
5507         /*
5508          * XXX KDM
5509          * We don't support the immediate bit on a stop unit.  In order to
5510          * do that, we would need to code up a way to know that a stop is
5511          * pending, and hold off any new commands until it completes, one
5512          * way or another.  Then we could accept or reject those commands
5513          * depending on its status.  We would almost need to do the reverse
5514          * of what we do below for an immediate start -- return the copy of
5515          * the ctl_io to the FETD with status to send to the host (and to
5516          * free the copy!) and then free the original I/O once the stop
5517          * actually completes.  That way, the OOA queue mechanism can work
5518          * to block commands that shouldn't proceed.  Another alternative
5519          * would be to put the copy in the queue in place of the original,
5520          * and return the original back to the caller.  That could be
5521          * slightly safer..
5522          */
5523         if ((cdb->byte2 & SSS_IMMED)
5524          && ((cdb->how & SSS_START) == 0)) {
5525                 ctl_set_invalid_field(ctsio,
5526                                       /*sks_valid*/ 1,
5527                                       /*command*/ 1,
5528                                       /*field*/ 1,
5529                                       /*bit_valid*/ 1,
5530                                       /*bit*/ 0);
5531                 ctl_done((union ctl_io *)ctsio);
5532                 return (CTL_RETVAL_COMPLETE);
5533         }
5534
5535         if ((lun->flags & CTL_LUN_PR_RESERVED)
5536          && ((cdb->how & SSS_START)==0)) {
5537                 uint32_t residx;
5538
5539                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5540                 if (lun->pr_keys[residx] == 0
5541                  || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5542
5543                         ctl_set_reservation_conflict(ctsio);
5544                         ctl_done((union ctl_io *)ctsio);
5545                         return (CTL_RETVAL_COMPLETE);
5546                 }
5547         }
5548
5549         /*
5550          * If there is no backend on this device, we can't start or stop
5551          * it.  In theory we shouldn't get any start/stop commands in the
5552          * first place at this level if the LUN doesn't have a backend.
5553          * That should get stopped by the command decode code.
5554          */
5555         if (lun->backend == NULL) {
5556                 ctl_set_invalid_opcode(ctsio);
5557                 ctl_done((union ctl_io *)ctsio);
5558                 return (CTL_RETVAL_COMPLETE);
5559         }
5560
5561         /*
5562          * XXX KDM Copan-specific offline behavior.
5563          * Figure out a reasonable way to port this?
5564          */
5565 #ifdef NEEDTOPORT
5566         mtx_lock(&lun->lun_lock);
5567
5568         if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5569          && (lun->flags & CTL_LUN_OFFLINE)) {
5570                 /*
5571                  * If the LUN is offline, and the on/offline bit isn't set,
5572                  * reject the start or stop.  Otherwise, let it through.
5573                  */
5574                 mtx_unlock(&lun->lun_lock);
5575                 ctl_set_lun_not_ready(ctsio);
5576                 ctl_done((union ctl_io *)ctsio);
5577         } else {
5578                 mtx_unlock(&lun->lun_lock);
5579 #endif /* NEEDTOPORT */
5580                 /*
5581                  * This could be a start or a stop when we're online,
5582                  * or a stop/offline or start/online.  A start or stop when
5583                  * we're offline is covered in the case above.
5584                  */
5585                 /*
5586                  * In the non-immediate case, we send the request to
5587                  * the backend and return status to the user when
5588                  * it is done.
5589                  *
5590                  * In the immediate case, we allocate a new ctl_io
5591                  * to hold a copy of the request, and send that to
5592                  * the backend.  We then set good status on the
5593                  * user's request and return it immediately.
5594                  */
5595                 if (cdb->byte2 & SSS_IMMED) {
5596                         union ctl_io *new_io;
5597
5598                         new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5599                         if (new_io == NULL) {
5600                                 ctl_set_busy(ctsio);
5601                                 ctl_done((union ctl_io *)ctsio);
5602                         } else {
5603                                 ctl_copy_io((union ctl_io *)ctsio,
5604                                             new_io);
5605                                 retval = lun->backend->config_write(new_io);
5606                                 ctl_set_success(ctsio);
5607                                 ctl_done((union ctl_io *)ctsio);
5608                         }
5609                 } else {
5610                         retval = lun->backend->config_write(
5611                                 (union ctl_io *)ctsio);
5612                 }
5613 #ifdef NEEDTOPORT
5614         }
5615 #endif
5616         return (retval);
5617 }
5618
5619 /*
5620  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5621  * we don't really do anything with the LBA and length fields if the user
5622  * passes them in.  Instead we'll just flush out the cache for the entire
5623  * LUN.
5624  */
5625 int
5626 ctl_sync_cache(struct ctl_scsiio *ctsio)
5627 {
5628         struct ctl_lun *lun;
5629         struct ctl_softc *ctl_softc;
5630         uint64_t starting_lba;
5631         uint32_t block_count;
5632         int retval;
5633
5634         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5635
5636         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5637         ctl_softc = control_softc;
5638         retval = 0;
5639
5640         switch (ctsio->cdb[0]) {
5641         case SYNCHRONIZE_CACHE: {
5642                 struct scsi_sync_cache *cdb;
5643                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5644
5645                 starting_lba = scsi_4btoul(cdb->begin_lba);
5646                 block_count = scsi_2btoul(cdb->lb_count);
5647                 break;
5648         }
5649         case SYNCHRONIZE_CACHE_16: {
5650                 struct scsi_sync_cache_16 *cdb;
5651                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5652
5653                 starting_lba = scsi_8btou64(cdb->begin_lba);
5654                 block_count = scsi_4btoul(cdb->lb_count);
5655                 break;
5656         }
5657         default:
5658                 ctl_set_invalid_opcode(ctsio);
5659                 ctl_done((union ctl_io *)ctsio);
5660                 goto bailout;
5661                 break; /* NOTREACHED */
5662         }
5663
5664         /*
5665          * We check the LBA and length, but don't do anything with them.
5666          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5667          * get flushed.  This check will just help satisfy anyone who wants
5668          * to see an error for an out of range LBA.
5669          */
5670         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5671                 ctl_set_lba_out_of_range(ctsio);
5672                 ctl_done((union ctl_io *)ctsio);
5673                 goto bailout;
5674         }
5675
5676         /*
5677          * If this LUN has no backend, we can't flush the cache anyway.
5678          */
5679         if (lun->backend == NULL) {
5680                 ctl_set_invalid_opcode(ctsio);
5681                 ctl_done((union ctl_io *)ctsio);
5682                 goto bailout;
5683         }
5684
5685         /*
5686          * Check to see whether we're configured to send the SYNCHRONIZE
5687          * CACHE command directly to the back end.
5688          */
5689         mtx_lock(&lun->lun_lock);
5690         if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5691          && (++(lun->sync_count) >= lun->sync_interval)) {
5692                 lun->sync_count = 0;
5693                 mtx_unlock(&lun->lun_lock);
5694                 retval = lun->backend->config_write((union ctl_io *)ctsio);
5695         } else {
5696                 mtx_unlock(&lun->lun_lock);
5697                 ctl_set_success(ctsio);
5698                 ctl_done((union ctl_io *)ctsio);
5699         }
5700
5701 bailout:
5702
5703         return (retval);
5704 }
5705
5706 int
5707 ctl_format(struct ctl_scsiio *ctsio)
5708 {
5709         struct scsi_format *cdb;
5710         struct ctl_lun *lun;
5711         struct ctl_softc *ctl_softc;
5712         int length, defect_list_len;
5713
5714         CTL_DEBUG_PRINT(("ctl_format\n"));
5715
5716         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5717         ctl_softc = control_softc;
5718
5719         cdb = (struct scsi_format *)ctsio->cdb;
5720
5721         length = 0;
5722         if (cdb->byte2 & SF_FMTDATA) {
5723                 if (cdb->byte2 & SF_LONGLIST)
5724                         length = sizeof(struct scsi_format_header_long);
5725                 else
5726                         length = sizeof(struct scsi_format_header_short);
5727         }
5728
5729         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5730          && (length > 0)) {
5731                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5732                 ctsio->kern_data_len = length;
5733                 ctsio->kern_total_len = length;
5734                 ctsio->kern_data_resid = 0;
5735                 ctsio->kern_rel_offset = 0;
5736                 ctsio->kern_sg_entries = 0;
5737                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5738                 ctsio->be_move_done = ctl_config_move_done;
5739                 ctl_datamove((union ctl_io *)ctsio);
5740
5741                 return (CTL_RETVAL_COMPLETE);
5742         }
5743
5744         defect_list_len = 0;
5745
5746         if (cdb->byte2 & SF_FMTDATA) {
5747                 if (cdb->byte2 & SF_LONGLIST) {
5748                         struct scsi_format_header_long *header;
5749
5750                         header = (struct scsi_format_header_long *)
5751                                 ctsio->kern_data_ptr;
5752
5753                         defect_list_len = scsi_4btoul(header->defect_list_len);
5754                         if (defect_list_len != 0) {
5755                                 ctl_set_invalid_field(ctsio,
5756                                                       /*sks_valid*/ 1,
5757                                                       /*command*/ 0,
5758                                                       /*field*/ 2,
5759                                                       /*bit_valid*/ 0,
5760                                                       /*bit*/ 0);
5761                                 goto bailout;
5762                         }
5763                 } else {
5764                         struct scsi_format_header_short *header;
5765
5766                         header = (struct scsi_format_header_short *)
5767                                 ctsio->kern_data_ptr;
5768
5769                         defect_list_len = scsi_2btoul(header->defect_list_len);
5770                         if (defect_list_len != 0) {
5771                                 ctl_set_invalid_field(ctsio,
5772                                                       /*sks_valid*/ 1,
5773                                                       /*command*/ 0,
5774                                                       /*field*/ 2,
5775                                                       /*bit_valid*/ 0,
5776                                                       /*bit*/ 0);
5777                                 goto bailout;
5778                         }
5779                 }
5780         }
5781
5782         /*
5783          * The format command will clear out the "Medium format corrupted"
5784          * status if set by the configuration code.  That status is really
5785          * just a way to notify the host that we have lost the media, and
5786          * get them to issue a command that will basically make them think
5787          * they're blowing away the media.
5788          */
5789         mtx_lock(&lun->lun_lock);
5790         lun->flags &= ~CTL_LUN_INOPERABLE;
5791         mtx_unlock(&lun->lun_lock);
5792
5793         ctsio->scsi_status = SCSI_STATUS_OK;
5794         ctsio->io_hdr.status = CTL_SUCCESS;
5795 bailout:
5796
5797         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5798                 free(ctsio->kern_data_ptr, M_CTL);
5799                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5800         }
5801
5802         ctl_done((union ctl_io *)ctsio);
5803         return (CTL_RETVAL_COMPLETE);
5804 }
5805
5806 int
5807 ctl_read_buffer(struct ctl_scsiio *ctsio)
5808 {
5809         struct scsi_read_buffer *cdb;
5810         struct ctl_lun *lun;
5811         int buffer_offset, len;
5812         static uint8_t descr[4];
5813         static uint8_t echo_descr[4] = { 0 };
5814
5815         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5816
5817         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5818         cdb = (struct scsi_read_buffer *)ctsio->cdb;
5819
5820         if (lun->flags & CTL_LUN_PR_RESERVED) {
5821                 uint32_t residx;
5822
5823                 /*
5824                  * XXX KDM need a lock here.
5825                  */
5826                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5827                 if ((lun->res_type == SPR_TYPE_EX_AC
5828                   && residx != lun->pr_res_idx)
5829                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
5830                    || lun->res_type == SPR_TYPE_EX_AC_AR)
5831                   && lun->pr_keys[residx] == 0)) {
5832                         ctl_set_reservation_conflict(ctsio);
5833                         ctl_done((union ctl_io *)ctsio);
5834                         return (CTL_RETVAL_COMPLETE);
5835                 }
5836         }
5837
5838         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5839             (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5840             (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5841                 ctl_set_invalid_field(ctsio,
5842                                       /*sks_valid*/ 1,
5843                                       /*command*/ 1,
5844                                       /*field*/ 1,
5845                                       /*bit_valid*/ 1,
5846                                       /*bit*/ 4);
5847                 ctl_done((union ctl_io *)ctsio);
5848                 return (CTL_RETVAL_COMPLETE);
5849         }
5850
5851         len = scsi_3btoul(cdb->length);
5852         buffer_offset = scsi_3btoul(cdb->offset);
5853
5854         if (buffer_offset + len > sizeof(lun->write_buffer)) {
5855                 ctl_set_invalid_field(ctsio,
5856                                       /*sks_valid*/ 1,
5857                                       /*command*/ 1,
5858                                       /*field*/ 6,
5859                                       /*bit_valid*/ 0,
5860                                       /*bit*/ 0);
5861                 ctl_done((union ctl_io *)ctsio);
5862                 return (CTL_RETVAL_COMPLETE);
5863         }
5864
5865         if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5866                 descr[0] = 0;
5867                 scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5868                 ctsio->kern_data_ptr = descr;
5869                 len = min(len, sizeof(descr));
5870         } else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5871                 ctsio->kern_data_ptr = echo_descr;
5872                 len = min(len, sizeof(echo_descr));
5873         } else
5874                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5875         ctsio->kern_data_len = len;
5876         ctsio->kern_total_len = len;
5877         ctsio->kern_data_resid = 0;
5878         ctsio->kern_rel_offset = 0;
5879         ctsio->kern_sg_entries = 0;
5880         ctsio->be_move_done = ctl_config_move_done;
5881         ctl_datamove((union ctl_io *)ctsio);
5882
5883         return (CTL_RETVAL_COMPLETE);
5884 }
5885
5886 int
5887 ctl_write_buffer(struct ctl_scsiio *ctsio)
5888 {
5889         struct scsi_write_buffer *cdb;
5890         struct ctl_lun *lun;
5891         int buffer_offset, len;
5892
5893         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5894
5895         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5896         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5897
5898         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5899                 ctl_set_invalid_field(ctsio,
5900                                       /*sks_valid*/ 1,
5901                                       /*command*/ 1,
5902                                       /*field*/ 1,
5903                                       /*bit_valid*/ 1,
5904                                       /*bit*/ 4);
5905                 ctl_done((union ctl_io *)ctsio);
5906                 return (CTL_RETVAL_COMPLETE);
5907         }
5908
5909         len = scsi_3btoul(cdb->length);
5910         buffer_offset = scsi_3btoul(cdb->offset);
5911
5912         if (buffer_offset + len > sizeof(lun->write_buffer)) {
5913                 ctl_set_invalid_field(ctsio,
5914                                       /*sks_valid*/ 1,
5915                                       /*command*/ 1,
5916                                       /*field*/ 6,
5917                                       /*bit_valid*/ 0,
5918                                       /*bit*/ 0);
5919                 ctl_done((union ctl_io *)ctsio);
5920                 return (CTL_RETVAL_COMPLETE);
5921         }
5922
5923         /*
5924          * If we've got a kernel request that hasn't been malloced yet,
5925          * malloc it and tell the caller the data buffer is here.
5926          */
5927         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5928                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5929                 ctsio->kern_data_len = len;
5930                 ctsio->kern_total_len = len;
5931                 ctsio->kern_data_resid = 0;
5932                 ctsio->kern_rel_offset = 0;
5933                 ctsio->kern_sg_entries = 0;
5934                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5935                 ctsio->be_move_done = ctl_config_move_done;
5936                 ctl_datamove((union ctl_io *)ctsio);
5937
5938                 return (CTL_RETVAL_COMPLETE);
5939         }
5940
5941         ctl_done((union ctl_io *)ctsio);
5942
5943         return (CTL_RETVAL_COMPLETE);
5944 }
5945
5946 int
5947 ctl_write_same(struct ctl_scsiio *ctsio)
5948 {
5949         struct ctl_lun *lun;
5950         struct ctl_lba_len_flags *lbalen;
5951         uint64_t lba;
5952         uint32_t num_blocks;
5953         int len, retval;
5954         uint8_t byte2;
5955
5956         retval = CTL_RETVAL_COMPLETE;
5957
5958         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5959
5960         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5961
5962         switch (ctsio->cdb[0]) {
5963         case WRITE_SAME_10: {
5964                 struct scsi_write_same_10 *cdb;
5965
5966                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5967
5968                 lba = scsi_4btoul(cdb->addr);
5969                 num_blocks = scsi_2btoul(cdb->length);
5970                 byte2 = cdb->byte2;
5971                 break;
5972         }
5973         case WRITE_SAME_16: {
5974                 struct scsi_write_same_16 *cdb;
5975
5976                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5977
5978                 lba = scsi_8btou64(cdb->addr);
5979                 num_blocks = scsi_4btoul(cdb->length);
5980                 byte2 = cdb->byte2;
5981                 break;
5982         }
5983         default:
5984                 /*
5985                  * We got a command we don't support.  This shouldn't
5986                  * happen, commands should be filtered out above us.
5987                  */
5988                 ctl_set_invalid_opcode(ctsio);
5989                 ctl_done((union ctl_io *)ctsio);
5990
5991                 return (CTL_RETVAL_COMPLETE);
5992                 break; /* NOTREACHED */
5993         }
5994
5995         /* NDOB and ANCHOR flags can be used only together with UNMAP */
5996         if ((byte2 & SWS_UNMAP) == 0 &&
5997             (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5998                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5999                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
6000                 ctl_done((union ctl_io *)ctsio);
6001                 return (CTL_RETVAL_COMPLETE);
6002         }
6003
6004         /*
6005          * The first check is to make sure we're in bounds, the second
6006          * check is to catch wrap-around problems.  If the lba + num blocks
6007          * is less than the lba, then we've wrapped around and the block
6008          * range is invalid anyway.
6009          */
6010         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6011          || ((lba + num_blocks) < lba)) {
6012                 ctl_set_lba_out_of_range(ctsio);
6013                 ctl_done((union ctl_io *)ctsio);
6014                 return (CTL_RETVAL_COMPLETE);
6015         }
6016
6017         /* Zero number of blocks means "to the last logical block" */
6018         if (num_blocks == 0) {
6019                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
6020                         ctl_set_invalid_field(ctsio,
6021                                               /*sks_valid*/ 0,
6022                                               /*command*/ 1,
6023                                               /*field*/ 0,
6024                                               /*bit_valid*/ 0,
6025                                               /*bit*/ 0);
6026                         ctl_done((union ctl_io *)ctsio);
6027                         return (CTL_RETVAL_COMPLETE);
6028                 }
6029                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
6030         }
6031
6032         len = lun->be_lun->blocksize;
6033
6034         /*
6035          * If we've got a kernel request that hasn't been malloced yet,
6036          * malloc it and tell the caller the data buffer is here.
6037          */
6038         if ((byte2 & SWS_NDOB) == 0 &&
6039             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6040                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6041                 ctsio->kern_data_len = len;
6042                 ctsio->kern_total_len = len;
6043                 ctsio->kern_data_resid = 0;
6044                 ctsio->kern_rel_offset = 0;
6045                 ctsio->kern_sg_entries = 0;
6046                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6047                 ctsio->be_move_done = ctl_config_move_done;
6048                 ctl_datamove((union ctl_io *)ctsio);
6049
6050                 return (CTL_RETVAL_COMPLETE);
6051         }
6052
6053         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6054         lbalen->lba = lba;
6055         lbalen->len = num_blocks;
6056         lbalen->flags = byte2;
6057         retval = lun->backend->config_write((union ctl_io *)ctsio);
6058
6059         return (retval);
6060 }
6061
6062 int
6063 ctl_unmap(struct ctl_scsiio *ctsio)
6064 {
6065         struct ctl_lun *lun;
6066         struct scsi_unmap *cdb;
6067         struct ctl_ptr_len_flags *ptrlen;
6068         struct scsi_unmap_header *hdr;
6069         struct scsi_unmap_desc *buf, *end, *endnz, *range;
6070         uint64_t lba;
6071         uint32_t num_blocks;
6072         int len, retval;
6073         uint8_t byte2;
6074
6075         retval = CTL_RETVAL_COMPLETE;
6076
6077         CTL_DEBUG_PRINT(("ctl_unmap\n"));
6078
6079         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6080         cdb = (struct scsi_unmap *)ctsio->cdb;
6081
6082         len = scsi_2btoul(cdb->length);
6083         byte2 = cdb->byte2;
6084
6085         /*
6086          * If we've got a kernel request that hasn't been malloced yet,
6087          * malloc it and tell the caller the data buffer is here.
6088          */
6089         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6090                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6091                 ctsio->kern_data_len = len;
6092                 ctsio->kern_total_len = len;
6093                 ctsio->kern_data_resid = 0;
6094                 ctsio->kern_rel_offset = 0;
6095                 ctsio->kern_sg_entries = 0;
6096                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6097                 ctsio->be_move_done = ctl_config_move_done;
6098                 ctl_datamove((union ctl_io *)ctsio);
6099
6100                 return (CTL_RETVAL_COMPLETE);
6101         }
6102
6103         len = ctsio->kern_total_len - ctsio->kern_data_resid;
6104         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6105         if (len < sizeof (*hdr) ||
6106             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6107             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6108             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6109                 ctl_set_invalid_field(ctsio,
6110                                       /*sks_valid*/ 0,
6111                                       /*command*/ 0,
6112                                       /*field*/ 0,
6113                                       /*bit_valid*/ 0,
6114                                       /*bit*/ 0);
6115                 ctl_done((union ctl_io *)ctsio);
6116                 return (CTL_RETVAL_COMPLETE);
6117         }
6118         len = scsi_2btoul(hdr->desc_length);
6119         buf = (struct scsi_unmap_desc *)(hdr + 1);
6120         end = buf + len / sizeof(*buf);
6121
6122         endnz = buf;
6123         for (range = buf; range < end; range++) {
6124                 lba = scsi_8btou64(range->lba);
6125                 num_blocks = scsi_4btoul(range->length);
6126                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6127                  || ((lba + num_blocks) < lba)) {
6128                         ctl_set_lba_out_of_range(ctsio);
6129                         ctl_done((union ctl_io *)ctsio);
6130                         return (CTL_RETVAL_COMPLETE);
6131                 }
6132                 if (num_blocks != 0)
6133                         endnz = range + 1;
6134         }
6135
6136         /*
6137          * Block backend can not handle zero last range.
6138          * Filter it out and return if there is nothing left.
6139          */
6140         len = (uint8_t *)endnz - (uint8_t *)buf;
6141         if (len == 0) {
6142                 ctl_set_success(ctsio);
6143                 ctl_done((union ctl_io *)ctsio);
6144                 return (CTL_RETVAL_COMPLETE);
6145         }
6146
6147         mtx_lock(&lun->lun_lock);
6148         ptrlen = (struct ctl_ptr_len_flags *)
6149             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6150         ptrlen->ptr = (void *)buf;
6151         ptrlen->len = len;
6152         ptrlen->flags = byte2;
6153         ctl_check_blocked(lun);
6154         mtx_unlock(&lun->lun_lock);
6155
6156         retval = lun->backend->config_write((union ctl_io *)ctsio);
6157         return (retval);
6158 }
6159
6160 /*
6161  * Note that this function currently doesn't actually do anything inside
6162  * CTL to enforce things if the DQue bit is turned on.
6163  *
6164  * Also note that this function can't be used in the default case, because
6165  * the DQue bit isn't set in the changeable mask for the control mode page
6166  * anyway.  This is just here as an example for how to implement a page
6167  * handler, and a placeholder in case we want to allow the user to turn
6168  * tagged queueing on and off.
6169  *
6170  * The D_SENSE bit handling is functional, however, and will turn
6171  * descriptor sense on and off for a given LUN.
6172  */
6173 int
6174 ctl_control_page_handler(struct ctl_scsiio *ctsio,
6175                          struct ctl_page_index *page_index, uint8_t *page_ptr)
6176 {
6177         struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6178         struct ctl_lun *lun;
6179         struct ctl_softc *softc;
6180         int set_ua;
6181         uint32_t initidx;
6182
6183         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6184         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6185         set_ua = 0;
6186
6187         user_cp = (struct scsi_control_page *)page_ptr;
6188         current_cp = (struct scsi_control_page *)
6189                 (page_index->page_data + (page_index->page_len *
6190                 CTL_PAGE_CURRENT));
6191         saved_cp = (struct scsi_control_page *)
6192                 (page_index->page_data + (page_index->page_len *
6193                 CTL_PAGE_SAVED));
6194
6195         softc = control_softc;
6196
6197         mtx_lock(&lun->lun_lock);
6198         if (((current_cp->rlec & SCP_DSENSE) == 0)
6199          && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6200                 /*
6201                  * Descriptor sense is currently turned off and the user
6202                  * wants to turn it on.
6203                  */
6204                 current_cp->rlec |= SCP_DSENSE;
6205                 saved_cp->rlec |= SCP_DSENSE;
6206                 lun->flags |= CTL_LUN_SENSE_DESC;
6207                 set_ua = 1;
6208         } else if (((current_cp->rlec & SCP_DSENSE) != 0)
6209                 && ((user_cp->rlec & SCP_DSENSE) == 0)) {
6210                 /*
6211                  * Descriptor sense is currently turned on, and the user
6212                  * wants to turn it off.
6213                  */
6214                 current_cp->rlec &= ~SCP_DSENSE;
6215                 saved_cp->rlec &= ~SCP_DSENSE;
6216                 lun->flags &= ~CTL_LUN_SENSE_DESC;
6217                 set_ua = 1;
6218         }
6219         if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6220             (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6221                 current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6222                 current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6223                 saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6224                 saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6225                 set_ua = 1;
6226         }
6227         if ((current_cp->eca_and_aen & SCP_SWP) !=
6228             (user_cp->eca_and_aen & SCP_SWP)) {
6229                 current_cp->eca_and_aen &= ~SCP_SWP;
6230                 current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6231                 saved_cp->eca_and_aen &= ~SCP_SWP;
6232                 saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6233                 set_ua = 1;
6234         }
6235         if (set_ua != 0) {
6236                 int i;
6237                 /*
6238                  * Let other initiators know that the mode
6239                  * parameters for this LUN have changed.
6240                  */
6241                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6242                         if (i == initidx)
6243                                 continue;
6244
6245                         lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6246                 }
6247         }
6248         mtx_unlock(&lun->lun_lock);
6249
6250         return (0);
6251 }
6252
6253 int
6254 ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6255                      struct ctl_page_index *page_index, uint8_t *page_ptr)
6256 {
6257         struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6258         struct ctl_lun *lun;
6259         int set_ua;
6260         uint32_t initidx;
6261
6262         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6263         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6264         set_ua = 0;
6265
6266         user_cp = (struct scsi_caching_page *)page_ptr;
6267         current_cp = (struct scsi_caching_page *)
6268                 (page_index->page_data + (page_index->page_len *
6269                 CTL_PAGE_CURRENT));
6270         saved_cp = (struct scsi_caching_page *)
6271                 (page_index->page_data + (page_index->page_len *
6272                 CTL_PAGE_SAVED));
6273
6274         mtx_lock(&lun->lun_lock);
6275         if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6276             (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6277                 current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6278                 current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6279                 saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6280                 saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6281                 set_ua = 1;
6282         }
6283         if (set_ua != 0) {
6284                 int i;
6285                 /*
6286                  * Let other initiators know that the mode
6287                  * parameters for this LUN have changed.
6288                  */
6289                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6290                         if (i == initidx)
6291                                 continue;
6292
6293                         lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6294                 }
6295         }
6296         mtx_unlock(&lun->lun_lock);
6297
6298         return (0);
6299 }
6300
6301 int
6302 ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6303                      struct ctl_page_index *page_index, uint8_t *page_ptr)
6304 {
6305         return (0);
6306 }
6307
6308 int
6309 ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6310                            struct ctl_page_index *page_index, int pc)
6311 {
6312         struct copan_power_subpage *page;
6313
6314         page = (struct copan_power_subpage *)page_index->page_data +
6315                 (page_index->page_len * pc);
6316
6317         switch (pc) {
6318         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6319                 /*
6320                  * We don't update the changable bits for this page.
6321                  */
6322                 break;
6323         case SMS_PAGE_CTRL_CURRENT >> 6:
6324         case SMS_PAGE_CTRL_DEFAULT >> 6:
6325         case SMS_PAGE_CTRL_SAVED >> 6:
6326 #ifdef NEEDTOPORT
6327                 ctl_update_power_subpage(page);
6328 #endif
6329                 break;
6330         default:
6331 #ifdef NEEDTOPORT
6332                 EPRINT(0, "Invalid PC %d!!", pc);
6333 #endif
6334                 break;
6335         }
6336         return (0);
6337 }
6338
6339
6340 int
6341 ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6342                    struct ctl_page_index *page_index, uint8_t *page_ptr)
6343 {
6344         struct copan_aps_subpage *user_sp;
6345         struct copan_aps_subpage *current_sp;
6346         union ctl_modepage_info *modepage_info;
6347         struct ctl_softc *softc;
6348         struct ctl_lun *lun;
6349         int retval;
6350
6351         retval = CTL_RETVAL_COMPLETE;
6352         current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6353                      (page_index->page_len * CTL_PAGE_CURRENT));
6354         softc = control_softc;
6355         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6356
6357         user_sp = (struct copan_aps_subpage *)page_ptr;
6358
6359         modepage_info = (union ctl_modepage_info *)
6360                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6361
6362         modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6363         modepage_info->header.subpage = page_index->subpage;
6364         modepage_info->aps.lock_active = user_sp->lock_active;
6365
6366         mtx_lock(&softc->ctl_lock);
6367
6368         /*
6369          * If there is a request to lock the LUN and another LUN is locked
6370          * this is an error. If the requested LUN is already locked ignore
6371          * the request. If no LUN is locked attempt to lock it.
6372          * if there is a request to unlock the LUN and the LUN is currently
6373          * locked attempt to unlock it. Otherwise ignore the request. i.e.
6374          * if another LUN is locked or no LUN is locked.
6375          */
6376         if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6377                 if (softc->aps_locked_lun == lun->lun) {
6378                         /*
6379                          * This LUN is already locked, so we're done.
6380                          */
6381                         retval = CTL_RETVAL_COMPLETE;
6382                 } else if (softc->aps_locked_lun == 0) {
6383                         /*
6384                          * No one has the lock, pass the request to the
6385                          * backend.
6386                          */
6387                         retval = lun->backend->config_write(
6388                                 (union ctl_io *)ctsio);
6389                 } else {
6390                         /*
6391                          * Someone else has the lock, throw out the request.
6392                          */
6393                         ctl_set_already_locked(ctsio);
6394                         free(ctsio->kern_data_ptr, M_CTL);
6395                         ctl_done((union ctl_io *)ctsio);
6396
6397                         /*
6398                          * Set the return value so that ctl_do_mode_select()
6399                          * won't try to complete the command.  We already
6400                          * completed it here.
6401                          */
6402                         retval = CTL_RETVAL_ERROR;
6403                 }
6404         } else if (softc->aps_locked_lun == lun->lun) {
6405                 /*
6406                  * This LUN is locked, so pass the unlock request to the
6407                  * backend.
6408                  */
6409                 retval = lun->backend->config_write((union ctl_io *)ctsio);
6410         }
6411         mtx_unlock(&softc->ctl_lock);
6412
6413         return (retval);
6414 }
6415
6416 int
6417 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6418                                 struct ctl_page_index *page_index,
6419                                 uint8_t *page_ptr)
6420 {
6421         uint8_t *c;
6422         int i;
6423
6424         c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6425         ctl_time_io_secs =
6426                 (c[0] << 8) |
6427                 (c[1] << 0) |
6428                 0;
6429         CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6430         printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6431         printf("page data:");
6432         for (i=0; i<8; i++)
6433                 printf(" %.2x",page_ptr[i]);
6434         printf("\n");
6435         return (0);
6436 }
6437
6438 int
6439 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6440                                struct ctl_page_index *page_index,
6441                                int pc)
6442 {
6443         struct copan_debugconf_subpage *page;
6444
6445         page = (struct copan_debugconf_subpage *)page_index->page_data +
6446                 (page_index->page_len * pc);
6447
6448         switch (pc) {
6449         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6450         case SMS_PAGE_CTRL_DEFAULT >> 6:
6451         case SMS_PAGE_CTRL_SAVED >> 6:
6452                 /*
6453                  * We don't update the changable or default bits for this page.
6454                  */
6455                 break;
6456         case SMS_PAGE_CTRL_CURRENT >> 6:
6457                 page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6458                 page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6459                 break;
6460         default:
6461 #ifdef NEEDTOPORT
6462                 EPRINT(0, "Invalid PC %d!!", pc);
6463 #endif /* NEEDTOPORT */
6464                 break;
6465         }
6466         return (0);
6467 }
6468
6469
6470 static int
6471 ctl_do_mode_select(union ctl_io *io)
6472 {
6473         struct scsi_mode_page_header *page_header;
6474         struct ctl_page_index *page_index;
6475         struct ctl_scsiio *ctsio;
6476         int control_dev, page_len;
6477         int page_len_offset, page_len_size;
6478         union ctl_modepage_info *modepage_info;
6479         struct ctl_lun *lun;
6480         int *len_left, *len_used;
6481         int retval, i;
6482
6483         ctsio = &io->scsiio;
6484         page_index = NULL;
6485         page_len = 0;
6486         retval = CTL_RETVAL_COMPLETE;
6487
6488         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6489
6490         if (lun->be_lun->lun_type != T_DIRECT)
6491                 control_dev = 1;
6492         else
6493                 control_dev = 0;
6494
6495         modepage_info = (union ctl_modepage_info *)
6496                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6497         len_left = &modepage_info->header.len_left;
6498         len_used = &modepage_info->header.len_used;
6499
6500 do_next_page:
6501
6502         page_header = (struct scsi_mode_page_header *)
6503                 (ctsio->kern_data_ptr + *len_used);
6504
6505         if (*len_left == 0) {
6506                 free(ctsio->kern_data_ptr, M_CTL);
6507                 ctl_set_success(ctsio);
6508                 ctl_done((union ctl_io *)ctsio);
6509                 return (CTL_RETVAL_COMPLETE);
6510         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6511
6512                 free(ctsio->kern_data_ptr, M_CTL);
6513                 ctl_set_param_len_error(ctsio);
6514                 ctl_done((union ctl_io *)ctsio);
6515                 return (CTL_RETVAL_COMPLETE);
6516
6517         } else if ((page_header->page_code & SMPH_SPF)
6518                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6519
6520                 free(ctsio->kern_data_ptr, M_CTL);
6521                 ctl_set_param_len_error(ctsio);
6522                 ctl_done((union ctl_io *)ctsio);
6523                 return (CTL_RETVAL_COMPLETE);
6524         }
6525
6526
6527         /*
6528          * XXX KDM should we do something with the block descriptor?
6529          */
6530         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6531
6532                 if ((control_dev != 0)
6533                  && (lun->mode_pages.index[i].page_flags &
6534                      CTL_PAGE_FLAG_DISK_ONLY))
6535                         continue;
6536
6537                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6538                     (page_header->page_code & SMPH_PC_MASK))
6539                         continue;
6540
6541                 /*
6542                  * If neither page has a subpage code, then we've got a
6543                  * match.
6544                  */
6545                 if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6546                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6547                         page_index = &lun->mode_pages.index[i];
6548                         page_len = page_header->page_length;
6549                         break;
6550                 }
6551
6552                 /*
6553                  * If both pages have subpages, then the subpage numbers
6554                  * have to match.
6555                  */
6556                 if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6557                   && (page_header->page_code & SMPH_SPF)) {
6558                         struct scsi_mode_page_header_sp *sph;
6559
6560                         sph = (struct scsi_mode_page_header_sp *)page_header;
6561
6562                         if (lun->mode_pages.index[i].subpage ==
6563                             sph->subpage) {
6564                                 page_index = &lun->mode_pages.index[i];
6565                                 page_len = scsi_2btoul(sph->page_length);
6566                                 break;
6567                         }
6568                 }
6569         }
6570
6571         /*
6572          * If we couldn't find the page, or if we don't have a mode select
6573          * handler for it, send back an error to the user.
6574          */
6575         if ((page_index == NULL)
6576          || (page_index->select_handler == NULL)) {
6577                 ctl_set_invalid_field(ctsio,
6578                                       /*sks_valid*/ 1,
6579                                       /*command*/ 0,
6580                                       /*field*/ *len_used,
6581                                       /*bit_valid*/ 0,
6582                                       /*bit*/ 0);
6583                 free(ctsio->kern_data_ptr, M_CTL);
6584                 ctl_done((union ctl_io *)ctsio);
6585                 return (CTL_RETVAL_COMPLETE);
6586         }
6587
6588         if (page_index->page_code & SMPH_SPF) {
6589                 page_len_offset = 2;
6590                 page_len_size = 2;
6591         } else {
6592                 page_len_size = 1;
6593                 page_len_offset = 1;
6594         }
6595
6596         /*
6597          * If the length the initiator gives us isn't the one we specify in
6598          * the mode page header, or if they didn't specify enough data in
6599          * the CDB to avoid truncating this page, kick out the request.
6600          */
6601         if ((page_len != (page_index->page_len - page_len_offset -
6602                           page_len_size))
6603          || (*len_left < page_index->page_len)) {
6604
6605
6606                 ctl_set_invalid_field(ctsio,
6607                                       /*sks_valid*/ 1,
6608                                       /*command*/ 0,
6609                                       /*field*/ *len_used + page_len_offset,
6610                                       /*bit_valid*/ 0,
6611                                       /*bit*/ 0);
6612                 free(ctsio->kern_data_ptr, M_CTL);
6613                 ctl_done((union ctl_io *)ctsio);
6614                 return (CTL_RETVAL_COMPLETE);
6615         }
6616
6617         /*
6618          * Run through the mode page, checking to make sure that the bits
6619          * the user changed are actually legal for him to change.
6620          */
6621         for (i = 0; i < page_index->page_len; i++) {
6622                 uint8_t *user_byte, *change_mask, *current_byte;
6623                 int bad_bit;
6624                 int j;
6625
6626                 user_byte = (uint8_t *)page_header + i;
6627                 change_mask = page_index->page_data +
6628                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6629                 current_byte = page_index->page_data +
6630                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6631
6632                 /*
6633                  * Check to see whether the user set any bits in this byte
6634                  * that he is not allowed to set.
6635                  */
6636                 if ((*user_byte & ~(*change_mask)) ==
6637                     (*current_byte & ~(*change_mask)))
6638                         continue;
6639
6640                 /*
6641                  * Go through bit by bit to determine which one is illegal.
6642                  */
6643                 bad_bit = 0;
6644                 for (j = 7; j >= 0; j--) {
6645                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6646                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6647                                 bad_bit = i;
6648                                 break;
6649                         }
6650                 }
6651                 ctl_set_invalid_field(ctsio,
6652                                       /*sks_valid*/ 1,
6653                                       /*command*/ 0,
6654                                       /*field*/ *len_used + i,
6655                                       /*bit_valid*/ 1,
6656                                       /*bit*/ bad_bit);
6657                 free(ctsio->kern_data_ptr, M_CTL);
6658                 ctl_done((union ctl_io *)ctsio);
6659                 return (CTL_RETVAL_COMPLETE);
6660         }
6661
6662         /*
6663          * Decrement these before we call the page handler, since we may
6664          * end up getting called back one way or another before the handler
6665          * returns to this context.
6666          */
6667         *len_left -= page_index->page_len;
6668         *len_used += page_index->page_len;
6669
6670         retval = page_index->select_handler(ctsio, page_index,
6671                                             (uint8_t *)page_header);
6672
6673         /*
6674          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6675          * wait until this queued command completes to finish processing
6676          * the mode page.  If it returns anything other than
6677          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6678          * already set the sense information, freed the data pointer, and
6679          * completed the io for us.
6680          */
6681         if (retval != CTL_RETVAL_COMPLETE)
6682                 goto bailout_no_done;
6683
6684         /*
6685          * If the initiator sent us more than one page, parse the next one.
6686          */
6687         if (*len_left > 0)
6688                 goto do_next_page;
6689
6690         ctl_set_success(ctsio);
6691         free(ctsio->kern_data_ptr, M_CTL);
6692         ctl_done((union ctl_io *)ctsio);
6693
6694 bailout_no_done:
6695
6696         return (CTL_RETVAL_COMPLETE);
6697
6698 }
6699
6700 int
6701 ctl_mode_select(struct ctl_scsiio *ctsio)
6702 {
6703         int param_len, pf, sp;
6704         int header_size, bd_len;
6705         int len_left, len_used;
6706         struct ctl_page_index *page_index;
6707         struct ctl_lun *lun;
6708         int control_dev, page_len;
6709         union ctl_modepage_info *modepage_info;
6710         int retval;
6711
6712         pf = 0;
6713         sp = 0;
6714         page_len = 0;
6715         len_used = 0;
6716         len_left = 0;
6717         retval = 0;
6718         bd_len = 0;
6719         page_index = NULL;
6720
6721         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6722
6723         if (lun->be_lun->lun_type != T_DIRECT)
6724                 control_dev = 1;
6725         else
6726                 control_dev = 0;
6727
6728         switch (ctsio->cdb[0]) {
6729         case MODE_SELECT_6: {
6730                 struct scsi_mode_select_6 *cdb;
6731
6732                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6733
6734                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6735                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6736
6737                 param_len = cdb->length;
6738                 header_size = sizeof(struct scsi_mode_header_6);
6739                 break;
6740         }
6741         case MODE_SELECT_10: {
6742                 struct scsi_mode_select_10 *cdb;
6743
6744                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6745
6746                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6747                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6748
6749                 param_len = scsi_2btoul(cdb->length);
6750                 header_size = sizeof(struct scsi_mode_header_10);
6751                 break;
6752         }
6753         default:
6754                 ctl_set_invalid_opcode(ctsio);
6755                 ctl_done((union ctl_io *)ctsio);
6756                 return (CTL_RETVAL_COMPLETE);
6757                 break; /* NOTREACHED */
6758         }
6759
6760         /*
6761          * From SPC-3:
6762          * "A parameter list length of zero indicates that the Data-Out Buffer
6763          * shall be empty. This condition shall not be considered as an error."
6764          */
6765         if (param_len == 0) {
6766                 ctl_set_success(ctsio);
6767                 ctl_done((union ctl_io *)ctsio);
6768                 return (CTL_RETVAL_COMPLETE);
6769         }
6770
6771         /*
6772          * Since we'll hit this the first time through, prior to
6773          * allocation, we don't need to free a data buffer here.
6774          */
6775         if (param_len < header_size) {
6776                 ctl_set_param_len_error(ctsio);
6777                 ctl_done((union ctl_io *)ctsio);
6778                 return (CTL_RETVAL_COMPLETE);
6779         }
6780
6781         /*
6782          * Allocate the data buffer and grab the user's data.  In theory,
6783          * we shouldn't have to sanity check the parameter list length here
6784          * because the maximum size is 64K.  We should be able to malloc
6785          * that much without too many problems.
6786          */
6787         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6788                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6789                 ctsio->kern_data_len = param_len;
6790                 ctsio->kern_total_len = param_len;
6791                 ctsio->kern_data_resid = 0;
6792                 ctsio->kern_rel_offset = 0;
6793                 ctsio->kern_sg_entries = 0;
6794                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6795                 ctsio->be_move_done = ctl_config_move_done;
6796                 ctl_datamove((union ctl_io *)ctsio);
6797
6798                 return (CTL_RETVAL_COMPLETE);
6799         }
6800
6801         switch (ctsio->cdb[0]) {
6802         case MODE_SELECT_6: {
6803                 struct scsi_mode_header_6 *mh6;
6804
6805                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6806                 bd_len = mh6->blk_desc_len;
6807                 break;
6808         }
6809         case MODE_SELECT_10: {
6810                 struct scsi_mode_header_10 *mh10;
6811
6812                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6813                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6814                 break;
6815         }
6816         default:
6817                 panic("Invalid CDB type %#x", ctsio->cdb[0]);
6818                 break;
6819         }
6820
6821         if (param_len < (header_size + bd_len)) {
6822                 free(ctsio->kern_data_ptr, M_CTL);
6823                 ctl_set_param_len_error(ctsio);
6824                 ctl_done((union ctl_io *)ctsio);
6825                 return (CTL_RETVAL_COMPLETE);
6826         }
6827
6828         /*
6829          * Set the IO_CONT flag, so that if this I/O gets passed to
6830          * ctl_config_write_done(), it'll get passed back to
6831          * ctl_do_mode_select() for further processing, or completion if
6832          * we're all done.
6833          */
6834         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6835         ctsio->io_cont = ctl_do_mode_select;
6836
6837         modepage_info = (union ctl_modepage_info *)
6838                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6839
6840         memset(modepage_info, 0, sizeof(*modepage_info));
6841
6842         len_left = param_len - header_size - bd_len;
6843         len_used = header_size + bd_len;
6844
6845         modepage_info->header.len_left = len_left;
6846         modepage_info->header.len_used = len_used;
6847
6848         return (ctl_do_mode_select((union ctl_io *)ctsio));
6849 }
6850
6851 int
6852 ctl_mode_sense(struct ctl_scsiio *ctsio)
6853 {
6854         struct ctl_lun *lun;
6855         int pc, page_code, dbd, llba, subpage;
6856         int alloc_len, page_len, header_len, total_len;
6857         struct scsi_mode_block_descr *block_desc;
6858         struct ctl_page_index *page_index;
6859         int control_dev;
6860
6861         dbd = 0;
6862         llba = 0;
6863         block_desc = NULL;
6864         page_index = NULL;
6865
6866         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6867
6868         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6869
6870         if (lun->be_lun->lun_type != T_DIRECT)
6871                 control_dev = 1;
6872         else
6873                 control_dev = 0;
6874
6875         if (lun->flags & CTL_LUN_PR_RESERVED) {
6876                 uint32_t residx;
6877
6878                 /*
6879                  * XXX KDM need a lock here.
6880                  */
6881                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6882                 if ((lun->res_type == SPR_TYPE_EX_AC
6883                   && residx != lun->pr_res_idx)
6884                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
6885                    || lun->res_type == SPR_TYPE_EX_AC_AR)
6886                   && lun->pr_keys[residx] == 0)) {
6887                         ctl_set_reservation_conflict(ctsio);
6888                         ctl_done((union ctl_io *)ctsio);
6889                         return (CTL_RETVAL_COMPLETE);
6890                 }
6891         }
6892
6893         switch (ctsio->cdb[0]) {
6894         case MODE_SENSE_6: {
6895                 struct scsi_mode_sense_6 *cdb;
6896
6897                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6898
6899                 header_len = sizeof(struct scsi_mode_hdr_6);
6900                 if (cdb->byte2 & SMS_DBD)
6901                         dbd = 1;
6902                 else
6903                         header_len += sizeof(struct scsi_mode_block_descr);
6904
6905                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6906                 page_code = cdb->page & SMS_PAGE_CODE;
6907                 subpage = cdb->subpage;
6908                 alloc_len = cdb->length;
6909                 break;
6910         }
6911         case MODE_SENSE_10: {
6912                 struct scsi_mode_sense_10 *cdb;
6913
6914                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6915
6916                 header_len = sizeof(struct scsi_mode_hdr_10);
6917
6918                 if (cdb->byte2 & SMS_DBD)
6919                         dbd = 1;
6920                 else
6921                         header_len += sizeof(struct scsi_mode_block_descr);
6922                 if (cdb->byte2 & SMS10_LLBAA)
6923                         llba = 1;
6924                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6925                 page_code = cdb->page & SMS_PAGE_CODE;
6926                 subpage = cdb->subpage;
6927                 alloc_len = scsi_2btoul(cdb->length);
6928                 break;
6929         }
6930         default:
6931                 ctl_set_invalid_opcode(ctsio);
6932                 ctl_done((union ctl_io *)ctsio);
6933                 return (CTL_RETVAL_COMPLETE);
6934                 break; /* NOTREACHED */
6935         }
6936
6937         /*
6938          * We have to make a first pass through to calculate the size of
6939          * the pages that match the user's query.  Then we allocate enough
6940          * memory to hold it, and actually copy the data into the buffer.
6941          */
6942         switch (page_code) {
6943         case SMS_ALL_PAGES_PAGE: {
6944                 int i;
6945
6946                 page_len = 0;
6947
6948                 /*
6949                  * At the moment, values other than 0 and 0xff here are
6950                  * reserved according to SPC-3.
6951                  */
6952                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6953                  && (subpage != SMS_SUBPAGE_ALL)) {
6954                         ctl_set_invalid_field(ctsio,
6955                                               /*sks_valid*/ 1,
6956                                               /*command*/ 1,
6957                                               /*field*/ 3,
6958                                               /*bit_valid*/ 0,
6959                                               /*bit*/ 0);
6960                         ctl_done((union ctl_io *)ctsio);
6961                         return (CTL_RETVAL_COMPLETE);
6962                 }
6963
6964                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6965                         if ((control_dev != 0)
6966                          && (lun->mode_pages.index[i].page_flags &
6967                              CTL_PAGE_FLAG_DISK_ONLY))
6968                                 continue;
6969
6970                         /*
6971                          * We don't use this subpage if the user didn't
6972                          * request all subpages.
6973                          */
6974                         if ((lun->mode_pages.index[i].subpage != 0)
6975                          && (subpage == SMS_SUBPAGE_PAGE_0))
6976                                 continue;
6977
6978 #if 0
6979                         printf("found page %#x len %d\n",
6980                                lun->mode_pages.index[i].page_code &
6981                                SMPH_PC_MASK,
6982                                lun->mode_pages.index[i].page_len);
6983 #endif
6984                         page_len += lun->mode_pages.index[i].page_len;
6985                 }
6986                 break;
6987         }
6988         default: {
6989                 int i;
6990
6991                 page_len = 0;
6992
6993                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6994                         /* Look for the right page code */
6995                         if ((lun->mode_pages.index[i].page_code &
6996                              SMPH_PC_MASK) != page_code)
6997                                 continue;
6998
6999                         /* Look for the right subpage or the subpage wildcard*/
7000                         if ((lun->mode_pages.index[i].subpage != subpage)
7001                          && (subpage != SMS_SUBPAGE_ALL))
7002                                 continue;
7003
7004                         /* Make sure the page is supported for this dev type */
7005                         if ((control_dev != 0)
7006                          && (lun->mode_pages.index[i].page_flags &
7007                              CTL_PAGE_FLAG_DISK_ONLY))
7008                                 continue;
7009
7010 #if 0
7011                         printf("found page %#x len %d\n",
7012                                lun->mode_pages.index[i].page_code &
7013                                SMPH_PC_MASK,
7014                                lun->mode_pages.index[i].page_len);
7015 #endif
7016
7017                         page_len += lun->mode_pages.index[i].page_len;
7018                 }
7019
7020                 if (page_len == 0) {
7021                         ctl_set_invalid_field(ctsio,
7022                                               /*sks_valid*/ 1,
7023                                               /*command*/ 1,
7024                                               /*field*/ 2,
7025                                               /*bit_valid*/ 1,
7026                                               /*bit*/ 5);
7027                         ctl_done((union ctl_io *)ctsio);
7028                         return (CTL_RETVAL_COMPLETE);
7029                 }
7030                 break;
7031         }
7032         }
7033
7034         total_len = header_len + page_len;
7035 #if 0
7036         printf("header_len = %d, page_len = %d, total_len = %d\n",
7037                header_len, page_len, total_len);
7038 #endif
7039
7040         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7041         ctsio->kern_sg_entries = 0;
7042         ctsio->kern_data_resid = 0;
7043         ctsio->kern_rel_offset = 0;
7044         if (total_len < alloc_len) {
7045                 ctsio->residual = alloc_len - total_len;
7046                 ctsio->kern_data_len = total_len;
7047                 ctsio->kern_total_len = total_len;
7048         } else {
7049                 ctsio->residual = 0;
7050                 ctsio->kern_data_len = alloc_len;
7051                 ctsio->kern_total_len = alloc_len;
7052         }
7053
7054         switch (ctsio->cdb[0]) {
7055         case MODE_SENSE_6: {
7056                 struct scsi_mode_hdr_6 *header;
7057
7058                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7059
7060                 header->datalen = ctl_min(total_len - 1, 254);
7061                 if (control_dev == 0) {
7062                         header->dev_specific = 0x10; /* DPOFUA */
7063                         if ((lun->flags & CTL_LUN_READONLY) ||
7064                             (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
7065                             .eca_and_aen & SCP_SWP) != 0)
7066                                     header->dev_specific |= 0x80; /* WP */
7067                 }
7068                 if (dbd)
7069                         header->block_descr_len = 0;
7070                 else
7071                         header->block_descr_len =
7072                                 sizeof(struct scsi_mode_block_descr);
7073                 block_desc = (struct scsi_mode_block_descr *)&header[1];
7074                 break;
7075         }
7076         case MODE_SENSE_10: {
7077                 struct scsi_mode_hdr_10 *header;
7078                 int datalen;
7079
7080                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7081
7082                 datalen = ctl_min(total_len - 2, 65533);
7083                 scsi_ulto2b(datalen, header->datalen);
7084                 if (control_dev == 0) {
7085                         header->dev_specific = 0x10; /* DPOFUA */
7086                         if ((lun->flags & CTL_LUN_READONLY) ||
7087                             (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
7088                             .eca_and_aen & SCP_SWP) != 0)
7089                                     header->dev_specific |= 0x80; /* WP */
7090                 }
7091                 if (dbd)
7092                         scsi_ulto2b(0, header->block_descr_len);
7093                 else
7094                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7095                                     header->block_descr_len);
7096                 block_desc = (struct scsi_mode_block_descr *)&header[1];
7097                 break;
7098         }
7099         default:
7100                 panic("invalid CDB type %#x", ctsio->cdb[0]);
7101                 break; /* NOTREACHED */
7102         }
7103
7104         /*
7105          * If we've got a disk, use its blocksize in the block
7106          * descriptor.  Otherwise, just set it to 0.
7107          */
7108         if (dbd == 0) {
7109                 if (control_dev == 0)
7110                         scsi_ulto3b(lun->be_lun->blocksize,
7111                                     block_desc->block_len);
7112                 else
7113                         scsi_ulto3b(0, block_desc->block_len);
7114         }
7115
7116         switch (page_code) {
7117         case SMS_ALL_PAGES_PAGE: {
7118                 int i, data_used;
7119
7120                 data_used = header_len;
7121                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7122                         struct ctl_page_index *page_index;
7123
7124                         page_index = &lun->mode_pages.index[i];
7125
7126                         if ((control_dev != 0)
7127                          && (page_index->page_flags &
7128                             CTL_PAGE_FLAG_DISK_ONLY))
7129                                 continue;
7130
7131                         /*
7132                          * We don't use this subpage if the user didn't
7133                          * request all subpages.  We already checked (above)
7134                          * to make sure the user only specified a subpage
7135                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7136                          */
7137                         if ((page_index->subpage != 0)
7138                          && (subpage == SMS_SUBPAGE_PAGE_0))
7139                                 continue;
7140
7141                         /*
7142                          * Call the handler, if it exists, to update the
7143                          * page to the latest values.
7144                          */
7145                         if (page_index->sense_handler != NULL)
7146                                 page_index->sense_handler(ctsio, page_index,pc);
7147
7148                         memcpy(ctsio->kern_data_ptr + data_used,
7149                                page_index->page_data +
7150                                (page_index->page_len * pc),
7151                                page_index->page_len);
7152                         data_used += page_index->page_len;
7153                 }
7154                 break;
7155         }
7156         default: {
7157                 int i, data_used;
7158
7159                 data_used = header_len;
7160
7161                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7162                         struct ctl_page_index *page_index;
7163
7164                         page_index = &lun->mode_pages.index[i];
7165
7166                         /* Look for the right page code */
7167                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7168                                 continue;
7169
7170                         /* Look for the right subpage or the subpage wildcard*/
7171                         if ((page_index->subpage != subpage)
7172                          && (subpage != SMS_SUBPAGE_ALL))
7173                                 continue;
7174
7175                         /* Make sure the page is supported for this dev type */
7176                         if ((control_dev != 0)
7177                          && (page_index->page_flags &
7178                              CTL_PAGE_FLAG_DISK_ONLY))
7179                                 continue;
7180
7181                         /*
7182                          * Call the handler, if it exists, to update the
7183                          * page to the latest values.
7184                          */
7185                         if (page_index->sense_handler != NULL)
7186                                 page_index->sense_handler(ctsio, page_index,pc);
7187
7188                         memcpy(ctsio->kern_data_ptr + data_used,
7189                                page_index->page_data +
7190                                (page_index->page_len * pc),
7191                                page_index->page_len);
7192                         data_used += page_index->page_len;
7193                 }
7194                 break;
7195         }
7196         }
7197
7198         ctsio->scsi_status = SCSI_STATUS_OK;
7199
7200         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7201         ctsio->be_move_done = ctl_config_move_done;
7202         ctl_datamove((union ctl_io *)ctsio);
7203
7204         return (CTL_RETVAL_COMPLETE);
7205 }
7206
7207 int
7208 ctl_read_capacity(struct ctl_scsiio *ctsio)
7209 {
7210         struct scsi_read_capacity *cdb;
7211         struct scsi_read_capacity_data *data;
7212         struct ctl_lun *lun;
7213         uint32_t lba;
7214
7215         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7216
7217         cdb = (struct scsi_read_capacity *)ctsio->cdb;
7218
7219         lba = scsi_4btoul(cdb->addr);
7220         if (((cdb->pmi & SRC_PMI) == 0)
7221          && (lba != 0)) {
7222                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7223                                       /*sks_valid*/ 1,
7224                                       /*command*/ 1,
7225                                       /*field*/ 2,
7226                                       /*bit_valid*/ 0,
7227                                       /*bit*/ 0);
7228                 ctl_done((union ctl_io *)ctsio);
7229                 return (CTL_RETVAL_COMPLETE);
7230         }
7231
7232         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7233
7234         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7235         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7236         ctsio->residual = 0;
7237         ctsio->kern_data_len = sizeof(*data);
7238         ctsio->kern_total_len = sizeof(*data);
7239         ctsio->kern_data_resid = 0;
7240         ctsio->kern_rel_offset = 0;
7241         ctsio->kern_sg_entries = 0;
7242
7243         /*
7244          * If the maximum LBA is greater than 0xfffffffe, the user must
7245          * issue a SERVICE ACTION IN (16) command, with the read capacity
7246          * serivce action set.
7247          */
7248         if (lun->be_lun->maxlba > 0xfffffffe)
7249                 scsi_ulto4b(0xffffffff, data->addr);
7250         else
7251                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7252
7253         /*
7254          * XXX KDM this may not be 512 bytes...
7255          */
7256         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7257
7258         ctsio->scsi_status = SCSI_STATUS_OK;
7259
7260         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7261         ctsio->be_move_done = ctl_config_move_done;
7262         ctl_datamove((union ctl_io *)ctsio);
7263
7264         return (CTL_RETVAL_COMPLETE);
7265 }
7266
7267 int
7268 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7269 {
7270         struct scsi_read_capacity_16 *cdb;
7271         struct scsi_read_capacity_data_long *data;
7272         struct ctl_lun *lun;
7273         uint64_t lba;
7274         uint32_t alloc_len;
7275
7276         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7277
7278         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7279
7280         alloc_len = scsi_4btoul(cdb->alloc_len);
7281         lba = scsi_8btou64(cdb->addr);
7282
7283         if ((cdb->reladr & SRC16_PMI)
7284          && (lba != 0)) {
7285                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7286                                       /*sks_valid*/ 1,
7287                                       /*command*/ 1,
7288                                       /*field*/ 2,
7289                                       /*bit_valid*/ 0,
7290                                       /*bit*/ 0);
7291                 ctl_done((union ctl_io *)ctsio);
7292                 return (CTL_RETVAL_COMPLETE);
7293         }
7294
7295         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7296
7297         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7298         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7299
7300         if (sizeof(*data) < alloc_len) {
7301                 ctsio->residual = alloc_len - sizeof(*data);
7302                 ctsio->kern_data_len = sizeof(*data);
7303                 ctsio->kern_total_len = sizeof(*data);
7304         } else {
7305                 ctsio->residual = 0;
7306                 ctsio->kern_data_len = alloc_len;
7307                 ctsio->kern_total_len = alloc_len;
7308         }
7309         ctsio->kern_data_resid = 0;
7310         ctsio->kern_rel_offset = 0;
7311         ctsio->kern_sg_entries = 0;
7312
7313         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7314         /* XXX KDM this may not be 512 bytes... */
7315         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7316         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7317         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7318         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7319                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7320
7321         ctsio->scsi_status = SCSI_STATUS_OK;
7322
7323         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7324         ctsio->be_move_done = ctl_config_move_done;
7325         ctl_datamove((union ctl_io *)ctsio);
7326
7327         return (CTL_RETVAL_COMPLETE);
7328 }
7329
7330 int
7331 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7332 {
7333         struct scsi_maintenance_in *cdb;
7334         int retval;
7335         int alloc_len, ext, total_len = 0, g, p, pc, pg;
7336         int num_target_port_groups, num_target_ports, single;
7337         struct ctl_lun *lun;
7338         struct ctl_softc *softc;
7339         struct ctl_port *port;
7340         struct scsi_target_group_data *rtg_ptr;
7341         struct scsi_target_group_data_extended *rtg_ext_ptr;
7342         struct scsi_target_port_group_descriptor *tpg_desc;
7343
7344         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7345
7346         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7347         softc = control_softc;
7348         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7349
7350         retval = CTL_RETVAL_COMPLETE;
7351
7352         switch (cdb->byte2 & STG_PDF_MASK) {
7353         case STG_PDF_LENGTH:
7354                 ext = 0;
7355                 break;
7356         case STG_PDF_EXTENDED:
7357                 ext = 1;
7358                 break;
7359         default:
7360                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7361                                       /*sks_valid*/ 1,
7362                                       /*command*/ 1,
7363                                       /*field*/ 2,
7364                                       /*bit_valid*/ 1,
7365                                       /*bit*/ 5);
7366                 ctl_done((union ctl_io *)ctsio);
7367                 return(retval);
7368         }
7369
7370         single = ctl_is_single;
7371         if (single)
7372                 num_target_port_groups = 1;
7373         else
7374                 num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7375         num_target_ports = 0;
7376         mtx_lock(&softc->ctl_lock);
7377         STAILQ_FOREACH(port, &softc->port_list, links) {
7378                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7379                         continue;
7380                 if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7381                         continue;
7382                 num_target_ports++;
7383         }
7384         mtx_unlock(&softc->ctl_lock);
7385
7386         if (ext)
7387                 total_len = sizeof(struct scsi_target_group_data_extended);
7388         else
7389                 total_len = sizeof(struct scsi_target_group_data);
7390         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7391                 num_target_port_groups +
7392             sizeof(struct scsi_target_port_descriptor) *
7393                 num_target_ports * num_target_port_groups;
7394
7395         alloc_len = scsi_4btoul(cdb->length);
7396
7397         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7398
7399         ctsio->kern_sg_entries = 0;
7400
7401         if (total_len < alloc_len) {
7402                 ctsio->residual = alloc_len - total_len;
7403                 ctsio->kern_data_len = total_len;
7404                 ctsio->kern_total_len = total_len;
7405         } else {
7406                 ctsio->residual = 0;
7407                 ctsio->kern_data_len = alloc_len;
7408                 ctsio->kern_total_len = alloc_len;
7409         }
7410         ctsio->kern_data_resid = 0;
7411         ctsio->kern_rel_offset = 0;
7412
7413         if (ext) {
7414                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7415                     ctsio->kern_data_ptr;
7416                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7417                 rtg_ext_ptr->format_type = 0x10;
7418                 rtg_ext_ptr->implicit_transition_time = 0;
7419                 tpg_desc = &rtg_ext_ptr->groups[0];
7420         } else {
7421                 rtg_ptr = (struct scsi_target_group_data *)
7422                     ctsio->kern_data_ptr;
7423                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7424                 tpg_desc = &rtg_ptr->groups[0];
7425         }
7426
7427         pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7428         mtx_lock(&softc->ctl_lock);
7429         for (g = 0; g < num_target_port_groups; g++) {
7430                 if (g == pg)
7431                         tpg_desc->pref_state = TPG_PRIMARY |
7432                             TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7433                 else
7434                         tpg_desc->pref_state =
7435                             TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7436                 tpg_desc->support = TPG_AO_SUP;
7437                 if (!single)
7438                         tpg_desc->support |= TPG_AN_SUP;
7439                 scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7440                 tpg_desc->status = TPG_IMPLICIT;
7441                 pc = 0;
7442                 STAILQ_FOREACH(port, &softc->port_list, links) {
7443                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7444                                 continue;
7445                         if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7446                             CTL_MAX_LUNS)
7447                                 continue;
7448                         p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7449                         scsi_ulto2b(p, tpg_desc->descriptors[pc].
7450                             relative_target_port_identifier);
7451                         pc++;
7452                 }
7453                 tpg_desc->target_port_count = pc;
7454                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7455                     &tpg_desc->descriptors[pc];
7456         }
7457         mtx_unlock(&softc->ctl_lock);
7458
7459         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7460         ctsio->be_move_done = ctl_config_move_done;
7461
7462         CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7463                          ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7464                          ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7465                          ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7466                          ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7467
7468         ctl_datamove((union ctl_io *)ctsio);
7469         return(retval);
7470 }
7471
7472 int
7473 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7474 {
7475         struct ctl_lun *lun;
7476         struct scsi_report_supported_opcodes *cdb;
7477         const struct ctl_cmd_entry *entry, *sentry;
7478         struct scsi_report_supported_opcodes_all *all;
7479         struct scsi_report_supported_opcodes_descr *descr;
7480         struct scsi_report_supported_opcodes_one *one;
7481         int retval;
7482         int alloc_len, total_len;
7483         int opcode, service_action, i, j, num;
7484
7485         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7486
7487         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7488         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7489
7490         retval = CTL_RETVAL_COMPLETE;
7491
7492         opcode = cdb->requested_opcode;
7493         service_action = scsi_2btoul(cdb->requested_service_action);
7494         switch (cdb->options & RSO_OPTIONS_MASK) {
7495         case RSO_OPTIONS_ALL:
7496                 num = 0;
7497                 for (i = 0; i < 256; i++) {
7498                         entry = &ctl_cmd_table[i];
7499                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7500                                 for (j = 0; j < 32; j++) {
7501                                         sentry = &((const struct ctl_cmd_entry *)
7502                                             entry->execute)[j];
7503                                         if (ctl_cmd_applicable(
7504                                             lun->be_lun->lun_type, sentry))
7505                                                 num++;
7506                                 }
7507                         } else {
7508                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7509                                     entry))
7510                                         num++;
7511                         }
7512                 }
7513                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7514                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7515                 break;
7516         case RSO_OPTIONS_OC:
7517                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7518                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7519                                               /*sks_valid*/ 1,
7520                                               /*command*/ 1,
7521                                               /*field*/ 2,
7522                                               /*bit_valid*/ 1,
7523                                               /*bit*/ 2);
7524                         ctl_done((union ctl_io *)ctsio);
7525                         return (CTL_RETVAL_COMPLETE);
7526                 }
7527                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7528                 break;
7529         case RSO_OPTIONS_OC_SA:
7530                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7531                     service_action >= 32) {
7532                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7533                                               /*sks_valid*/ 1,
7534                                               /*command*/ 1,
7535                                               /*field*/ 2,
7536                                               /*bit_valid*/ 1,
7537                                               /*bit*/ 2);
7538                         ctl_done((union ctl_io *)ctsio);
7539                         return (CTL_RETVAL_COMPLETE);
7540                 }
7541                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7542                 break;
7543         default:
7544                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7545                                       /*sks_valid*/ 1,
7546                                       /*command*/ 1,
7547                                       /*field*/ 2,
7548                                       /*bit_valid*/ 1,
7549                                       /*bit*/ 2);
7550                 ctl_done((union ctl_io *)ctsio);
7551                 return (CTL_RETVAL_COMPLETE);
7552         }
7553
7554         alloc_len = scsi_4btoul(cdb->length);
7555
7556         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7557
7558         ctsio->kern_sg_entries = 0;
7559
7560         if (total_len < alloc_len) {
7561                 ctsio->residual = alloc_len - total_len;
7562                 ctsio->kern_data_len = total_len;
7563                 ctsio->kern_total_len = total_len;
7564         } else {
7565                 ctsio->residual = 0;
7566                 ctsio->kern_data_len = alloc_len;
7567                 ctsio->kern_total_len = alloc_len;
7568         }
7569         ctsio->kern_data_resid = 0;
7570         ctsio->kern_rel_offset = 0;
7571
7572         switch (cdb->options & RSO_OPTIONS_MASK) {
7573         case RSO_OPTIONS_ALL:
7574                 all = (struct scsi_report_supported_opcodes_all *)
7575                     ctsio->kern_data_ptr;
7576                 num = 0;
7577                 for (i = 0; i < 256; i++) {
7578                         entry = &ctl_cmd_table[i];
7579                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7580                                 for (j = 0; j < 32; j++) {
7581                                         sentry = &((const struct ctl_cmd_entry *)
7582                                             entry->execute)[j];
7583                                         if (!ctl_cmd_applicable(
7584                                             lun->be_lun->lun_type, sentry))
7585                                                 continue;
7586                                         descr = &all->descr[num++];
7587                                         descr->opcode = i;
7588                                         scsi_ulto2b(j, descr->service_action);
7589                                         descr->flags = RSO_SERVACTV;
7590                                         scsi_ulto2b(sentry->length,
7591                                             descr->cdb_length);
7592                                 }
7593                         } else {
7594                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7595                                     entry))
7596                                         continue;
7597                                 descr = &all->descr[num++];
7598                                 descr->opcode = i;
7599                                 scsi_ulto2b(0, descr->service_action);
7600                                 descr->flags = 0;
7601                                 scsi_ulto2b(entry->length, descr->cdb_length);
7602                         }
7603                 }
7604                 scsi_ulto4b(
7605                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7606                     all->length);
7607                 break;
7608         case RSO_OPTIONS_OC:
7609                 one = (struct scsi_report_supported_opcodes_one *)
7610                     ctsio->kern_data_ptr;
7611                 entry = &ctl_cmd_table[opcode];
7612                 goto fill_one;
7613         case RSO_OPTIONS_OC_SA:
7614                 one = (struct scsi_report_supported_opcodes_one *)
7615                     ctsio->kern_data_ptr;
7616                 entry = &ctl_cmd_table[opcode];
7617                 entry = &((const struct ctl_cmd_entry *)
7618                     entry->execute)[service_action];
7619 fill_one:
7620                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7621                         one->support = 3;
7622                         scsi_ulto2b(entry->length, one->cdb_length);
7623                         one->cdb_usage[0] = opcode;
7624                         memcpy(&one->cdb_usage[1], entry->usage,
7625                             entry->length - 1);
7626                 } else
7627                         one->support = 1;
7628                 break;
7629         }
7630
7631         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7632         ctsio->be_move_done = ctl_config_move_done;
7633
7634         ctl_datamove((union ctl_io *)ctsio);
7635         return(retval);
7636 }
7637
7638 int
7639 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7640 {
7641         struct ctl_lun *lun;
7642         struct scsi_report_supported_tmf *cdb;
7643         struct scsi_report_supported_tmf_data *data;
7644         int retval;
7645         int alloc_len, total_len;
7646
7647         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7648
7649         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7650         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7651
7652         retval = CTL_RETVAL_COMPLETE;
7653
7654         total_len = sizeof(struct scsi_report_supported_tmf_data);
7655         alloc_len = scsi_4btoul(cdb->length);
7656
7657         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7658
7659         ctsio->kern_sg_entries = 0;
7660
7661         if (total_len < alloc_len) {
7662                 ctsio->residual = alloc_len - total_len;
7663                 ctsio->kern_data_len = total_len;
7664                 ctsio->kern_total_len = total_len;
7665         } else {
7666                 ctsio->residual = 0;
7667                 ctsio->kern_data_len = alloc_len;
7668                 ctsio->kern_total_len = alloc_len;
7669         }
7670         ctsio->kern_data_resid = 0;
7671         ctsio->kern_rel_offset = 0;
7672
7673         data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7674         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7675         data->byte2 |= RST_ITNRS;
7676
7677         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7678         ctsio->be_move_done = ctl_config_move_done;
7679
7680         ctl_datamove((union ctl_io *)ctsio);
7681         return (retval);
7682 }
7683
7684 int
7685 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7686 {
7687         struct ctl_lun *lun;
7688         struct scsi_report_timestamp *cdb;
7689         struct scsi_report_timestamp_data *data;
7690         struct timeval tv;
7691         int64_t timestamp;
7692         int retval;
7693         int alloc_len, total_len;
7694
7695         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7696
7697         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7698         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7699
7700         retval = CTL_RETVAL_COMPLETE;
7701
7702         total_len = sizeof(struct scsi_report_timestamp_data);
7703         alloc_len = scsi_4btoul(cdb->length);
7704
7705         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7706
7707         ctsio->kern_sg_entries = 0;
7708
7709         if (total_len < alloc_len) {
7710                 ctsio->residual = alloc_len - total_len;
7711                 ctsio->kern_data_len = total_len;
7712                 ctsio->kern_total_len = total_len;
7713         } else {
7714                 ctsio->residual = 0;
7715                 ctsio->kern_data_len = alloc_len;
7716                 ctsio->kern_total_len = alloc_len;
7717         }
7718         ctsio->kern_data_resid = 0;
7719         ctsio->kern_rel_offset = 0;
7720
7721         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7722         scsi_ulto2b(sizeof(*data) - 2, data->length);
7723         data->origin = RTS_ORIG_OUTSIDE;
7724         getmicrotime(&tv);
7725         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7726         scsi_ulto4b(timestamp >> 16, data->timestamp);
7727         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7728
7729         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7730         ctsio->be_move_done = ctl_config_move_done;
7731
7732         ctl_datamove((union ctl_io *)ctsio);
7733         return (retval);
7734 }
7735
7736 int
7737 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7738 {
7739         struct scsi_per_res_in *cdb;
7740         int alloc_len, total_len = 0;
7741         /* struct scsi_per_res_in_rsrv in_data; */
7742         struct ctl_lun *lun;
7743         struct ctl_softc *softc;
7744
7745         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7746
7747         softc = control_softc;
7748
7749         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7750
7751         alloc_len = scsi_2btoul(cdb->length);
7752
7753         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7754
7755 retry:
7756         mtx_lock(&lun->lun_lock);
7757         switch (cdb->action) {
7758         case SPRI_RK: /* read keys */
7759                 total_len = sizeof(struct scsi_per_res_in_keys) +
7760                         lun->pr_key_count *
7761                         sizeof(struct scsi_per_res_key);
7762                 break;
7763         case SPRI_RR: /* read reservation */
7764                 if (lun->flags & CTL_LUN_PR_RESERVED)
7765                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7766                 else
7767                         total_len = sizeof(struct scsi_per_res_in_header);
7768                 break;
7769         case SPRI_RC: /* report capabilities */
7770                 total_len = sizeof(struct scsi_per_res_cap);
7771                 break;
7772         case SPRI_RS: /* read full status */
7773                 total_len = sizeof(struct scsi_per_res_in_header) +
7774                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7775                     lun->pr_key_count;
7776                 break;
7777         default:
7778                 panic("Invalid PR type %x", cdb->action);
7779         }
7780         mtx_unlock(&lun->lun_lock);
7781
7782         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7783
7784         if (total_len < alloc_len) {
7785                 ctsio->residual = alloc_len - total_len;
7786                 ctsio->kern_data_len = total_len;
7787                 ctsio->kern_total_len = total_len;
7788         } else {
7789                 ctsio->residual = 0;
7790                 ctsio->kern_data_len = alloc_len;
7791                 ctsio->kern_total_len = alloc_len;
7792         }
7793
7794         ctsio->kern_data_resid = 0;
7795         ctsio->kern_rel_offset = 0;
7796         ctsio->kern_sg_entries = 0;
7797
7798         mtx_lock(&lun->lun_lock);
7799         switch (cdb->action) {
7800         case SPRI_RK: { // read keys
7801         struct scsi_per_res_in_keys *res_keys;
7802                 int i, key_count;
7803
7804                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7805
7806                 /*
7807                  * We had to drop the lock to allocate our buffer, which
7808                  * leaves time for someone to come in with another
7809                  * persistent reservation.  (That is unlikely, though,
7810                  * since this should be the only persistent reservation
7811                  * command active right now.)
7812                  */
7813                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7814                     (lun->pr_key_count *
7815                      sizeof(struct scsi_per_res_key)))){
7816                         mtx_unlock(&lun->lun_lock);
7817                         free(ctsio->kern_data_ptr, M_CTL);
7818                         printf("%s: reservation length changed, retrying\n",
7819                                __func__);
7820                         goto retry;
7821                 }
7822
7823                 scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7824
7825                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7826                              lun->pr_key_count, res_keys->header.length);
7827
7828                 for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7829                         if (lun->pr_keys[i] == 0)
7830                                 continue;
7831
7832                         /*
7833                          * We used lun->pr_key_count to calculate the
7834                          * size to allocate.  If it turns out the number of
7835                          * initiators with the registered flag set is
7836                          * larger than that (i.e. they haven't been kept in
7837                          * sync), we've got a problem.
7838                          */
7839                         if (key_count >= lun->pr_key_count) {
7840 #ifdef NEEDTOPORT
7841                                 csevent_log(CSC_CTL | CSC_SHELF_SW |
7842                                             CTL_PR_ERROR,
7843                                             csevent_LogType_Fault,
7844                                             csevent_AlertLevel_Yellow,
7845                                             csevent_FRU_ShelfController,
7846                                             csevent_FRU_Firmware,
7847                                         csevent_FRU_Unknown,
7848                                             "registered keys %d >= key "
7849                                             "count %d", key_count,
7850                                             lun->pr_key_count);
7851 #endif
7852                                 key_count++;
7853                                 continue;
7854                         }
7855                         scsi_u64to8b(lun->pr_keys[i],
7856                             res_keys->keys[key_count].key);
7857                         key_count++;
7858                 }
7859                 break;
7860         }
7861         case SPRI_RR: { // read reservation
7862                 struct scsi_per_res_in_rsrv *res;
7863                 int tmp_len, header_only;
7864
7865                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7866
7867                 scsi_ulto4b(lun->PRGeneration, res->header.generation);
7868
7869                 if (lun->flags & CTL_LUN_PR_RESERVED)
7870                 {
7871                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7872                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7873                                     res->header.length);
7874                         header_only = 0;
7875                 } else {
7876                         tmp_len = sizeof(struct scsi_per_res_in_header);
7877                         scsi_ulto4b(0, res->header.length);
7878                         header_only = 1;
7879                 }
7880
7881                 /*
7882                  * We had to drop the lock to allocate our buffer, which
7883                  * leaves time for someone to come in with another
7884                  * persistent reservation.  (That is unlikely, though,
7885                  * since this should be the only persistent reservation
7886                  * command active right now.)
7887                  */
7888                 if (tmp_len != total_len) {
7889                         mtx_unlock(&lun->lun_lock);
7890                         free(ctsio->kern_data_ptr, M_CTL);
7891                         printf("%s: reservation status changed, retrying\n",
7892                                __func__);
7893                         goto retry;
7894                 }
7895
7896                 /*
7897                  * No reservation held, so we're done.
7898                  */
7899                 if (header_only != 0)
7900                         break;
7901
7902                 /*
7903                  * If the registration is an All Registrants type, the key
7904                  * is 0, since it doesn't really matter.
7905                  */
7906                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7907                         scsi_u64to8b(lun->pr_keys[lun->pr_res_idx],
7908                             res->data.reservation);
7909                 }
7910                 res->data.scopetype = lun->res_type;
7911                 break;
7912         }
7913         case SPRI_RC:     //report capabilities
7914         {
7915                 struct scsi_per_res_cap *res_cap;
7916                 uint16_t type_mask;
7917
7918                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7919                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7920                 res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7921                 type_mask = SPRI_TM_WR_EX_AR |
7922                             SPRI_TM_EX_AC_RO |
7923                             SPRI_TM_WR_EX_RO |
7924                             SPRI_TM_EX_AC |
7925                             SPRI_TM_WR_EX |
7926                             SPRI_TM_EX_AC_AR;
7927                 scsi_ulto2b(type_mask, res_cap->type_mask);
7928                 break;
7929         }
7930         case SPRI_RS: { // read full status
7931                 struct scsi_per_res_in_full *res_status;
7932                 struct scsi_per_res_in_full_desc *res_desc;
7933                 struct ctl_port *port;
7934                 int i, len;
7935
7936                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7937
7938                 /*
7939                  * We had to drop the lock to allocate our buffer, which
7940                  * leaves time for someone to come in with another
7941                  * persistent reservation.  (That is unlikely, though,
7942                  * since this should be the only persistent reservation
7943                  * command active right now.)
7944                  */
7945                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7946                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7947                      lun->pr_key_count)){
7948                         mtx_unlock(&lun->lun_lock);
7949                         free(ctsio->kern_data_ptr, M_CTL);
7950                         printf("%s: reservation length changed, retrying\n",
7951                                __func__);
7952                         goto retry;
7953                 }
7954
7955                 scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7956
7957                 res_desc = &res_status->desc[0];
7958                 for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7959                         if (lun->pr_keys[i] == 0)
7960                                 continue;
7961
7962                         scsi_u64to8b(lun->pr_keys[i], res_desc->res_key.key);
7963                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7964                             (lun->pr_res_idx == i ||
7965                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7966                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7967                                 res_desc->scopetype = lun->res_type;
7968                         }
7969                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7970                             res_desc->rel_trgt_port_id);
7971                         len = 0;
7972                         port = softc->ctl_ports[
7973                             ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7974                         if (port != NULL)
7975                                 len = ctl_create_iid(port,
7976                                     i % CTL_MAX_INIT_PER_PORT,
7977                                     res_desc->transport_id);
7978                         scsi_ulto4b(len, res_desc->additional_length);
7979                         res_desc = (struct scsi_per_res_in_full_desc *)
7980                             &res_desc->transport_id[len];
7981                 }
7982                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7983                     res_status->header.length);
7984                 break;
7985         }
7986         default:
7987                 /*
7988                  * This is a bug, because we just checked for this above,
7989                  * and should have returned an error.
7990                  */
7991                 panic("Invalid PR type %x", cdb->action);
7992                 break; /* NOTREACHED */
7993         }
7994         mtx_unlock(&lun->lun_lock);
7995
7996         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7997         ctsio->be_move_done = ctl_config_move_done;
7998
7999         CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
8000                          ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
8001                          ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
8002                          ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
8003                          ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
8004
8005         ctl_datamove((union ctl_io *)ctsio);
8006
8007         return (CTL_RETVAL_COMPLETE);
8008 }
8009
8010 /*
8011  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
8012  * it should return.
8013  */
8014 static int
8015 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
8016                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
8017                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
8018                 struct scsi_per_res_out_parms* param)
8019 {
8020         union ctl_ha_msg persis_io;
8021         int retval, i;
8022         int isc_retval;
8023
8024         retval = 0;
8025
8026         mtx_lock(&lun->lun_lock);
8027         if (sa_res_key == 0) {
8028                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8029                         /* validate scope and type */
8030                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8031                              SPR_LU_SCOPE) {
8032                                 mtx_unlock(&lun->lun_lock);
8033                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8034                                                       /*sks_valid*/ 1,
8035                                                       /*command*/ 1,
8036                                                       /*field*/ 2,
8037                                                       /*bit_valid*/ 1,
8038                                                       /*bit*/ 4);
8039                                 ctl_done((union ctl_io *)ctsio);
8040                                 return (1);
8041                         }
8042
8043                         if (type>8 || type==2 || type==4 || type==0) {
8044                                 mtx_unlock(&lun->lun_lock);
8045                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8046                                                       /*sks_valid*/ 1,
8047                                                       /*command*/ 1,
8048                                                       /*field*/ 2,
8049                                                       /*bit_valid*/ 1,
8050                                                       /*bit*/ 0);
8051                                 ctl_done((union ctl_io *)ctsio);
8052                                 return (1);
8053                         }
8054
8055                         /*
8056                          * Unregister everybody else and build UA for
8057                          * them
8058                          */
8059                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8060                                 if (i == residx || lun->pr_keys[i] == 0)
8061                                         continue;
8062
8063                                 if (!persis_offset
8064                                  && i <CTL_MAX_INITIATORS)
8065                                         lun->pending_ua[i] |=
8066                                                 CTL_UA_REG_PREEMPT;
8067                                 else if (persis_offset
8068                                       && i >= persis_offset)
8069                                         lun->pending_ua[i-persis_offset] |=
8070                                                 CTL_UA_REG_PREEMPT;
8071                                 lun->pr_keys[i] = 0;
8072                         }
8073                         lun->pr_key_count = 1;
8074                         lun->res_type = type;
8075                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8076                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8077                                 lun->pr_res_idx = residx;
8078
8079                         /* send msg to other side */
8080                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8081                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8082                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8083                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8084                         persis_io.pr.pr_info.res_type = type;
8085                         memcpy(persis_io.pr.pr_info.sa_res_key,
8086                                param->serv_act_res_key,
8087                                sizeof(param->serv_act_res_key));
8088                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8089                              &persis_io, sizeof(persis_io), 0)) >
8090                              CTL_HA_STATUS_SUCCESS) {
8091                                 printf("CTL:Persis Out error returned "
8092                                        "from ctl_ha_msg_send %d\n",
8093                                        isc_retval);
8094                         }
8095                 } else {
8096                         /* not all registrants */
8097                         mtx_unlock(&lun->lun_lock);
8098                         free(ctsio->kern_data_ptr, M_CTL);
8099                         ctl_set_invalid_field(ctsio,
8100                                               /*sks_valid*/ 1,
8101                                               /*command*/ 0,
8102                                               /*field*/ 8,
8103                                               /*bit_valid*/ 0,
8104                                               /*bit*/ 0);
8105                         ctl_done((union ctl_io *)ctsio);
8106                         return (1);
8107                 }
8108         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8109                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
8110                 int found = 0;
8111
8112                 if (res_key == sa_res_key) {
8113                         /* special case */
8114                         /*
8115                          * The spec implies this is not good but doesn't
8116                          * say what to do. There are two choices either
8117                          * generate a res conflict or check condition
8118                          * with illegal field in parameter data. Since
8119                          * that is what is done when the sa_res_key is
8120                          * zero I'll take that approach since this has
8121                          * to do with the sa_res_key.
8122                          */
8123                         mtx_unlock(&lun->lun_lock);
8124                         free(ctsio->kern_data_ptr, M_CTL);
8125                         ctl_set_invalid_field(ctsio,
8126                                               /*sks_valid*/ 1,
8127                                               /*command*/ 0,
8128                                               /*field*/ 8,
8129                                               /*bit_valid*/ 0,
8130                                               /*bit*/ 0);
8131                         ctl_done((union ctl_io *)ctsio);
8132                         return (1);
8133                 }
8134
8135                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8136                         if (lun->pr_keys[i] != sa_res_key)
8137                                 continue;
8138
8139                         found = 1;
8140                         lun->pr_keys[i] = 0;
8141                         lun->pr_key_count--;
8142
8143                         if (!persis_offset && i < CTL_MAX_INITIATORS)
8144                                 lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8145                         else if (persis_offset && i >= persis_offset)
8146                                 lun->pending_ua[i-persis_offset] |=
8147                                         CTL_UA_REG_PREEMPT;
8148                 }
8149                 if (!found) {
8150                         mtx_unlock(&lun->lun_lock);
8151                         free(ctsio->kern_data_ptr, M_CTL);
8152                         ctl_set_reservation_conflict(ctsio);
8153                         ctl_done((union ctl_io *)ctsio);
8154                         return (CTL_RETVAL_COMPLETE);
8155                 }
8156                 /* send msg to other side */
8157                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8158                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8159                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8160                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
8161                 persis_io.pr.pr_info.res_type = type;
8162                 memcpy(persis_io.pr.pr_info.sa_res_key,
8163                        param->serv_act_res_key,
8164                        sizeof(param->serv_act_res_key));
8165                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8166                      &persis_io, sizeof(persis_io), 0)) >
8167                      CTL_HA_STATUS_SUCCESS) {
8168                         printf("CTL:Persis Out error returned from "
8169                                "ctl_ha_msg_send %d\n", isc_retval);
8170                 }
8171         } else {
8172                 /* Reserved but not all registrants */
8173                 /* sa_res_key is res holder */
8174                 if (sa_res_key == lun->pr_keys[lun->pr_res_idx]) {
8175                         /* validate scope and type */
8176                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8177                              SPR_LU_SCOPE) {
8178                                 mtx_unlock(&lun->lun_lock);
8179                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8180                                                       /*sks_valid*/ 1,
8181                                                       /*command*/ 1,
8182                                                       /*field*/ 2,
8183                                                       /*bit_valid*/ 1,
8184                                                       /*bit*/ 4);
8185                                 ctl_done((union ctl_io *)ctsio);
8186                                 return (1);
8187                         }
8188
8189                         if (type>8 || type==2 || type==4 || type==0) {
8190                                 mtx_unlock(&lun->lun_lock);
8191                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8192                                                       /*sks_valid*/ 1,
8193                                                       /*command*/ 1,
8194                                                       /*field*/ 2,
8195                                                       /*bit_valid*/ 1,
8196                                                       /*bit*/ 0);
8197                                 ctl_done((union ctl_io *)ctsio);
8198                                 return (1);
8199                         }
8200
8201                         /*
8202                          * Do the following:
8203                          * if sa_res_key != res_key remove all
8204                          * registrants w/sa_res_key and generate UA
8205                          * for these registrants(Registrations
8206                          * Preempted) if it wasn't an exclusive
8207                          * reservation generate UA(Reservations
8208                          * Preempted) for all other registered nexuses
8209                          * if the type has changed. Establish the new
8210                          * reservation and holder. If res_key and
8211                          * sa_res_key are the same do the above
8212                          * except don't unregister the res holder.
8213                          */
8214
8215                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8216                                 if (i == residx || lun->pr_keys[i] == 0)
8217                                         continue;
8218
8219                                 if (sa_res_key == lun->pr_keys[i]) {
8220                                         lun->pr_keys[i] = 0;
8221                                         lun->pr_key_count--;
8222
8223                                         if (!persis_offset
8224                                          && i < CTL_MAX_INITIATORS)
8225                                                 lun->pending_ua[i] |=
8226                                                         CTL_UA_REG_PREEMPT;
8227                                         else if (persis_offset
8228                                               && i >= persis_offset)
8229                                                 lun->pending_ua[i-persis_offset] |=
8230                                                   CTL_UA_REG_PREEMPT;
8231                                 } else if (type != lun->res_type
8232                                         && (lun->res_type == SPR_TYPE_WR_EX_RO
8233                                          || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8234                                                 if (!persis_offset
8235                                                  && i < CTL_MAX_INITIATORS)
8236                                                         lun->pending_ua[i] |=
8237                                                         CTL_UA_RES_RELEASE;
8238                                                 else if (persis_offset
8239                                                       && i >= persis_offset)
8240                                                         lun->pending_ua[
8241                                                         i-persis_offset] |=
8242                                                         CTL_UA_RES_RELEASE;
8243                                 }
8244                         }
8245                         lun->res_type = type;
8246                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8247                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8248                                 lun->pr_res_idx = residx;
8249                         else
8250                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8251
8252                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8253                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8254                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8255                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8256                         persis_io.pr.pr_info.res_type = type;
8257                         memcpy(persis_io.pr.pr_info.sa_res_key,
8258                                param->serv_act_res_key,
8259                                sizeof(param->serv_act_res_key));
8260                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8261                              &persis_io, sizeof(persis_io), 0)) >
8262                              CTL_HA_STATUS_SUCCESS) {
8263                                 printf("CTL:Persis Out error returned "
8264                                        "from ctl_ha_msg_send %d\n",
8265                                        isc_retval);
8266                         }
8267                 } else {
8268                         /*
8269                          * sa_res_key is not the res holder just
8270                          * remove registrants
8271                          */
8272                         int found=0;
8273
8274                         for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8275                                 if (sa_res_key != lun->pr_keys[i])
8276                                         continue;
8277
8278                                 found = 1;
8279                                 lun->pr_keys[i] = 0;
8280                                 lun->pr_key_count--;
8281
8282                                 if (!persis_offset
8283                                  && i < CTL_MAX_INITIATORS)
8284                                         lun->pending_ua[i] |=
8285                                                 CTL_UA_REG_PREEMPT;
8286                                 else if (persis_offset
8287                                       && i >= persis_offset)
8288                                         lun->pending_ua[i-persis_offset] |=
8289                                                 CTL_UA_REG_PREEMPT;
8290                         }
8291
8292                         if (!found) {
8293                                 mtx_unlock(&lun->lun_lock);
8294                                 free(ctsio->kern_data_ptr, M_CTL);
8295                                 ctl_set_reservation_conflict(ctsio);
8296                                 ctl_done((union ctl_io *)ctsio);
8297                                 return (1);
8298                         }
8299                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8300                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8301                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8302                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8303                         persis_io.pr.pr_info.res_type = type;
8304                         memcpy(persis_io.pr.pr_info.sa_res_key,
8305                                param->serv_act_res_key,
8306                                sizeof(param->serv_act_res_key));
8307                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8308                              &persis_io, sizeof(persis_io), 0)) >
8309                              CTL_HA_STATUS_SUCCESS) {
8310                                 printf("CTL:Persis Out error returned "
8311                                        "from ctl_ha_msg_send %d\n",
8312                                 isc_retval);
8313                         }
8314                 }
8315         }
8316
8317         lun->PRGeneration++;
8318         mtx_unlock(&lun->lun_lock);
8319
8320         return (retval);
8321 }
8322
8323 static void
8324 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8325 {
8326         uint64_t sa_res_key;
8327         int i;
8328
8329         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8330
8331         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8332          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8333          || sa_res_key != lun->pr_keys[lun->pr_res_idx]) {
8334                 if (sa_res_key == 0) {
8335                         /*
8336                          * Unregister everybody else and build UA for
8337                          * them
8338                          */
8339                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8340                                 if (i == msg->pr.pr_info.residx ||
8341                                     lun->pr_keys[i] == 0)
8342                                         continue;
8343
8344                                 if (!persis_offset
8345                                  && i < CTL_MAX_INITIATORS)
8346                                         lun->pending_ua[i] |=
8347                                                 CTL_UA_REG_PREEMPT;
8348                                 else if (persis_offset && i >= persis_offset)
8349                                         lun->pending_ua[i - persis_offset] |=
8350                                                 CTL_UA_REG_PREEMPT;
8351                                 lun->pr_keys[i] = 0;
8352                         }
8353
8354                         lun->pr_key_count = 1;
8355                         lun->res_type = msg->pr.pr_info.res_type;
8356                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8357                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8358                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8359                 } else {
8360                         for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8361                                 if (sa_res_key == lun->pr_keys[i])
8362                                         continue;
8363
8364                                 lun->pr_keys[i] = 0;
8365                                 lun->pr_key_count--;
8366
8367                                 if (!persis_offset
8368                                  && i < persis_offset)
8369                                         lun->pending_ua[i] |=
8370                                                 CTL_UA_REG_PREEMPT;
8371                                 else if (persis_offset
8372                                       && i >= persis_offset)
8373                                         lun->pending_ua[i - persis_offset] |=
8374                                                 CTL_UA_REG_PREEMPT;
8375                         }
8376                 }
8377         } else {
8378                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8379                         if (i == msg->pr.pr_info.residx ||
8380                             lun->pr_keys[i] == 0)
8381                                 continue;
8382
8383                         if (sa_res_key == lun->pr_keys[i]) {
8384                                 lun->pr_keys[i] = 0;
8385                                 lun->pr_key_count--;
8386                                 if (!persis_offset
8387                                  && i < CTL_MAX_INITIATORS)
8388                                         lun->pending_ua[i] |=
8389                                                 CTL_UA_REG_PREEMPT;
8390                                 else if (persis_offset
8391                                       && i >= persis_offset)
8392                                         lun->pending_ua[i - persis_offset] |=
8393                                                 CTL_UA_REG_PREEMPT;
8394                         } else if (msg->pr.pr_info.res_type != lun->res_type
8395                                 && (lun->res_type == SPR_TYPE_WR_EX_RO
8396                                  || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8397                                         if (!persis_offset
8398                                          && i < persis_offset)
8399                                                 lun->pending_ua[i] |=
8400                                                         CTL_UA_RES_RELEASE;
8401                                         else if (persis_offset
8402                                               && i >= persis_offset)
8403                                         lun->pending_ua[i - persis_offset] |=
8404                                                 CTL_UA_RES_RELEASE;
8405                         }
8406                 }
8407                 lun->res_type = msg->pr.pr_info.res_type;
8408                 if (lun->res_type != SPR_TYPE_WR_EX_AR
8409                  && lun->res_type != SPR_TYPE_EX_AC_AR)
8410                         lun->pr_res_idx = msg->pr.pr_info.residx;
8411                 else
8412                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8413         }
8414         lun->PRGeneration++;
8415
8416 }
8417
8418
8419 int
8420 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8421 {
8422         int retval;
8423         int isc_retval;
8424         u_int32_t param_len;
8425         struct scsi_per_res_out *cdb;
8426         struct ctl_lun *lun;
8427         struct scsi_per_res_out_parms* param;
8428         struct ctl_softc *softc;
8429         uint32_t residx;
8430         uint64_t res_key, sa_res_key;
8431         uint8_t type;
8432         union ctl_ha_msg persis_io;
8433         int    i;
8434
8435         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8436
8437         retval = CTL_RETVAL_COMPLETE;
8438
8439         softc = control_softc;
8440
8441         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8442         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8443
8444         /*
8445          * We only support whole-LUN scope.  The scope & type are ignored for
8446          * register, register and ignore existing key and clear.
8447          * We sometimes ignore scope and type on preempts too!!
8448          * Verify reservation type here as well.
8449          */
8450         type = cdb->scope_type & SPR_TYPE_MASK;
8451         if ((cdb->action == SPRO_RESERVE)
8452          || (cdb->action == SPRO_RELEASE)) {
8453                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8454                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8455                                               /*sks_valid*/ 1,
8456                                               /*command*/ 1,
8457                                               /*field*/ 2,
8458                                               /*bit_valid*/ 1,
8459                                               /*bit*/ 4);
8460                         ctl_done((union ctl_io *)ctsio);
8461                         return (CTL_RETVAL_COMPLETE);
8462                 }
8463
8464                 if (type>8 || type==2 || type==4 || type==0) {
8465                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8466                                               /*sks_valid*/ 1,
8467                                               /*command*/ 1,
8468                                               /*field*/ 2,
8469                                               /*bit_valid*/ 1,
8470                                               /*bit*/ 0);
8471                         ctl_done((union ctl_io *)ctsio);
8472                         return (CTL_RETVAL_COMPLETE);
8473                 }
8474         }
8475
8476         param_len = scsi_4btoul(cdb->length);
8477
8478         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8479                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8480                 ctsio->kern_data_len = param_len;
8481                 ctsio->kern_total_len = param_len;
8482                 ctsio->kern_data_resid = 0;
8483                 ctsio->kern_rel_offset = 0;
8484                 ctsio->kern_sg_entries = 0;
8485                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8486                 ctsio->be_move_done = ctl_config_move_done;
8487                 ctl_datamove((union ctl_io *)ctsio);
8488
8489                 return (CTL_RETVAL_COMPLETE);
8490         }
8491
8492         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8493
8494         residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8495         res_key = scsi_8btou64(param->res_key.key);
8496         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8497
8498         /*
8499          * Validate the reservation key here except for SPRO_REG_IGNO
8500          * This must be done for all other service actions
8501          */
8502         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8503                 mtx_lock(&lun->lun_lock);
8504                 if (lun->pr_keys[residx] != 0) {
8505                     if (res_key != lun->pr_keys[residx]) {
8506                                 /*
8507                                  * The current key passed in doesn't match
8508                                  * the one the initiator previously
8509                                  * registered.
8510                                  */
8511                                 mtx_unlock(&lun->lun_lock);
8512                                 free(ctsio->kern_data_ptr, M_CTL);
8513                                 ctl_set_reservation_conflict(ctsio);
8514                                 ctl_done((union ctl_io *)ctsio);
8515                                 return (CTL_RETVAL_COMPLETE);
8516                         }
8517                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8518                         /*
8519                          * We are not registered
8520                          */
8521                         mtx_unlock(&lun->lun_lock);
8522                         free(ctsio->kern_data_ptr, M_CTL);
8523                         ctl_set_reservation_conflict(ctsio);
8524                         ctl_done((union ctl_io *)ctsio);
8525                         return (CTL_RETVAL_COMPLETE);
8526                 } else if (res_key != 0) {
8527                         /*
8528                          * We are not registered and trying to register but
8529                          * the register key isn't zero.
8530                          */
8531                         mtx_unlock(&lun->lun_lock);
8532                         free(ctsio->kern_data_ptr, M_CTL);
8533                         ctl_set_reservation_conflict(ctsio);
8534                         ctl_done((union ctl_io *)ctsio);
8535                         return (CTL_RETVAL_COMPLETE);
8536                 }
8537                 mtx_unlock(&lun->lun_lock);
8538         }
8539
8540         switch (cdb->action & SPRO_ACTION_MASK) {
8541         case SPRO_REGISTER:
8542         case SPRO_REG_IGNO: {
8543
8544 #if 0
8545                 printf("Registration received\n");
8546 #endif
8547
8548                 /*
8549                  * We don't support any of these options, as we report in
8550                  * the read capabilities request (see
8551                  * ctl_persistent_reserve_in(), above).
8552                  */
8553                 if ((param->flags & SPR_SPEC_I_PT)
8554                  || (param->flags & SPR_ALL_TG_PT)
8555                  || (param->flags & SPR_APTPL)) {
8556                         int bit_ptr;
8557
8558                         if (param->flags & SPR_APTPL)
8559                                 bit_ptr = 0;
8560                         else if (param->flags & SPR_ALL_TG_PT)
8561                                 bit_ptr = 2;
8562                         else /* SPR_SPEC_I_PT */
8563                                 bit_ptr = 3;
8564
8565                         free(ctsio->kern_data_ptr, M_CTL);
8566                         ctl_set_invalid_field(ctsio,
8567                                               /*sks_valid*/ 1,
8568                                               /*command*/ 0,
8569                                               /*field*/ 20,
8570                                               /*bit_valid*/ 1,
8571                                               /*bit*/ bit_ptr);
8572                         ctl_done((union ctl_io *)ctsio);
8573                         return (CTL_RETVAL_COMPLETE);
8574                 }
8575
8576                 mtx_lock(&lun->lun_lock);
8577
8578                 /*
8579                  * The initiator wants to clear the
8580                  * key/unregister.
8581                  */
8582                 if (sa_res_key == 0) {
8583                         if ((res_key == 0
8584                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8585                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8586                           && lun->pr_keys[residx] == 0)) {
8587                                 mtx_unlock(&lun->lun_lock);
8588                                 goto done;
8589                         }
8590
8591                         lun->pr_keys[residx] = 0;
8592                         lun->pr_key_count--;
8593
8594                         if (residx == lun->pr_res_idx) {
8595                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8596                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8597
8598                                 if ((lun->res_type == SPR_TYPE_WR_EX_RO
8599                                   || lun->res_type == SPR_TYPE_EX_AC_RO)
8600                                  && lun->pr_key_count) {
8601                                         /*
8602                                          * If the reservation is a registrants
8603                                          * only type we need to generate a UA
8604                                          * for other registered inits.  The
8605                                          * sense code should be RESERVATIONS
8606                                          * RELEASED
8607                                          */
8608
8609                                         for (i = 0; i < CTL_MAX_INITIATORS;i++){
8610                                                 if (lun->pr_keys[
8611                                                     i + persis_offset] == 0)
8612                                                         continue;
8613                                                 lun->pending_ua[i] |=
8614                                                         CTL_UA_RES_RELEASE;
8615                                         }
8616                                 }
8617                                 lun->res_type = 0;
8618                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8619                                 if (lun->pr_key_count==0) {
8620                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8621                                         lun->res_type = 0;
8622                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8623                                 }
8624                         }
8625                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8626                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8627                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8628                         persis_io.pr.pr_info.residx = residx;
8629                         if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8630                              &persis_io, sizeof(persis_io), 0 )) >
8631                              CTL_HA_STATUS_SUCCESS) {
8632                                 printf("CTL:Persis Out error returned from "
8633                                        "ctl_ha_msg_send %d\n", isc_retval);
8634                         }
8635                 } else /* sa_res_key != 0 */ {
8636
8637                         /*
8638                          * If we aren't registered currently then increment
8639                          * the key count and set the registered flag.
8640                          */
8641                         if (lun->pr_keys[residx] == 0)
8642                                 lun->pr_key_count++;
8643                         lun->pr_keys[residx] = sa_res_key;
8644
8645                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8646                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8647                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8648                         persis_io.pr.pr_info.residx = residx;
8649                         memcpy(persis_io.pr.pr_info.sa_res_key,
8650                                param->serv_act_res_key,
8651                                sizeof(param->serv_act_res_key));
8652                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8653                              &persis_io, sizeof(persis_io), 0)) >
8654                              CTL_HA_STATUS_SUCCESS) {
8655                                 printf("CTL:Persis Out error returned from "
8656                                        "ctl_ha_msg_send %d\n", isc_retval);
8657                         }
8658                 }
8659                 lun->PRGeneration++;
8660                 mtx_unlock(&lun->lun_lock);
8661
8662                 break;
8663         }
8664         case SPRO_RESERVE:
8665 #if 0
8666                 printf("Reserve executed type %d\n", type);
8667 #endif
8668                 mtx_lock(&lun->lun_lock);
8669                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8670                         /*
8671                          * if this isn't the reservation holder and it's
8672                          * not a "all registrants" type or if the type is
8673                          * different then we have a conflict
8674                          */
8675                         if ((lun->pr_res_idx != residx
8676                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8677                          || lun->res_type != type) {
8678                                 mtx_unlock(&lun->lun_lock);
8679                                 free(ctsio->kern_data_ptr, M_CTL);
8680                                 ctl_set_reservation_conflict(ctsio);
8681                                 ctl_done((union ctl_io *)ctsio);
8682                                 return (CTL_RETVAL_COMPLETE);
8683                         }
8684                         mtx_unlock(&lun->lun_lock);
8685                 } else /* create a reservation */ {
8686                         /*
8687                          * If it's not an "all registrants" type record
8688                          * reservation holder
8689                          */
8690                         if (type != SPR_TYPE_WR_EX_AR
8691                          && type != SPR_TYPE_EX_AC_AR)
8692                                 lun->pr_res_idx = residx; /* Res holder */
8693                         else
8694                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8695
8696                         lun->flags |= CTL_LUN_PR_RESERVED;
8697                         lun->res_type = type;
8698
8699                         mtx_unlock(&lun->lun_lock);
8700
8701                         /* send msg to other side */
8702                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8703                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8704                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8705                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8706                         persis_io.pr.pr_info.res_type = type;
8707                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8708                              &persis_io, sizeof(persis_io), 0)) >
8709                              CTL_HA_STATUS_SUCCESS) {
8710                                 printf("CTL:Persis Out error returned from "
8711                                        "ctl_ha_msg_send %d\n", isc_retval);
8712                         }
8713                 }
8714                 break;
8715
8716         case SPRO_RELEASE:
8717                 mtx_lock(&lun->lun_lock);
8718                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8719                         /* No reservation exists return good status */
8720                         mtx_unlock(&lun->lun_lock);
8721                         goto done;
8722                 }
8723                 /*
8724                  * Is this nexus a reservation holder?
8725                  */
8726                 if (lun->pr_res_idx != residx
8727                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8728                         /*
8729                          * not a res holder return good status but
8730                          * do nothing
8731                          */
8732                         mtx_unlock(&lun->lun_lock);
8733                         goto done;
8734                 }
8735
8736                 if (lun->res_type != type) {
8737                         mtx_unlock(&lun->lun_lock);
8738                         free(ctsio->kern_data_ptr, M_CTL);
8739                         ctl_set_illegal_pr_release(ctsio);
8740                         ctl_done((union ctl_io *)ctsio);
8741                         return (CTL_RETVAL_COMPLETE);
8742                 }
8743
8744                 /* okay to release */
8745                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8746                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8747                 lun->res_type = 0;
8748
8749                 /*
8750                  * if this isn't an exclusive access
8751                  * res generate UA for all other
8752                  * registrants.
8753                  */
8754                 if (type != SPR_TYPE_EX_AC
8755                  && type != SPR_TYPE_WR_EX) {
8756                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8757                                 if (i == residx ||
8758                                     lun->pr_keys[i + persis_offset] == 0)
8759                                         continue;
8760                                 lun->pending_ua[i] |= CTL_UA_RES_RELEASE;
8761                         }
8762                 }
8763                 mtx_unlock(&lun->lun_lock);
8764                 /* Send msg to other side */
8765                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8766                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8767                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8768                 if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8769                      sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8770                         printf("CTL:Persis Out error returned from "
8771                                "ctl_ha_msg_send %d\n", isc_retval);
8772                 }
8773                 break;
8774
8775         case SPRO_CLEAR:
8776                 /* send msg to other side */
8777
8778                 mtx_lock(&lun->lun_lock);
8779                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8780                 lun->res_type = 0;
8781                 lun->pr_key_count = 0;
8782                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8783
8784                 lun->pr_keys[residx] = 0;
8785
8786                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8787                         if (lun->pr_keys[i] != 0) {
8788                                 if (!persis_offset && i < CTL_MAX_INITIATORS)
8789                                         lun->pending_ua[i] |=
8790                                                 CTL_UA_RES_PREEMPT;
8791                                 else if (persis_offset && i >= persis_offset)
8792                                         lun->pending_ua[i-persis_offset] |=
8793                                             CTL_UA_RES_PREEMPT;
8794
8795                                 lun->pr_keys[i] = 0;
8796                         }
8797                 lun->PRGeneration++;
8798                 mtx_unlock(&lun->lun_lock);
8799                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8800                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8801                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8802                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8803                      sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8804                         printf("CTL:Persis Out error returned from "
8805                                "ctl_ha_msg_send %d\n", isc_retval);
8806                 }
8807                 break;
8808
8809         case SPRO_PREEMPT: {
8810                 int nretval;
8811
8812                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8813                                           residx, ctsio, cdb, param);
8814                 if (nretval != 0)
8815                         return (CTL_RETVAL_COMPLETE);
8816                 break;
8817         }
8818         default:
8819                 panic("Invalid PR type %x", cdb->action);
8820         }
8821
8822 done:
8823         free(ctsio->kern_data_ptr, M_CTL);
8824         ctl_set_success(ctsio);
8825         ctl_done((union ctl_io *)ctsio);
8826
8827         return (retval);
8828 }
8829
8830 /*
8831  * This routine is for handling a message from the other SC pertaining to
8832  * persistent reserve out. All the error checking will have been done
8833  * so only perorming the action need be done here to keep the two
8834  * in sync.
8835  */
8836 static void
8837 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8838 {
8839         struct ctl_lun *lun;
8840         struct ctl_softc *softc;
8841         int i;
8842         uint32_t targ_lun;
8843
8844         softc = control_softc;
8845
8846         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8847         lun = softc->ctl_luns[targ_lun];
8848         mtx_lock(&lun->lun_lock);
8849         switch(msg->pr.pr_info.action) {
8850         case CTL_PR_REG_KEY:
8851                 if (lun->pr_keys[msg->pr.pr_info.residx] == 0)
8852                         lun->pr_key_count++;
8853                 lun->pr_keys[msg->pr.pr_info.residx] =
8854                     scsi_8btou64(msg->pr.pr_info.sa_res_key);
8855                 lun->PRGeneration++;
8856                 break;
8857
8858         case CTL_PR_UNREG_KEY:
8859                 lun->pr_keys[msg->pr.pr_info.residx] = 0;
8860                 lun->pr_key_count--;
8861
8862                 /* XXX Need to see if the reservation has been released */
8863                 /* if so do we need to generate UA? */
8864                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8865                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8866                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8867
8868                         if ((lun->res_type == SPR_TYPE_WR_EX_RO
8869                           || lun->res_type == SPR_TYPE_EX_AC_RO)
8870                          && lun->pr_key_count) {
8871                                 /*
8872                                  * If the reservation is a registrants
8873                                  * only type we need to generate a UA
8874                                  * for other registered inits.  The
8875                                  * sense code should be RESERVATIONS
8876                                  * RELEASED
8877                                  */
8878
8879                                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8880                                         if (lun->pr_keys[i+
8881                                             persis_offset] == 0)
8882                                                 continue;
8883
8884                                         lun->pending_ua[i] |=
8885                                                 CTL_UA_RES_RELEASE;
8886                                 }
8887                         }
8888                         lun->res_type = 0;
8889                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8890                         if (lun->pr_key_count==0) {
8891                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8892                                 lun->res_type = 0;
8893                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8894                         }
8895                 }
8896                 lun->PRGeneration++;
8897                 break;
8898
8899         case CTL_PR_RESERVE:
8900                 lun->flags |= CTL_LUN_PR_RESERVED;
8901                 lun->res_type = msg->pr.pr_info.res_type;
8902                 lun->pr_res_idx = msg->pr.pr_info.residx;
8903
8904                 break;
8905
8906         case CTL_PR_RELEASE:
8907                 /*
8908                  * if this isn't an exclusive access res generate UA for all
8909                  * other registrants.
8910                  */
8911                 if (lun->res_type != SPR_TYPE_EX_AC
8912                  && lun->res_type != SPR_TYPE_WR_EX) {
8913                         for (i = 0; i < CTL_MAX_INITIATORS; i++)
8914                                 if (lun->pr_keys[i+persis_offset] != 0)
8915                                         lun->pending_ua[i] |=
8916                                                 CTL_UA_RES_RELEASE;
8917                 }
8918
8919                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8920                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8921                 lun->res_type = 0;
8922                 break;
8923
8924         case CTL_PR_PREEMPT:
8925                 ctl_pro_preempt_other(lun, msg);
8926                 break;
8927         case CTL_PR_CLEAR:
8928                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8929                 lun->res_type = 0;
8930                 lun->pr_key_count = 0;
8931                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8932
8933                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8934                         if (lun->pr_keys[i] == 0)
8935                                 continue;
8936                         if (!persis_offset
8937                          && i < CTL_MAX_INITIATORS)
8938                                 lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
8939                         else if (persis_offset
8940                               && i >= persis_offset)
8941                                 lun->pending_ua[i-persis_offset] |=
8942                                         CTL_UA_RES_PREEMPT;
8943                         lun->pr_keys[i] = 0;
8944                 }
8945                 lun->PRGeneration++;
8946                 break;
8947         }
8948
8949         mtx_unlock(&lun->lun_lock);
8950 }
8951
8952 int
8953 ctl_read_write(struct ctl_scsiio *ctsio)
8954 {
8955         struct ctl_lun *lun;
8956         struct ctl_lba_len_flags *lbalen;
8957         uint64_t lba;
8958         uint32_t num_blocks;
8959         int flags, retval;
8960         int isread;
8961
8962         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8963
8964         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8965
8966         flags = 0;
8967         retval = CTL_RETVAL_COMPLETE;
8968
8969         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8970               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8971         if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8972                 uint32_t residx;
8973
8974                 /*
8975                  * XXX KDM need a lock here.
8976                  */
8977                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8978                 if ((lun->res_type == SPR_TYPE_EX_AC
8979                   && residx != lun->pr_res_idx)
8980                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
8981                    || lun->res_type == SPR_TYPE_EX_AC_AR)
8982                   && lun->pr_keys[residx] == 0)) {
8983                         ctl_set_reservation_conflict(ctsio);
8984                         ctl_done((union ctl_io *)ctsio);
8985                         return (CTL_RETVAL_COMPLETE);
8986                 }
8987         }
8988
8989         switch (ctsio->cdb[0]) {
8990         case READ_6:
8991         case WRITE_6: {
8992                 struct scsi_rw_6 *cdb;
8993
8994                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8995
8996                 lba = scsi_3btoul(cdb->addr);
8997                 /* only 5 bits are valid in the most significant address byte */
8998                 lba &= 0x1fffff;
8999                 num_blocks = cdb->length;
9000                 /*
9001                  * This is correct according to SBC-2.
9002                  */
9003                 if (num_blocks == 0)
9004                         num_blocks = 256;
9005                 break;
9006         }
9007         case READ_10:
9008         case WRITE_10: {
9009                 struct scsi_rw_10 *cdb;
9010
9011                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
9012                 if (cdb->byte2 & SRW10_FUA)
9013                         flags |= CTL_LLF_FUA;
9014                 if (cdb->byte2 & SRW10_DPO)
9015                         flags |= CTL_LLF_DPO;
9016                 lba = scsi_4btoul(cdb->addr);
9017                 num_blocks = scsi_2btoul(cdb->length);
9018                 break;
9019         }
9020         case WRITE_VERIFY_10: {
9021                 struct scsi_write_verify_10 *cdb;
9022
9023                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9024                 flags |= CTL_LLF_FUA;
9025                 if (cdb->byte2 & SWV_DPO)
9026                         flags |= CTL_LLF_DPO;
9027                 lba = scsi_4btoul(cdb->addr);
9028                 num_blocks = scsi_2btoul(cdb->length);
9029                 break;
9030         }
9031         case READ_12:
9032         case WRITE_12: {
9033                 struct scsi_rw_12 *cdb;
9034
9035                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
9036                 if (cdb->byte2 & SRW12_FUA)
9037                         flags |= CTL_LLF_FUA;
9038                 if (cdb->byte2 & SRW12_DPO)
9039                         flags |= CTL_LLF_DPO;
9040                 lba = scsi_4btoul(cdb->addr);
9041                 num_blocks = scsi_4btoul(cdb->length);
9042                 break;
9043         }
9044         case WRITE_VERIFY_12: {
9045                 struct scsi_write_verify_12 *cdb;
9046
9047                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9048                 flags |= CTL_LLF_FUA;
9049                 if (cdb->byte2 & SWV_DPO)
9050                         flags |= CTL_LLF_DPO;
9051                 lba = scsi_4btoul(cdb->addr);
9052                 num_blocks = scsi_4btoul(cdb->length);
9053                 break;
9054         }
9055         case READ_16:
9056         case WRITE_16: {
9057                 struct scsi_rw_16 *cdb;
9058
9059                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9060                 if (cdb->byte2 & SRW12_FUA)
9061                         flags |= CTL_LLF_FUA;
9062                 if (cdb->byte2 & SRW12_DPO)
9063                         flags |= CTL_LLF_DPO;
9064                 lba = scsi_8btou64(cdb->addr);
9065                 num_blocks = scsi_4btoul(cdb->length);
9066                 break;
9067         }
9068         case WRITE_ATOMIC_16: {
9069                 struct scsi_rw_16 *cdb;
9070
9071                 if (lun->be_lun->atomicblock == 0) {
9072                         ctl_set_invalid_opcode(ctsio);
9073                         ctl_done((union ctl_io *)ctsio);
9074                         return (CTL_RETVAL_COMPLETE);
9075                 }
9076
9077                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9078                 if (cdb->byte2 & SRW12_FUA)
9079                         flags |= CTL_LLF_FUA;
9080                 if (cdb->byte2 & SRW12_DPO)
9081                         flags |= CTL_LLF_DPO;
9082                 lba = scsi_8btou64(cdb->addr);
9083                 num_blocks = scsi_4btoul(cdb->length);
9084                 if (num_blocks > lun->be_lun->atomicblock) {
9085                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
9086                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
9087                             /*bit*/ 0);
9088                         ctl_done((union ctl_io *)ctsio);
9089                         return (CTL_RETVAL_COMPLETE);
9090                 }
9091                 break;
9092         }
9093         case WRITE_VERIFY_16: {
9094                 struct scsi_write_verify_16 *cdb;
9095
9096                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9097                 flags |= CTL_LLF_FUA;
9098                 if (cdb->byte2 & SWV_DPO)
9099                         flags |= CTL_LLF_DPO;
9100                 lba = scsi_8btou64(cdb->addr);
9101                 num_blocks = scsi_4btoul(cdb->length);
9102                 break;
9103         }
9104         default:
9105                 /*
9106                  * We got a command we don't support.  This shouldn't
9107                  * happen, commands should be filtered out above us.
9108                  */
9109                 ctl_set_invalid_opcode(ctsio);
9110                 ctl_done((union ctl_io *)ctsio);
9111
9112                 return (CTL_RETVAL_COMPLETE);
9113                 break; /* NOTREACHED */
9114         }
9115
9116         /*
9117          * The first check is to make sure we're in bounds, the second
9118          * check is to catch wrap-around problems.  If the lba + num blocks
9119          * is less than the lba, then we've wrapped around and the block
9120          * range is invalid anyway.
9121          */
9122         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9123          || ((lba + num_blocks) < lba)) {
9124                 ctl_set_lba_out_of_range(ctsio);
9125                 ctl_done((union ctl_io *)ctsio);
9126                 return (CTL_RETVAL_COMPLETE);
9127         }
9128
9129         /*
9130          * According to SBC-3, a transfer length of 0 is not an error.
9131          * Note that this cannot happen with WRITE(6) or READ(6), since 0
9132          * translates to 256 blocks for those commands.
9133          */
9134         if (num_blocks == 0) {
9135                 ctl_set_success(ctsio);
9136                 ctl_done((union ctl_io *)ctsio);
9137                 return (CTL_RETVAL_COMPLETE);
9138         }
9139
9140         /* Set FUA and/or DPO if caches are disabled. */
9141         if (isread) {
9142                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9143                     SCP_RCD) != 0)
9144                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9145         } else {
9146                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9147                     SCP_WCE) == 0)
9148                         flags |= CTL_LLF_FUA;
9149         }
9150
9151         lbalen = (struct ctl_lba_len_flags *)
9152             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9153         lbalen->lba = lba;
9154         lbalen->len = num_blocks;
9155         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9156
9157         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9158         ctsio->kern_rel_offset = 0;
9159
9160         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9161
9162         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9163
9164         return (retval);
9165 }
9166
9167 static int
9168 ctl_cnw_cont(union ctl_io *io)
9169 {
9170         struct ctl_scsiio *ctsio;
9171         struct ctl_lun *lun;
9172         struct ctl_lba_len_flags *lbalen;
9173         int retval;
9174
9175         ctsio = &io->scsiio;
9176         ctsio->io_hdr.status = CTL_STATUS_NONE;
9177         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9178         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9179         lbalen = (struct ctl_lba_len_flags *)
9180             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9181         lbalen->flags &= ~CTL_LLF_COMPARE;
9182         lbalen->flags |= CTL_LLF_WRITE;
9183
9184         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9185         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9186         return (retval);
9187 }
9188
9189 int
9190 ctl_cnw(struct ctl_scsiio *ctsio)
9191 {
9192         struct ctl_lun *lun;
9193         struct ctl_lba_len_flags *lbalen;
9194         uint64_t lba;
9195         uint32_t num_blocks;
9196         int flags, retval;
9197
9198         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9199
9200         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9201
9202         flags = 0;
9203         retval = CTL_RETVAL_COMPLETE;
9204
9205         switch (ctsio->cdb[0]) {
9206         case COMPARE_AND_WRITE: {
9207                 struct scsi_compare_and_write *cdb;
9208
9209                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9210                 if (cdb->byte2 & SRW10_FUA)
9211                         flags |= CTL_LLF_FUA;
9212                 if (cdb->byte2 & SRW10_DPO)
9213                         flags |= CTL_LLF_DPO;
9214                 lba = scsi_8btou64(cdb->addr);
9215                 num_blocks = cdb->length;
9216                 break;
9217         }
9218         default:
9219                 /*
9220                  * We got a command we don't support.  This shouldn't
9221                  * happen, commands should be filtered out above us.
9222                  */
9223                 ctl_set_invalid_opcode(ctsio);
9224                 ctl_done((union ctl_io *)ctsio);
9225
9226                 return (CTL_RETVAL_COMPLETE);
9227                 break; /* NOTREACHED */
9228         }
9229
9230         /*
9231          * The first check is to make sure we're in bounds, the second
9232          * check is to catch wrap-around problems.  If the lba + num blocks
9233          * is less than the lba, then we've wrapped around and the block
9234          * range is invalid anyway.
9235          */
9236         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9237          || ((lba + num_blocks) < lba)) {
9238                 ctl_set_lba_out_of_range(ctsio);
9239                 ctl_done((union ctl_io *)ctsio);
9240                 return (CTL_RETVAL_COMPLETE);
9241         }
9242
9243         /*
9244          * According to SBC-3, a transfer length of 0 is not an error.
9245          */
9246         if (num_blocks == 0) {
9247                 ctl_set_success(ctsio);
9248                 ctl_done((union ctl_io *)ctsio);
9249                 return (CTL_RETVAL_COMPLETE);
9250         }
9251
9252         /* Set FUA if write cache is disabled. */
9253         if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9254             SCP_WCE) == 0)
9255                 flags |= CTL_LLF_FUA;
9256
9257         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9258         ctsio->kern_rel_offset = 0;
9259
9260         /*
9261          * Set the IO_CONT flag, so that if this I/O gets passed to
9262          * ctl_data_submit_done(), it'll get passed back to
9263          * ctl_ctl_cnw_cont() for further processing.
9264          */
9265         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9266         ctsio->io_cont = ctl_cnw_cont;
9267
9268         lbalen = (struct ctl_lba_len_flags *)
9269             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9270         lbalen->lba = lba;
9271         lbalen->len = num_blocks;
9272         lbalen->flags = CTL_LLF_COMPARE | flags;
9273
9274         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9275         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9276         return (retval);
9277 }
9278
9279 int
9280 ctl_verify(struct ctl_scsiio *ctsio)
9281 {
9282         struct ctl_lun *lun;
9283         struct ctl_lba_len_flags *lbalen;
9284         uint64_t lba;
9285         uint32_t num_blocks;
9286         int bytchk, flags;
9287         int retval;
9288
9289         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9290
9291         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9292
9293         bytchk = 0;
9294         flags = CTL_LLF_FUA;
9295         retval = CTL_RETVAL_COMPLETE;
9296
9297         switch (ctsio->cdb[0]) {
9298         case VERIFY_10: {
9299                 struct scsi_verify_10 *cdb;
9300
9301                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
9302                 if (cdb->byte2 & SVFY_BYTCHK)
9303                         bytchk = 1;
9304                 if (cdb->byte2 & SVFY_DPO)
9305                         flags |= CTL_LLF_DPO;
9306                 lba = scsi_4btoul(cdb->addr);
9307                 num_blocks = scsi_2btoul(cdb->length);
9308                 break;
9309         }
9310         case VERIFY_12: {
9311                 struct scsi_verify_12 *cdb;
9312
9313                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
9314                 if (cdb->byte2 & SVFY_BYTCHK)
9315                         bytchk = 1;
9316                 if (cdb->byte2 & SVFY_DPO)
9317                         flags |= CTL_LLF_DPO;
9318                 lba = scsi_4btoul(cdb->addr);
9319                 num_blocks = scsi_4btoul(cdb->length);
9320                 break;
9321         }
9322         case VERIFY_16: {
9323                 struct scsi_rw_16 *cdb;
9324
9325                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9326                 if (cdb->byte2 & SVFY_BYTCHK)
9327                         bytchk = 1;
9328                 if (cdb->byte2 & SVFY_DPO)
9329                         flags |= CTL_LLF_DPO;
9330                 lba = scsi_8btou64(cdb->addr);
9331                 num_blocks = scsi_4btoul(cdb->length);
9332                 break;
9333         }
9334         default:
9335                 /*
9336                  * We got a command we don't support.  This shouldn't
9337                  * happen, commands should be filtered out above us.
9338                  */
9339                 ctl_set_invalid_opcode(ctsio);
9340                 ctl_done((union ctl_io *)ctsio);
9341                 return (CTL_RETVAL_COMPLETE);
9342         }
9343
9344         /*
9345          * The first check is to make sure we're in bounds, the second
9346          * check is to catch wrap-around problems.  If the lba + num blocks
9347          * is less than the lba, then we've wrapped around and the block
9348          * range is invalid anyway.
9349          */
9350         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9351          || ((lba + num_blocks) < lba)) {
9352                 ctl_set_lba_out_of_range(ctsio);
9353                 ctl_done((union ctl_io *)ctsio);
9354                 return (CTL_RETVAL_COMPLETE);
9355         }
9356
9357         /*
9358          * According to SBC-3, a transfer length of 0 is not an error.
9359          */
9360         if (num_blocks == 0) {
9361                 ctl_set_success(ctsio);
9362                 ctl_done((union ctl_io *)ctsio);
9363                 return (CTL_RETVAL_COMPLETE);
9364         }
9365
9366         lbalen = (struct ctl_lba_len_flags *)
9367             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9368         lbalen->lba = lba;
9369         lbalen->len = num_blocks;
9370         if (bytchk) {
9371                 lbalen->flags = CTL_LLF_COMPARE | flags;
9372                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9373         } else {
9374                 lbalen->flags = CTL_LLF_VERIFY | flags;
9375                 ctsio->kern_total_len = 0;
9376         }
9377         ctsio->kern_rel_offset = 0;
9378
9379         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9380         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9381         return (retval);
9382 }
9383
9384 int
9385 ctl_report_luns(struct ctl_scsiio *ctsio)
9386 {
9387         struct scsi_report_luns *cdb;
9388         struct scsi_report_luns_data *lun_data;
9389         struct ctl_lun *lun, *request_lun;
9390         int num_luns, retval;
9391         uint32_t alloc_len, lun_datalen;
9392         int num_filled, well_known;
9393         uint32_t initidx, targ_lun_id, lun_id;
9394
9395         retval = CTL_RETVAL_COMPLETE;
9396         well_known = 0;
9397
9398         cdb = (struct scsi_report_luns *)ctsio->cdb;
9399
9400         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9401
9402         mtx_lock(&control_softc->ctl_lock);
9403         num_luns = control_softc->num_luns;
9404         mtx_unlock(&control_softc->ctl_lock);
9405
9406         switch (cdb->select_report) {
9407         case RPL_REPORT_DEFAULT:
9408         case RPL_REPORT_ALL:
9409                 break;
9410         case RPL_REPORT_WELLKNOWN:
9411                 well_known = 1;
9412                 num_luns = 0;
9413                 break;
9414         default:
9415                 ctl_set_invalid_field(ctsio,
9416                                       /*sks_valid*/ 1,
9417                                       /*command*/ 1,
9418                                       /*field*/ 2,
9419                                       /*bit_valid*/ 0,
9420                                       /*bit*/ 0);
9421                 ctl_done((union ctl_io *)ctsio);
9422                 return (retval);
9423                 break; /* NOTREACHED */
9424         }
9425
9426         alloc_len = scsi_4btoul(cdb->length);
9427         /*
9428          * The initiator has to allocate at least 16 bytes for this request,
9429          * so he can at least get the header and the first LUN.  Otherwise
9430          * we reject the request (per SPC-3 rev 14, section 6.21).
9431          */
9432         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9433             sizeof(struct scsi_report_luns_lundata))) {
9434                 ctl_set_invalid_field(ctsio,
9435                                       /*sks_valid*/ 1,
9436                                       /*command*/ 1,
9437                                       /*field*/ 6,
9438                                       /*bit_valid*/ 0,
9439                                       /*bit*/ 0);
9440                 ctl_done((union ctl_io *)ctsio);
9441                 return (retval);
9442         }
9443
9444         request_lun = (struct ctl_lun *)
9445                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9446
9447         lun_datalen = sizeof(*lun_data) +
9448                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9449
9450         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9451         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9452         ctsio->kern_sg_entries = 0;
9453
9454         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9455
9456         mtx_lock(&control_softc->ctl_lock);
9457         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9458                 lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9459                 if (lun_id >= CTL_MAX_LUNS)
9460                         continue;
9461                 lun = control_softc->ctl_luns[lun_id];
9462                 if (lun == NULL)
9463                         continue;
9464
9465                 if (targ_lun_id <= 0xff) {
9466                         /*
9467                          * Peripheral addressing method, bus number 0.
9468                          */
9469                         lun_data->luns[num_filled].lundata[0] =
9470                                 RPL_LUNDATA_ATYP_PERIPH;
9471                         lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9472                         num_filled++;
9473                 } else if (targ_lun_id <= 0x3fff) {
9474                         /*
9475                          * Flat addressing method.
9476                          */
9477                         lun_data->luns[num_filled].lundata[0] =
9478                                 RPL_LUNDATA_ATYP_FLAT |
9479                                 (targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9480 #ifdef OLDCTLHEADERS
9481                                 (SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9482                                 (targ_lun_id & SRLD_BUS_LUN_MASK);
9483 #endif
9484                         lun_data->luns[num_filled].lundata[1] =
9485 #ifdef OLDCTLHEADERS
9486                                 targ_lun_id >> SRLD_BUS_LUN_BITS;
9487 #endif
9488                                 targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9489                         num_filled++;
9490                 } else {
9491                         printf("ctl_report_luns: bogus LUN number %jd, "
9492                                "skipping\n", (intmax_t)targ_lun_id);
9493                 }
9494                 /*
9495                  * According to SPC-3, rev 14 section 6.21:
9496                  *
9497                  * "The execution of a REPORT LUNS command to any valid and
9498                  * installed logical unit shall clear the REPORTED LUNS DATA
9499                  * HAS CHANGED unit attention condition for all logical
9500                  * units of that target with respect to the requesting
9501                  * initiator. A valid and installed logical unit is one
9502                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9503                  * INQUIRY data (see 6.4.2)."
9504                  *
9505                  * If request_lun is NULL, the LUN this report luns command
9506                  * was issued to is either disabled or doesn't exist. In that
9507                  * case, we shouldn't clear any pending lun change unit
9508                  * attention.
9509                  */
9510                 if (request_lun != NULL) {
9511                         mtx_lock(&lun->lun_lock);
9512                         lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9513                         mtx_unlock(&lun->lun_lock);
9514                 }
9515         }
9516         mtx_unlock(&control_softc->ctl_lock);
9517
9518         /*
9519          * It's quite possible that we've returned fewer LUNs than we allocated
9520          * space for.  Trim it.
9521          */
9522         lun_datalen = sizeof(*lun_data) +
9523                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9524
9525         if (lun_datalen < alloc_len) {
9526                 ctsio->residual = alloc_len - lun_datalen;
9527                 ctsio->kern_data_len = lun_datalen;
9528                 ctsio->kern_total_len = lun_datalen;
9529         } else {
9530                 ctsio->residual = 0;
9531                 ctsio->kern_data_len = alloc_len;
9532                 ctsio->kern_total_len = alloc_len;
9533         }
9534         ctsio->kern_data_resid = 0;
9535         ctsio->kern_rel_offset = 0;
9536         ctsio->kern_sg_entries = 0;
9537
9538         /*
9539          * We set this to the actual data length, regardless of how much
9540          * space we actually have to return results.  If the user looks at
9541          * this value, he'll know whether or not he allocated enough space
9542          * and reissue the command if necessary.  We don't support well
9543          * known logical units, so if the user asks for that, return none.
9544          */
9545         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9546
9547         /*
9548          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9549          * this request.
9550          */
9551         ctsio->scsi_status = SCSI_STATUS_OK;
9552
9553         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9554         ctsio->be_move_done = ctl_config_move_done;
9555         ctl_datamove((union ctl_io *)ctsio);
9556
9557         return (retval);
9558 }
9559
9560 int
9561 ctl_request_sense(struct ctl_scsiio *ctsio)
9562 {
9563         struct scsi_request_sense *cdb;
9564         struct scsi_sense_data *sense_ptr;
9565         struct ctl_lun *lun;
9566         uint32_t initidx;
9567         int have_error;
9568         scsi_sense_data_type sense_format;
9569
9570         cdb = (struct scsi_request_sense *)ctsio->cdb;
9571
9572         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9573
9574         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9575
9576         /*
9577          * Determine which sense format the user wants.
9578          */
9579         if (cdb->byte2 & SRS_DESC)
9580                 sense_format = SSD_TYPE_DESC;
9581         else
9582                 sense_format = SSD_TYPE_FIXED;
9583
9584         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9585         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9586         ctsio->kern_sg_entries = 0;
9587
9588         /*
9589          * struct scsi_sense_data, which is currently set to 256 bytes, is
9590          * larger than the largest allowed value for the length field in the
9591          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9592          */
9593         ctsio->residual = 0;
9594         ctsio->kern_data_len = cdb->length;
9595         ctsio->kern_total_len = cdb->length;
9596
9597         ctsio->kern_data_resid = 0;
9598         ctsio->kern_rel_offset = 0;
9599         ctsio->kern_sg_entries = 0;
9600
9601         /*
9602          * If we don't have a LUN, we don't have any pending sense.
9603          */
9604         if (lun == NULL)
9605                 goto no_sense;
9606
9607         have_error = 0;
9608         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9609         /*
9610          * Check for pending sense, and then for pending unit attentions.
9611          * Pending sense gets returned first, then pending unit attentions.
9612          */
9613         mtx_lock(&lun->lun_lock);
9614 #ifdef CTL_WITH_CA
9615         if (ctl_is_set(lun->have_ca, initidx)) {
9616                 scsi_sense_data_type stored_format;
9617
9618                 /*
9619                  * Check to see which sense format was used for the stored
9620                  * sense data.
9621                  */
9622                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9623
9624                 /*
9625                  * If the user requested a different sense format than the
9626                  * one we stored, then we need to convert it to the other
9627                  * format.  If we're going from descriptor to fixed format
9628                  * sense data, we may lose things in translation, depending
9629                  * on what options were used.
9630                  *
9631                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9632                  * for some reason we'll just copy it out as-is.
9633                  */
9634                 if ((stored_format == SSD_TYPE_FIXED)
9635                  && (sense_format == SSD_TYPE_DESC))
9636                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9637                             &lun->pending_sense[initidx],
9638                             (struct scsi_sense_data_desc *)sense_ptr);
9639                 else if ((stored_format == SSD_TYPE_DESC)
9640                       && (sense_format == SSD_TYPE_FIXED))
9641                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9642                             &lun->pending_sense[initidx],
9643                             (struct scsi_sense_data_fixed *)sense_ptr);
9644                 else
9645                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9646                                ctl_min(sizeof(*sense_ptr),
9647                                sizeof(lun->pending_sense[initidx])));
9648
9649                 ctl_clear_mask(lun->have_ca, initidx);
9650                 have_error = 1;
9651         } else
9652 #endif
9653         if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9654                 ctl_ua_type ua_type;
9655
9656                 ua_type = ctl_build_ua(&lun->pending_ua[initidx],
9657                                        sense_ptr, sense_format);
9658                 if (ua_type != CTL_UA_NONE)
9659                         have_error = 1;
9660         }
9661         mtx_unlock(&lun->lun_lock);
9662
9663         /*
9664          * We already have a pending error, return it.
9665          */
9666         if (have_error != 0) {
9667                 /*
9668                  * We report the SCSI status as OK, since the status of the
9669                  * request sense command itself is OK.
9670                  */
9671                 ctsio->scsi_status = SCSI_STATUS_OK;
9672
9673                 /*
9674                  * We report 0 for the sense length, because we aren't doing
9675                  * autosense in this case.  We're reporting sense as
9676                  * parameter data.
9677                  */
9678                 ctsio->sense_len = 0;
9679                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9680                 ctsio->be_move_done = ctl_config_move_done;
9681                 ctl_datamove((union ctl_io *)ctsio);
9682
9683                 return (CTL_RETVAL_COMPLETE);
9684         }
9685
9686 no_sense:
9687
9688         /*
9689          * No sense information to report, so we report that everything is
9690          * okay.
9691          */
9692         ctl_set_sense_data(sense_ptr,
9693                            lun,
9694                            sense_format,
9695                            /*current_error*/ 1,
9696                            /*sense_key*/ SSD_KEY_NO_SENSE,
9697                            /*asc*/ 0x00,
9698                            /*ascq*/ 0x00,
9699                            SSD_ELEM_NONE);
9700
9701         ctsio->scsi_status = SCSI_STATUS_OK;
9702
9703         /*
9704          * We report 0 for the sense length, because we aren't doing
9705          * autosense in this case.  We're reporting sense as parameter data.
9706          */
9707         ctsio->sense_len = 0;
9708         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9709         ctsio->be_move_done = ctl_config_move_done;
9710         ctl_datamove((union ctl_io *)ctsio);
9711
9712         return (CTL_RETVAL_COMPLETE);
9713 }
9714
9715 int
9716 ctl_tur(struct ctl_scsiio *ctsio)
9717 {
9718         struct ctl_lun *lun;
9719
9720         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9721
9722         CTL_DEBUG_PRINT(("ctl_tur\n"));
9723
9724         if (lun == NULL)
9725                 return (EINVAL);
9726
9727         ctsio->scsi_status = SCSI_STATUS_OK;
9728         ctsio->io_hdr.status = CTL_SUCCESS;
9729
9730         ctl_done((union ctl_io *)ctsio);
9731
9732         return (CTL_RETVAL_COMPLETE);
9733 }
9734
9735 #ifdef notyet
9736 static int
9737 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9738 {
9739
9740 }
9741 #endif
9742
9743 static int
9744 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9745 {
9746         struct scsi_vpd_supported_pages *pages;
9747         int sup_page_size;
9748         struct ctl_lun *lun;
9749
9750         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9751
9752         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9753             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9754         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9755         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9756         ctsio->kern_sg_entries = 0;
9757
9758         if (sup_page_size < alloc_len) {
9759                 ctsio->residual = alloc_len - sup_page_size;
9760                 ctsio->kern_data_len = sup_page_size;
9761                 ctsio->kern_total_len = sup_page_size;
9762         } else {
9763                 ctsio->residual = 0;
9764                 ctsio->kern_data_len = alloc_len;
9765                 ctsio->kern_total_len = alloc_len;
9766         }
9767         ctsio->kern_data_resid = 0;
9768         ctsio->kern_rel_offset = 0;
9769         ctsio->kern_sg_entries = 0;
9770
9771         /*
9772          * The control device is always connected.  The disk device, on the
9773          * other hand, may not be online all the time.  Need to change this
9774          * to figure out whether the disk device is actually online or not.
9775          */
9776         if (lun != NULL)
9777                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9778                                 lun->be_lun->lun_type;
9779         else
9780                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9781
9782         pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9783         /* Supported VPD pages */
9784         pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9785         /* Serial Number */
9786         pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9787         /* Device Identification */
9788         pages->page_list[2] = SVPD_DEVICE_ID;
9789         /* Extended INQUIRY Data */
9790         pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9791         /* Mode Page Policy */
9792         pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9793         /* SCSI Ports */
9794         pages->page_list[5] = SVPD_SCSI_PORTS;
9795         /* Third-party Copy */
9796         pages->page_list[6] = SVPD_SCSI_TPC;
9797         /* Block limits */
9798         pages->page_list[7] = SVPD_BLOCK_LIMITS;
9799         /* Block Device Characteristics */
9800         pages->page_list[8] = SVPD_BDC;
9801         /* Logical Block Provisioning */
9802         pages->page_list[9] = SVPD_LBP;
9803
9804         ctsio->scsi_status = SCSI_STATUS_OK;
9805
9806         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9807         ctsio->be_move_done = ctl_config_move_done;
9808         ctl_datamove((union ctl_io *)ctsio);
9809
9810         return (CTL_RETVAL_COMPLETE);
9811 }
9812
9813 static int
9814 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9815 {
9816         struct scsi_vpd_unit_serial_number *sn_ptr;
9817         struct ctl_lun *lun;
9818
9819         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9820
9821         ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9822         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9823         ctsio->kern_sg_entries = 0;
9824
9825         if (sizeof(*sn_ptr) < alloc_len) {
9826                 ctsio->residual = alloc_len - sizeof(*sn_ptr);
9827                 ctsio->kern_data_len = sizeof(*sn_ptr);
9828                 ctsio->kern_total_len = sizeof(*sn_ptr);
9829         } else {
9830                 ctsio->residual = 0;
9831                 ctsio->kern_data_len = alloc_len;
9832                 ctsio->kern_total_len = alloc_len;
9833         }
9834         ctsio->kern_data_resid = 0;
9835         ctsio->kern_rel_offset = 0;
9836         ctsio->kern_sg_entries = 0;
9837
9838         /*
9839          * The control device is always connected.  The disk device, on the
9840          * other hand, may not be online all the time.  Need to change this
9841          * to figure out whether the disk device is actually online or not.
9842          */
9843         if (lun != NULL)
9844                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9845                                   lun->be_lun->lun_type;
9846         else
9847                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9848
9849         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9850         sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9851         /*
9852          * If we don't have a LUN, we just leave the serial number as
9853          * all spaces.
9854          */
9855         memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9856         if (lun != NULL) {
9857                 strncpy((char *)sn_ptr->serial_num,
9858                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9859         }
9860         ctsio->scsi_status = SCSI_STATUS_OK;
9861
9862         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9863         ctsio->be_move_done = ctl_config_move_done;
9864         ctl_datamove((union ctl_io *)ctsio);
9865
9866         return (CTL_RETVAL_COMPLETE);
9867 }
9868
9869
9870 static int
9871 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9872 {
9873         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9874         struct ctl_lun *lun;
9875         int data_len;
9876
9877         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9878
9879         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9880         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9881         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9882         ctsio->kern_sg_entries = 0;
9883
9884         if (data_len < alloc_len) {
9885                 ctsio->residual = alloc_len - data_len;
9886                 ctsio->kern_data_len = data_len;
9887                 ctsio->kern_total_len = data_len;
9888         } else {
9889                 ctsio->residual = 0;
9890                 ctsio->kern_data_len = alloc_len;
9891                 ctsio->kern_total_len = alloc_len;
9892         }
9893         ctsio->kern_data_resid = 0;
9894         ctsio->kern_rel_offset = 0;
9895         ctsio->kern_sg_entries = 0;
9896
9897         /*
9898          * The control device is always connected.  The disk device, on the
9899          * other hand, may not be online all the time.
9900          */
9901         if (lun != NULL)
9902                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9903                                      lun->be_lun->lun_type;
9904         else
9905                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9906         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9907         eid_ptr->page_length = data_len - 4;
9908         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9909         eid_ptr->flags3 = SVPD_EID_V_SUP;
9910
9911         ctsio->scsi_status = SCSI_STATUS_OK;
9912         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9913         ctsio->be_move_done = ctl_config_move_done;
9914         ctl_datamove((union ctl_io *)ctsio);
9915
9916         return (CTL_RETVAL_COMPLETE);
9917 }
9918
9919 static int
9920 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9921 {
9922         struct scsi_vpd_mode_page_policy *mpp_ptr;
9923         struct ctl_lun *lun;
9924         int data_len;
9925
9926         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9927
9928         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9929             sizeof(struct scsi_vpd_mode_page_policy_descr);
9930
9931         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9932         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9933         ctsio->kern_sg_entries = 0;
9934
9935         if (data_len < alloc_len) {
9936                 ctsio->residual = alloc_len - data_len;
9937                 ctsio->kern_data_len = data_len;
9938                 ctsio->kern_total_len = data_len;
9939         } else {
9940                 ctsio->residual = 0;
9941                 ctsio->kern_data_len = alloc_len;
9942                 ctsio->kern_total_len = alloc_len;
9943         }
9944         ctsio->kern_data_resid = 0;
9945         ctsio->kern_rel_offset = 0;
9946         ctsio->kern_sg_entries = 0;
9947
9948         /*
9949          * The control device is always connected.  The disk device, on the
9950          * other hand, may not be online all the time.
9951          */
9952         if (lun != NULL)
9953                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9954                                      lun->be_lun->lun_type;
9955         else
9956                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9957         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9958         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9959         mpp_ptr->descr[0].page_code = 0x3f;
9960         mpp_ptr->descr[0].subpage_code = 0xff;
9961         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9962
9963         ctsio->scsi_status = SCSI_STATUS_OK;
9964         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9965         ctsio->be_move_done = ctl_config_move_done;
9966         ctl_datamove((union ctl_io *)ctsio);
9967
9968         return (CTL_RETVAL_COMPLETE);
9969 }
9970
9971 static int
9972 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9973 {
9974         struct scsi_vpd_device_id *devid_ptr;
9975         struct scsi_vpd_id_descriptor *desc;
9976         struct ctl_softc *ctl_softc;
9977         struct ctl_lun *lun;
9978         struct ctl_port *port;
9979         int data_len;
9980         uint8_t proto;
9981
9982         ctl_softc = control_softc;
9983
9984         port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9985         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9986
9987         data_len = sizeof(struct scsi_vpd_device_id) +
9988             sizeof(struct scsi_vpd_id_descriptor) +
9989                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9990             sizeof(struct scsi_vpd_id_descriptor) +
9991                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9992         if (lun && lun->lun_devid)
9993                 data_len += lun->lun_devid->len;
9994         if (port->port_devid)
9995                 data_len += port->port_devid->len;
9996         if (port->target_devid)
9997                 data_len += port->target_devid->len;
9998
9999         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10000         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
10001         ctsio->kern_sg_entries = 0;
10002
10003         if (data_len < alloc_len) {
10004                 ctsio->residual = alloc_len - data_len;
10005                 ctsio->kern_data_len = data_len;
10006                 ctsio->kern_total_len = data_len;
10007         } else {
10008                 ctsio->residual = 0;
10009                 ctsio->kern_data_len = alloc_len;
10010                 ctsio->kern_total_len = alloc_len;
10011         }
10012         ctsio->kern_data_resid = 0;
10013         ctsio->kern_rel_offset = 0;
10014         ctsio->kern_sg_entries = 0;
10015
10016         /*
10017          * The control device is always connected.  The disk device, on the
10018          * other hand, may not be online all the time.
10019          */
10020         if (lun != NULL)
10021                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10022                                      lun->be_lun->lun_type;
10023         else
10024                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10025         devid_ptr->page_code = SVPD_DEVICE_ID;
10026         scsi_ulto2b(data_len - 4, devid_ptr->length);
10027
10028         if (port->port_type == CTL_PORT_FC)
10029                 proto = SCSI_PROTO_FC << 4;
10030         else if (port->port_type == CTL_PORT_ISCSI)
10031                 proto = SCSI_PROTO_ISCSI << 4;
10032         else
10033                 proto = SCSI_PROTO_SPI << 4;
10034         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
10035
10036         /*
10037          * We're using a LUN association here.  i.e., this device ID is a
10038          * per-LUN identifier.
10039          */
10040         if (lun && lun->lun_devid) {
10041                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
10042                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10043                     lun->lun_devid->len);
10044         }
10045
10046         /*
10047          * This is for the WWPN which is a port association.
10048          */
10049         if (port->port_devid) {
10050                 memcpy(desc, port->port_devid->data, port->port_devid->len);
10051                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10052                     port->port_devid->len);
10053         }
10054
10055         /*
10056          * This is for the Relative Target Port(type 4h) identifier
10057          */
10058         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10059         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10060             SVPD_ID_TYPE_RELTARG;
10061         desc->length = 4;
10062         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
10063         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10064             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
10065
10066         /*
10067          * This is for the Target Port Group(type 5h) identifier
10068          */
10069         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10070         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10071             SVPD_ID_TYPE_TPORTGRP;
10072         desc->length = 4;
10073         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
10074             &desc->identifier[2]);
10075         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10076             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10077
10078         /*
10079          * This is for the Target identifier
10080          */
10081         if (port->target_devid) {
10082                 memcpy(desc, port->target_devid->data, port->target_devid->len);
10083         }
10084
10085         ctsio->scsi_status = SCSI_STATUS_OK;
10086         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10087         ctsio->be_move_done = ctl_config_move_done;
10088         ctl_datamove((union ctl_io *)ctsio);
10089
10090         return (CTL_RETVAL_COMPLETE);
10091 }
10092
10093 static int
10094 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10095 {
10096         struct ctl_softc *softc = control_softc;
10097         struct scsi_vpd_scsi_ports *sp;
10098         struct scsi_vpd_port_designation *pd;
10099         struct scsi_vpd_port_designation_cont *pdc;
10100         struct ctl_lun *lun;
10101         struct ctl_port *port;
10102         int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10103         int num_target_port_groups, single;
10104
10105         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10106
10107         single = ctl_is_single;
10108         if (single)
10109                 num_target_port_groups = 1;
10110         else
10111                 num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10112         num_target_ports = 0;
10113         iid_len = 0;
10114         id_len = 0;
10115         mtx_lock(&softc->ctl_lock);
10116         STAILQ_FOREACH(port, &softc->port_list, links) {
10117                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10118                         continue;
10119                 if (lun != NULL &&
10120                     ctl_map_lun_back(port->targ_port, lun->lun) >=
10121                     CTL_MAX_LUNS)
10122                         continue;
10123                 num_target_ports++;
10124                 if (port->init_devid)
10125                         iid_len += port->init_devid->len;
10126                 if (port->port_devid)
10127                         id_len += port->port_devid->len;
10128         }
10129         mtx_unlock(&softc->ctl_lock);
10130
10131         data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10132             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10133              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10134         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10135         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10136         ctsio->kern_sg_entries = 0;
10137
10138         if (data_len < alloc_len) {
10139                 ctsio->residual = alloc_len - data_len;
10140                 ctsio->kern_data_len = data_len;
10141                 ctsio->kern_total_len = data_len;
10142         } else {
10143                 ctsio->residual = 0;
10144                 ctsio->kern_data_len = alloc_len;
10145                 ctsio->kern_total_len = alloc_len;
10146         }
10147         ctsio->kern_data_resid = 0;
10148         ctsio->kern_rel_offset = 0;
10149         ctsio->kern_sg_entries = 0;
10150
10151         /*
10152          * The control device is always connected.  The disk device, on the
10153          * other hand, may not be online all the time.  Need to change this
10154          * to figure out whether the disk device is actually online or not.
10155          */
10156         if (lun != NULL)
10157                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10158                                   lun->be_lun->lun_type;
10159         else
10160                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10161
10162         sp->page_code = SVPD_SCSI_PORTS;
10163         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10164             sp->page_length);
10165         pd = &sp->design[0];
10166
10167         mtx_lock(&softc->ctl_lock);
10168         if (softc->flags & CTL_FLAG_MASTER_SHELF)
10169                 pg = 0;
10170         else
10171                 pg = 1;
10172         for (g = 0; g < num_target_port_groups; g++) {
10173                 STAILQ_FOREACH(port, &softc->port_list, links) {
10174                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10175                                 continue;
10176                         if (lun != NULL &&
10177                             ctl_map_lun_back(port->targ_port, lun->lun) >=
10178                             CTL_MAX_LUNS)
10179                                 continue;
10180                         p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10181                         scsi_ulto2b(p, pd->relative_port_id);
10182                         if (port->init_devid && g == pg) {
10183                                 iid_len = port->init_devid->len;
10184                                 memcpy(pd->initiator_transportid,
10185                                     port->init_devid->data, port->init_devid->len);
10186                         } else
10187                                 iid_len = 0;
10188                         scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10189                         pdc = (struct scsi_vpd_port_designation_cont *)
10190                             (&pd->initiator_transportid[iid_len]);
10191                         if (port->port_devid && g == pg) {
10192                                 id_len = port->port_devid->len;
10193                                 memcpy(pdc->target_port_descriptors,
10194                                     port->port_devid->data, port->port_devid->len);
10195                         } else
10196                                 id_len = 0;
10197                         scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10198                         pd = (struct scsi_vpd_port_designation *)
10199                             ((uint8_t *)pdc->target_port_descriptors + id_len);
10200                 }
10201         }
10202         mtx_unlock(&softc->ctl_lock);
10203
10204         ctsio->scsi_status = SCSI_STATUS_OK;
10205         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10206         ctsio->be_move_done = ctl_config_move_done;
10207         ctl_datamove((union ctl_io *)ctsio);
10208
10209         return (CTL_RETVAL_COMPLETE);
10210 }
10211
10212 static int
10213 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10214 {
10215         struct scsi_vpd_block_limits *bl_ptr;
10216         struct ctl_lun *lun;
10217         int bs;
10218
10219         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10220
10221         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10222         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10223         ctsio->kern_sg_entries = 0;
10224
10225         if (sizeof(*bl_ptr) < alloc_len) {
10226                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
10227                 ctsio->kern_data_len = sizeof(*bl_ptr);
10228                 ctsio->kern_total_len = sizeof(*bl_ptr);
10229         } else {
10230                 ctsio->residual = 0;
10231                 ctsio->kern_data_len = alloc_len;
10232                 ctsio->kern_total_len = alloc_len;
10233         }
10234         ctsio->kern_data_resid = 0;
10235         ctsio->kern_rel_offset = 0;
10236         ctsio->kern_sg_entries = 0;
10237
10238         /*
10239          * The control device is always connected.  The disk device, on the
10240          * other hand, may not be online all the time.  Need to change this
10241          * to figure out whether the disk device is actually online or not.
10242          */
10243         if (lun != NULL)
10244                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10245                                   lun->be_lun->lun_type;
10246         else
10247                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10248
10249         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10250         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
10251         bl_ptr->max_cmp_write_len = 0xff;
10252         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10253         if (lun != NULL) {
10254                 bs = lun->be_lun->blocksize;
10255                 scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10256                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10257                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10258                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10259                         if (lun->be_lun->pblockexp != 0) {
10260                                 scsi_ulto4b((1 << lun->be_lun->pblockexp),
10261                                     bl_ptr->opt_unmap_grain);
10262                                 scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10263                                     bl_ptr->unmap_grain_align);
10264                         }
10265                 }
10266                 scsi_ulto4b(lun->be_lun->atomicblock,
10267                     bl_ptr->max_atomic_transfer_length);
10268                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
10269                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10270         }
10271         scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10272
10273         ctsio->scsi_status = SCSI_STATUS_OK;
10274         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10275         ctsio->be_move_done = ctl_config_move_done;
10276         ctl_datamove((union ctl_io *)ctsio);
10277
10278         return (CTL_RETVAL_COMPLETE);
10279 }
10280
10281 static int
10282 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10283 {
10284         struct scsi_vpd_block_device_characteristics *bdc_ptr;
10285         struct ctl_lun *lun;
10286
10287         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10288
10289         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10290         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10291         ctsio->kern_sg_entries = 0;
10292
10293         if (sizeof(*bdc_ptr) < alloc_len) {
10294                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10295                 ctsio->kern_data_len = sizeof(*bdc_ptr);
10296                 ctsio->kern_total_len = sizeof(*bdc_ptr);
10297         } else {
10298                 ctsio->residual = 0;
10299                 ctsio->kern_data_len = alloc_len;
10300                 ctsio->kern_total_len = alloc_len;
10301         }
10302         ctsio->kern_data_resid = 0;
10303         ctsio->kern_rel_offset = 0;
10304         ctsio->kern_sg_entries = 0;
10305
10306         /*
10307          * The control device is always connected.  The disk device, on the
10308          * other hand, may not be online all the time.  Need to change this
10309          * to figure out whether the disk device is actually online or not.
10310          */
10311         if (lun != NULL)
10312                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10313                                   lun->be_lun->lun_type;
10314         else
10315                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10316         bdc_ptr->page_code = SVPD_BDC;
10317         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10318         scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10319         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10320
10321         ctsio->scsi_status = SCSI_STATUS_OK;
10322         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10323         ctsio->be_move_done = ctl_config_move_done;
10324         ctl_datamove((union ctl_io *)ctsio);
10325
10326         return (CTL_RETVAL_COMPLETE);
10327 }
10328
10329 static int
10330 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10331 {
10332         struct scsi_vpd_logical_block_prov *lbp_ptr;
10333         struct ctl_lun *lun;
10334
10335         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10336
10337         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10338         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10339         ctsio->kern_sg_entries = 0;
10340
10341         if (sizeof(*lbp_ptr) < alloc_len) {
10342                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10343                 ctsio->kern_data_len = sizeof(*lbp_ptr);
10344                 ctsio->kern_total_len = sizeof(*lbp_ptr);
10345         } else {
10346                 ctsio->residual = 0;
10347                 ctsio->kern_data_len = alloc_len;
10348                 ctsio->kern_total_len = alloc_len;
10349         }
10350         ctsio->kern_data_resid = 0;
10351         ctsio->kern_rel_offset = 0;
10352         ctsio->kern_sg_entries = 0;
10353
10354         /*
10355          * The control device is always connected.  The disk device, on the
10356          * other hand, may not be online all the time.  Need to change this
10357          * to figure out whether the disk device is actually online or not.
10358          */
10359         if (lun != NULL)
10360                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10361                                   lun->be_lun->lun_type;
10362         else
10363                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10364
10365         lbp_ptr->page_code = SVPD_LBP;
10366         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10367         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10368                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10369                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10370                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10371         }
10372
10373         ctsio->scsi_status = SCSI_STATUS_OK;
10374         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10375         ctsio->be_move_done = ctl_config_move_done;
10376         ctl_datamove((union ctl_io *)ctsio);
10377
10378         return (CTL_RETVAL_COMPLETE);
10379 }
10380
10381 static int
10382 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10383 {
10384         struct scsi_inquiry *cdb;
10385         struct ctl_lun *lun;
10386         int alloc_len, retval;
10387
10388         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10389         cdb = (struct scsi_inquiry *)ctsio->cdb;
10390
10391         retval = CTL_RETVAL_COMPLETE;
10392
10393         alloc_len = scsi_2btoul(cdb->length);
10394
10395         switch (cdb->page_code) {
10396         case SVPD_SUPPORTED_PAGES:
10397                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10398                 break;
10399         case SVPD_UNIT_SERIAL_NUMBER:
10400                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10401                 break;
10402         case SVPD_DEVICE_ID:
10403                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10404                 break;
10405         case SVPD_EXTENDED_INQUIRY_DATA:
10406                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10407                 break;
10408         case SVPD_MODE_PAGE_POLICY:
10409                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10410                 break;
10411         case SVPD_SCSI_PORTS:
10412                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10413                 break;
10414         case SVPD_SCSI_TPC:
10415                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10416                 break;
10417         case SVPD_BLOCK_LIMITS:
10418                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10419                 break;
10420         case SVPD_BDC:
10421                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10422                 break;
10423         case SVPD_LBP:
10424                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10425                 break;
10426         default:
10427                 ctl_set_invalid_field(ctsio,
10428                                       /*sks_valid*/ 1,
10429                                       /*command*/ 1,
10430                                       /*field*/ 2,
10431                                       /*bit_valid*/ 0,
10432                                       /*bit*/ 0);
10433                 ctl_done((union ctl_io *)ctsio);
10434                 retval = CTL_RETVAL_COMPLETE;
10435                 break;
10436         }
10437
10438         return (retval);
10439 }
10440
10441 static int
10442 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10443 {
10444         struct scsi_inquiry_data *inq_ptr;
10445         struct scsi_inquiry *cdb;
10446         struct ctl_softc *ctl_softc;
10447         struct ctl_lun *lun;
10448         char *val;
10449         uint32_t alloc_len, data_len;
10450         ctl_port_type port_type;
10451
10452         ctl_softc = control_softc;
10453
10454         /*
10455          * Figure out whether we're talking to a Fibre Channel port or not.
10456          * We treat the ioctl front end, and any SCSI adapters, as packetized
10457          * SCSI front ends.
10458          */
10459         port_type = ctl_softc->ctl_ports[
10460             ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10461         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10462                 port_type = CTL_PORT_SCSI;
10463
10464         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10465         cdb = (struct scsi_inquiry *)ctsio->cdb;
10466         alloc_len = scsi_2btoul(cdb->length);
10467
10468         /*
10469          * We malloc the full inquiry data size here and fill it
10470          * in.  If the user only asks for less, we'll give him
10471          * that much.
10472          */
10473         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10474         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10475         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10476         ctsio->kern_sg_entries = 0;
10477         ctsio->kern_data_resid = 0;
10478         ctsio->kern_rel_offset = 0;
10479
10480         if (data_len < alloc_len) {
10481                 ctsio->residual = alloc_len - data_len;
10482                 ctsio->kern_data_len = data_len;
10483                 ctsio->kern_total_len = data_len;
10484         } else {
10485                 ctsio->residual = 0;
10486                 ctsio->kern_data_len = alloc_len;
10487                 ctsio->kern_total_len = alloc_len;
10488         }
10489
10490         /*
10491          * If we have a LUN configured, report it as connected.  Otherwise,
10492          * report that it is offline or no device is supported, depending
10493          * on the value of inquiry_pq_no_lun.
10494          *
10495          * According to the spec (SPC-4 r34), the peripheral qualifier
10496          * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10497          *
10498          * "A peripheral device having the specified peripheral device type
10499          * is not connected to this logical unit. However, the device
10500          * server is capable of supporting the specified peripheral device
10501          * type on this logical unit."
10502          *
10503          * According to the same spec, the peripheral qualifier
10504          * SID_QUAL_BAD_LU (011b) is used in this scenario:
10505          *
10506          * "The device server is not capable of supporting a peripheral
10507          * device on this logical unit. For this peripheral qualifier the
10508          * peripheral device type shall be set to 1Fh. All other peripheral
10509          * device type values are reserved for this peripheral qualifier."
10510          *
10511          * Given the text, it would seem that we probably want to report that
10512          * the LUN is offline here.  There is no LUN connected, but we can
10513          * support a LUN at the given LUN number.
10514          *
10515          * In the real world, though, it sounds like things are a little
10516          * different:
10517          *
10518          * - Linux, when presented with a LUN with the offline peripheral
10519          *   qualifier, will create an sg driver instance for it.  So when
10520          *   you attach it to CTL, you wind up with a ton of sg driver
10521          *   instances.  (One for every LUN that Linux bothered to probe.)
10522          *   Linux does this despite the fact that it issues a REPORT LUNs
10523          *   to LUN 0 to get the inventory of supported LUNs.
10524          *
10525          * - There is other anecdotal evidence (from Emulex folks) about
10526          *   arrays that use the offline peripheral qualifier for LUNs that
10527          *   are on the "passive" path in an active/passive array.
10528          *
10529          * So the solution is provide a hopefully reasonable default
10530          * (return bad/no LUN) and allow the user to change the behavior
10531          * with a tunable/sysctl variable.
10532          */
10533         if (lun != NULL)
10534                 inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10535                                   lun->be_lun->lun_type;
10536         else if (ctl_softc->inquiry_pq_no_lun == 0)
10537                 inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10538         else
10539                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10540
10541         /* RMB in byte 2 is 0 */
10542         inq_ptr->version = SCSI_REV_SPC4;
10543
10544         /*
10545          * According to SAM-3, even if a device only supports a single
10546          * level of LUN addressing, it should still set the HISUP bit:
10547          *
10548          * 4.9.1 Logical unit numbers overview
10549          *
10550          * All logical unit number formats described in this standard are
10551          * hierarchical in structure even when only a single level in that
10552          * hierarchy is used. The HISUP bit shall be set to one in the
10553          * standard INQUIRY data (see SPC-2) when any logical unit number
10554          * format described in this standard is used.  Non-hierarchical
10555          * formats are outside the scope of this standard.
10556          *
10557          * Therefore we set the HiSup bit here.
10558          *
10559          * The reponse format is 2, per SPC-3.
10560          */
10561         inq_ptr->response_format = SID_HiSup | 2;
10562
10563         inq_ptr->additional_length = data_len -
10564             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10565         CTL_DEBUG_PRINT(("additional_length = %d\n",
10566                          inq_ptr->additional_length));
10567
10568         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10569         /* 16 bit addressing */
10570         if (port_type == CTL_PORT_SCSI)
10571                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10572         /* XXX set the SID_MultiP bit here if we're actually going to
10573            respond on multiple ports */
10574         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10575
10576         /* 16 bit data bus, synchronous transfers */
10577         if (port_type == CTL_PORT_SCSI)
10578                 inq_ptr->flags = SID_WBus16 | SID_Sync;
10579         /*
10580          * XXX KDM do we want to support tagged queueing on the control
10581          * device at all?
10582          */
10583         if ((lun == NULL)
10584          || (lun->be_lun->lun_type != T_PROCESSOR))
10585                 inq_ptr->flags |= SID_CmdQue;
10586         /*
10587          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10588          * We have 8 bytes for the vendor name, and 16 bytes for the device
10589          * name and 4 bytes for the revision.
10590          */
10591         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10592             "vendor")) == NULL) {
10593                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10594         } else {
10595                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10596                 strncpy(inq_ptr->vendor, val,
10597                     min(sizeof(inq_ptr->vendor), strlen(val)));
10598         }
10599         if (lun == NULL) {
10600                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10601                     sizeof(inq_ptr->product));
10602         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10603                 switch (lun->be_lun->lun_type) {
10604                 case T_DIRECT:
10605                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10606                             sizeof(inq_ptr->product));
10607                         break;
10608                 case T_PROCESSOR:
10609                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10610                             sizeof(inq_ptr->product));
10611                         break;
10612                 default:
10613                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10614                             sizeof(inq_ptr->product));
10615                         break;
10616                 }
10617         } else {
10618                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10619                 strncpy(inq_ptr->product, val,
10620                     min(sizeof(inq_ptr->product), strlen(val)));
10621         }
10622
10623         /*
10624          * XXX make this a macro somewhere so it automatically gets
10625          * incremented when we make changes.
10626          */
10627         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10628             "revision")) == NULL) {
10629                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10630         } else {
10631                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10632                 strncpy(inq_ptr->revision, val,
10633                     min(sizeof(inq_ptr->revision), strlen(val)));
10634         }
10635
10636         /*
10637          * For parallel SCSI, we support double transition and single
10638          * transition clocking.  We also support QAS (Quick Arbitration
10639          * and Selection) and Information Unit transfers on both the
10640          * control and array devices.
10641          */
10642         if (port_type == CTL_PORT_SCSI)
10643                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10644                                     SID_SPI_IUS;
10645
10646         /* SAM-5 (no version claimed) */
10647         scsi_ulto2b(0x00A0, inq_ptr->version1);
10648         /* SPC-4 (no version claimed) */
10649         scsi_ulto2b(0x0460, inq_ptr->version2);
10650         if (port_type == CTL_PORT_FC) {
10651                 /* FCP-2 ANSI INCITS.350:2003 */
10652                 scsi_ulto2b(0x0917, inq_ptr->version3);
10653         } else if (port_type == CTL_PORT_SCSI) {
10654                 /* SPI-4 ANSI INCITS.362:200x */
10655                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10656         } else if (port_type == CTL_PORT_ISCSI) {
10657                 /* iSCSI (no version claimed) */
10658                 scsi_ulto2b(0x0960, inq_ptr->version3);
10659         } else if (port_type == CTL_PORT_SAS) {
10660                 /* SAS (no version claimed) */
10661                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10662         }
10663
10664         if (lun == NULL) {
10665                 /* SBC-4 (no version claimed) */
10666                 scsi_ulto2b(0x0600, inq_ptr->version4);
10667         } else {
10668                 switch (lun->be_lun->lun_type) {
10669                 case T_DIRECT:
10670                         /* SBC-4 (no version claimed) */
10671                         scsi_ulto2b(0x0600, inq_ptr->version4);
10672                         break;
10673                 case T_PROCESSOR:
10674                 default:
10675                         break;
10676                 }
10677         }
10678
10679         ctsio->scsi_status = SCSI_STATUS_OK;
10680         if (ctsio->kern_data_len > 0) {
10681                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10682                 ctsio->be_move_done = ctl_config_move_done;
10683                 ctl_datamove((union ctl_io *)ctsio);
10684         } else {
10685                 ctsio->io_hdr.status = CTL_SUCCESS;
10686                 ctl_done((union ctl_io *)ctsio);
10687         }
10688
10689         return (CTL_RETVAL_COMPLETE);
10690 }
10691
10692 int
10693 ctl_inquiry(struct ctl_scsiio *ctsio)
10694 {
10695         struct scsi_inquiry *cdb;
10696         int retval;
10697
10698         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10699
10700         cdb = (struct scsi_inquiry *)ctsio->cdb;
10701         if (cdb->byte2 & SI_EVPD)
10702                 retval = ctl_inquiry_evpd(ctsio);
10703         else if (cdb->page_code == 0)
10704                 retval = ctl_inquiry_std(ctsio);
10705         else {
10706                 ctl_set_invalid_field(ctsio,
10707                                       /*sks_valid*/ 1,
10708                                       /*command*/ 1,
10709                                       /*field*/ 2,
10710                                       /*bit_valid*/ 0,
10711                                       /*bit*/ 0);
10712                 ctl_done((union ctl_io *)ctsio);
10713                 return (CTL_RETVAL_COMPLETE);
10714         }
10715
10716         return (retval);
10717 }
10718
10719 /*
10720  * For known CDB types, parse the LBA and length.
10721  */
10722 static int
10723 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10724 {
10725         if (io->io_hdr.io_type != CTL_IO_SCSI)
10726                 return (1);
10727
10728         switch (io->scsiio.cdb[0]) {
10729         case COMPARE_AND_WRITE: {
10730                 struct scsi_compare_and_write *cdb;
10731
10732                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10733
10734                 *lba = scsi_8btou64(cdb->addr);
10735                 *len = cdb->length;
10736                 break;
10737         }
10738         case READ_6:
10739         case WRITE_6: {
10740                 struct scsi_rw_6 *cdb;
10741
10742                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10743
10744                 *lba = scsi_3btoul(cdb->addr);
10745                 /* only 5 bits are valid in the most significant address byte */
10746                 *lba &= 0x1fffff;
10747                 *len = cdb->length;
10748                 break;
10749         }
10750         case READ_10:
10751         case WRITE_10: {
10752                 struct scsi_rw_10 *cdb;
10753
10754                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10755
10756                 *lba = scsi_4btoul(cdb->addr);
10757                 *len = scsi_2btoul(cdb->length);
10758                 break;
10759         }
10760         case WRITE_VERIFY_10: {
10761                 struct scsi_write_verify_10 *cdb;
10762
10763                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10764
10765                 *lba = scsi_4btoul(cdb->addr);
10766                 *len = scsi_2btoul(cdb->length);
10767                 break;
10768         }
10769         case READ_12:
10770         case WRITE_12: {
10771                 struct scsi_rw_12 *cdb;
10772
10773                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10774
10775                 *lba = scsi_4btoul(cdb->addr);
10776                 *len = scsi_4btoul(cdb->length);
10777                 break;
10778         }
10779         case WRITE_VERIFY_12: {
10780                 struct scsi_write_verify_12 *cdb;
10781
10782                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10783
10784                 *lba = scsi_4btoul(cdb->addr);
10785                 *len = scsi_4btoul(cdb->length);
10786                 break;
10787         }
10788         case READ_16:
10789         case WRITE_16:
10790         case WRITE_ATOMIC_16: {
10791                 struct scsi_rw_16 *cdb;
10792
10793                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10794
10795                 *lba = scsi_8btou64(cdb->addr);
10796                 *len = scsi_4btoul(cdb->length);
10797                 break;
10798         }
10799         case WRITE_VERIFY_16: {
10800                 struct scsi_write_verify_16 *cdb;
10801
10802                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10803
10804                 *lba = scsi_8btou64(cdb->addr);
10805                 *len = scsi_4btoul(cdb->length);
10806                 break;
10807         }
10808         case WRITE_SAME_10: {
10809                 struct scsi_write_same_10 *cdb;
10810
10811                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10812
10813                 *lba = scsi_4btoul(cdb->addr);
10814                 *len = scsi_2btoul(cdb->length);
10815                 break;
10816         }
10817         case WRITE_SAME_16: {
10818                 struct scsi_write_same_16 *cdb;
10819
10820                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10821
10822                 *lba = scsi_8btou64(cdb->addr);
10823                 *len = scsi_4btoul(cdb->length);
10824                 break;
10825         }
10826         case VERIFY_10: {
10827                 struct scsi_verify_10 *cdb;
10828
10829                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10830
10831                 *lba = scsi_4btoul(cdb->addr);
10832                 *len = scsi_2btoul(cdb->length);
10833                 break;
10834         }
10835         case VERIFY_12: {
10836                 struct scsi_verify_12 *cdb;
10837
10838                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10839
10840                 *lba = scsi_4btoul(cdb->addr);
10841                 *len = scsi_4btoul(cdb->length);
10842                 break;
10843         }
10844         case VERIFY_16: {
10845                 struct scsi_verify_16 *cdb;
10846
10847                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10848
10849                 *lba = scsi_8btou64(cdb->addr);
10850                 *len = scsi_4btoul(cdb->length);
10851                 break;
10852         }
10853         case UNMAP: {
10854                 *lba = 0;
10855                 *len = UINT64_MAX;
10856                 break;
10857         }
10858         default:
10859                 return (1);
10860                 break; /* NOTREACHED */
10861         }
10862
10863         return (0);
10864 }
10865
10866 static ctl_action
10867 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10868 {
10869         uint64_t endlba1, endlba2;
10870
10871         endlba1 = lba1 + len1 - 1;
10872         endlba2 = lba2 + len2 - 1;
10873
10874         if ((endlba1 < lba2)
10875          || (endlba2 < lba1))
10876                 return (CTL_ACTION_PASS);
10877         else
10878                 return (CTL_ACTION_BLOCK);
10879 }
10880
10881 static int
10882 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10883 {
10884         struct ctl_ptr_len_flags *ptrlen;
10885         struct scsi_unmap_desc *buf, *end, *range;
10886         uint64_t lba;
10887         uint32_t len;
10888
10889         /* If not UNMAP -- go other way. */
10890         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10891             io->scsiio.cdb[0] != UNMAP)
10892                 return (CTL_ACTION_ERROR);
10893
10894         /* If UNMAP without data -- block and wait for data. */
10895         ptrlen = (struct ctl_ptr_len_flags *)
10896             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10897         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10898             ptrlen->ptr == NULL)
10899                 return (CTL_ACTION_BLOCK);
10900
10901         /* UNMAP with data -- check for collision. */
10902         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10903         end = buf + ptrlen->len / sizeof(*buf);
10904         for (range = buf; range < end; range++) {
10905                 lba = scsi_8btou64(range->lba);
10906                 len = scsi_4btoul(range->length);
10907                 if ((lba < lba2 + len2) && (lba + len > lba2))
10908                         return (CTL_ACTION_BLOCK);
10909         }
10910         return (CTL_ACTION_PASS);
10911 }
10912
10913 static ctl_action
10914 ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10915 {
10916         uint64_t lba1, lba2;
10917         uint64_t len1, len2;
10918         int retval;
10919
10920         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10921                 return (CTL_ACTION_ERROR);
10922
10923         retval = ctl_extent_check_unmap(io2, lba1, len1);
10924         if (retval != CTL_ACTION_ERROR)
10925                 return (retval);
10926
10927         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10928                 return (CTL_ACTION_ERROR);
10929
10930         return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10931 }
10932
10933 static ctl_action
10934 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10935     union ctl_io *ooa_io)
10936 {
10937         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10938         ctl_serialize_action *serialize_row;
10939
10940         /*
10941          * The initiator attempted multiple untagged commands at the same
10942          * time.  Can't do that.
10943          */
10944         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10945          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10946          && ((pending_io->io_hdr.nexus.targ_port ==
10947               ooa_io->io_hdr.nexus.targ_port)
10948           && (pending_io->io_hdr.nexus.initid.id ==
10949               ooa_io->io_hdr.nexus.initid.id))
10950          && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10951                 return (CTL_ACTION_OVERLAP);
10952
10953         /*
10954          * The initiator attempted to send multiple tagged commands with
10955          * the same ID.  (It's fine if different initiators have the same
10956          * tag ID.)
10957          *
10958          * Even if all of those conditions are true, we don't kill the I/O
10959          * if the command ahead of us has been aborted.  We won't end up
10960          * sending it to the FETD, and it's perfectly legal to resend a
10961          * command with the same tag number as long as the previous
10962          * instance of this tag number has been aborted somehow.
10963          */
10964         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10965          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10966          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10967          && ((pending_io->io_hdr.nexus.targ_port ==
10968               ooa_io->io_hdr.nexus.targ_port)
10969           && (pending_io->io_hdr.nexus.initid.id ==
10970               ooa_io->io_hdr.nexus.initid.id))
10971          && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10972                 return (CTL_ACTION_OVERLAP_TAG);
10973
10974         /*
10975          * If we get a head of queue tag, SAM-3 says that we should
10976          * immediately execute it.
10977          *
10978          * What happens if this command would normally block for some other
10979          * reason?  e.g. a request sense with a head of queue tag
10980          * immediately after a write.  Normally that would block, but this
10981          * will result in its getting executed immediately...
10982          *
10983          * We currently return "pass" instead of "skip", so we'll end up
10984          * going through the rest of the queue to check for overlapped tags.
10985          *
10986          * XXX KDM check for other types of blockage first??
10987          */
10988         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10989                 return (CTL_ACTION_PASS);
10990
10991         /*
10992          * Ordered tags have to block until all items ahead of them
10993          * have completed.  If we get called with an ordered tag, we always
10994          * block, if something else is ahead of us in the queue.
10995          */
10996         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10997                 return (CTL_ACTION_BLOCK);
10998
10999         /*
11000          * Simple tags get blocked until all head of queue and ordered tags
11001          * ahead of them have completed.  I'm lumping untagged commands in
11002          * with simple tags here.  XXX KDM is that the right thing to do?
11003          */
11004         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11005           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11006          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11007           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11008                 return (CTL_ACTION_BLOCK);
11009
11010         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11011         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11012
11013         serialize_row = ctl_serialize_table[ooa_entry->seridx];
11014
11015         switch (serialize_row[pending_entry->seridx]) {
11016         case CTL_SER_BLOCK:
11017                 return (CTL_ACTION_BLOCK);
11018         case CTL_SER_EXTENT:
11019                 return (ctl_extent_check(pending_io, ooa_io));
11020         case CTL_SER_EXTENTOPT:
11021                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11022                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11023                         return (ctl_extent_check(pending_io, ooa_io));
11024                 /* FALLTHROUGH */
11025         case CTL_SER_PASS:
11026                 return (CTL_ACTION_PASS);
11027         case CTL_SER_BLOCKOPT:
11028                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11029                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11030                         return (CTL_ACTION_BLOCK);
11031                 return (CTL_ACTION_PASS);
11032         case CTL_SER_SKIP:
11033                 return (CTL_ACTION_SKIP);
11034         default:
11035                 panic("invalid serialization value %d",
11036                       serialize_row[pending_entry->seridx]);
11037         }
11038
11039         return (CTL_ACTION_ERROR);
11040 }
11041
11042 /*
11043  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11044  * Assumptions:
11045  * - pending_io is generally either incoming, or on the blocked queue
11046  * - starting I/O is the I/O we want to start the check with.
11047  */
11048 static ctl_action
11049 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11050               union ctl_io *starting_io)
11051 {
11052         union ctl_io *ooa_io;
11053         ctl_action action;
11054
11055         mtx_assert(&lun->lun_lock, MA_OWNED);
11056
11057         /*
11058          * Run back along the OOA queue, starting with the current
11059          * blocked I/O and going through every I/O before it on the
11060          * queue.  If starting_io is NULL, we'll just end up returning
11061          * CTL_ACTION_PASS.
11062          */
11063         for (ooa_io = starting_io; ooa_io != NULL;
11064              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11065              ooa_links)){
11066
11067                 /*
11068                  * This routine just checks to see whether
11069                  * cur_blocked is blocked by ooa_io, which is ahead
11070                  * of it in the queue.  It doesn't queue/dequeue
11071                  * cur_blocked.
11072                  */
11073                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11074                 switch (action) {
11075                 case CTL_ACTION_BLOCK:
11076                 case CTL_ACTION_OVERLAP:
11077                 case CTL_ACTION_OVERLAP_TAG:
11078                 case CTL_ACTION_SKIP:
11079                 case CTL_ACTION_ERROR:
11080                         return (action);
11081                         break; /* NOTREACHED */
11082                 case CTL_ACTION_PASS:
11083                         break;
11084                 default:
11085                         panic("invalid action %d", action);
11086                         break;  /* NOTREACHED */
11087                 }
11088         }
11089
11090         return (CTL_ACTION_PASS);
11091 }
11092
11093 /*
11094  * Assumptions:
11095  * - An I/O has just completed, and has been removed from the per-LUN OOA
11096  *   queue, so some items on the blocked queue may now be unblocked.
11097  */
11098 static int
11099 ctl_check_blocked(struct ctl_lun *lun)
11100 {
11101         union ctl_io *cur_blocked, *next_blocked;
11102
11103         mtx_assert(&lun->lun_lock, MA_OWNED);
11104
11105         /*
11106          * Run forward from the head of the blocked queue, checking each
11107          * entry against the I/Os prior to it on the OOA queue to see if
11108          * there is still any blockage.
11109          *
11110          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11111          * with our removing a variable on it while it is traversing the
11112          * list.
11113          */
11114         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11115              cur_blocked != NULL; cur_blocked = next_blocked) {
11116                 union ctl_io *prev_ooa;
11117                 ctl_action action;
11118
11119                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11120                                                           blocked_links);
11121
11122                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11123                                                       ctl_ooaq, ooa_links);
11124
11125                 /*
11126                  * If cur_blocked happens to be the first item in the OOA
11127                  * queue now, prev_ooa will be NULL, and the action
11128                  * returned will just be CTL_ACTION_PASS.
11129                  */
11130                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11131
11132                 switch (action) {
11133                 case CTL_ACTION_BLOCK:
11134                         /* Nothing to do here, still blocked */
11135                         break;
11136                 case CTL_ACTION_OVERLAP:
11137                 case CTL_ACTION_OVERLAP_TAG:
11138                         /*
11139                          * This shouldn't happen!  In theory we've already
11140                          * checked this command for overlap...
11141                          */
11142                         break;
11143                 case CTL_ACTION_PASS:
11144                 case CTL_ACTION_SKIP: {
11145                         struct ctl_softc *softc;
11146                         const struct ctl_cmd_entry *entry;
11147                         uint32_t initidx;
11148                         int isc_retval;
11149
11150                         /*
11151                          * The skip case shouldn't happen, this transaction
11152                          * should have never made it onto the blocked queue.
11153                          */
11154                         /*
11155                          * This I/O is no longer blocked, we can remove it
11156                          * from the blocked queue.  Since this is a TAILQ
11157                          * (doubly linked list), we can do O(1) removals
11158                          * from any place on the list.
11159                          */
11160                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11161                                      blocked_links);
11162                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11163
11164                         if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11165                                 /*
11166                                  * Need to send IO back to original side to
11167                                  * run
11168                                  */
11169                                 union ctl_ha_msg msg_info;
11170
11171                                 msg_info.hdr.original_sc =
11172                                         cur_blocked->io_hdr.original_sc;
11173                                 msg_info.hdr.serializing_sc = cur_blocked;
11174                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11175                                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11176                                      &msg_info, sizeof(msg_info), 0)) >
11177                                      CTL_HA_STATUS_SUCCESS) {
11178                                         printf("CTL:Check Blocked error from "
11179                                                "ctl_ha_msg_send %d\n",
11180                                                isc_retval);
11181                                 }
11182                                 break;
11183                         }
11184                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11185                         softc = control_softc;
11186
11187                         initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11188
11189                         /*
11190                          * Check this I/O for LUN state changes that may
11191                          * have happened while this command was blocked.
11192                          * The LUN state may have been changed by a command
11193                          * ahead of us in the queue, so we need to re-check
11194                          * for any states that can be caused by SCSI
11195                          * commands.
11196                          */
11197                         if (ctl_scsiio_lun_check(softc, lun, entry,
11198                                                  &cur_blocked->scsiio) == 0) {
11199                                 cur_blocked->io_hdr.flags |=
11200                                                       CTL_FLAG_IS_WAS_ON_RTR;
11201                                 ctl_enqueue_rtr(cur_blocked);
11202                         } else
11203                                 ctl_done(cur_blocked);
11204                         break;
11205                 }
11206                 default:
11207                         /*
11208                          * This probably shouldn't happen -- we shouldn't
11209                          * get CTL_ACTION_ERROR, or anything else.
11210                          */
11211                         break;
11212                 }
11213         }
11214
11215         return (CTL_RETVAL_COMPLETE);
11216 }
11217
11218 /*
11219  * This routine (with one exception) checks LUN flags that can be set by
11220  * commands ahead of us in the OOA queue.  These flags have to be checked
11221  * when a command initially comes in, and when we pull a command off the
11222  * blocked queue and are preparing to execute it.  The reason we have to
11223  * check these flags for commands on the blocked queue is that the LUN
11224  * state may have been changed by a command ahead of us while we're on the
11225  * blocked queue.
11226  *
11227  * Ordering is somewhat important with these checks, so please pay
11228  * careful attention to the placement of any new checks.
11229  */
11230 static int
11231 ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11232     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11233 {
11234         int retval;
11235         uint32_t residx;
11236
11237         retval = 0;
11238
11239         mtx_assert(&lun->lun_lock, MA_OWNED);
11240
11241         /*
11242          * If this shelf is a secondary shelf controller, we have to reject
11243          * any media access commands.
11244          */
11245 #if 0
11246         /* No longer needed for HA */
11247         if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11248          && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11249                 ctl_set_lun_standby(ctsio);
11250                 retval = 1;
11251                 goto bailout;
11252         }
11253 #endif
11254
11255         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11256                 if (lun->flags & CTL_LUN_READONLY) {
11257                         ctl_set_sense(ctsio, /*current_error*/ 1,
11258                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11259                             /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
11260                         retval = 1;
11261                         goto bailout;
11262                 }
11263                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11264                     .eca_and_aen & SCP_SWP) != 0) {
11265                         ctl_set_sense(ctsio, /*current_error*/ 1,
11266                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11267                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11268                         retval = 1;
11269                         goto bailout;
11270                 }
11271         }
11272
11273         /*
11274          * Check for a reservation conflict.  If this command isn't allowed
11275          * even on reserved LUNs, and if this initiator isn't the one who
11276          * reserved us, reject the command with a reservation conflict.
11277          */
11278         residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11279         if ((lun->flags & CTL_LUN_RESERVED)
11280          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11281                 if (lun->res_idx != residx) {
11282                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11283                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
11284                         retval = 1;
11285                         goto bailout;
11286                 }
11287         }
11288
11289         if ((lun->flags & CTL_LUN_PR_RESERVED)
11290          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11291                 /*
11292                  * if we aren't registered or it's a res holder type
11293                  * reservation and this isn't the res holder then set a
11294                  * conflict.
11295                  * NOTE: Commands which might be allowed on write exclusive
11296                  * type reservations are checked in the particular command
11297                  * for a conflict. Read and SSU are the only ones.
11298                  */
11299                 if (lun->pr_keys[residx] == 0
11300                  || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11301                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11302                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
11303                         retval = 1;
11304                         goto bailout;
11305                 }
11306
11307         }
11308
11309         if ((lun->flags & CTL_LUN_OFFLINE)
11310          && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11311                 ctl_set_lun_not_ready(ctsio);
11312                 retval = 1;
11313                 goto bailout;
11314         }
11315
11316         /*
11317          * If the LUN is stopped, see if this particular command is allowed
11318          * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11319          */
11320         if ((lun->flags & CTL_LUN_STOPPED)
11321          && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11322                 /* "Logical unit not ready, initializing cmd. required" */
11323                 ctl_set_lun_stopped(ctsio);
11324                 retval = 1;
11325                 goto bailout;
11326         }
11327
11328         if ((lun->flags & CTL_LUN_INOPERABLE)
11329          && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11330                 /* "Medium format corrupted" */
11331                 ctl_set_medium_format_corrupted(ctsio);
11332                 retval = 1;
11333                 goto bailout;
11334         }
11335
11336 bailout:
11337         return (retval);
11338
11339 }
11340
11341 static void
11342 ctl_failover_io(union ctl_io *io, int have_lock)
11343 {
11344         ctl_set_busy(&io->scsiio);
11345         ctl_done(io);
11346 }
11347
11348 static void
11349 ctl_failover(void)
11350 {
11351         struct ctl_lun *lun;
11352         struct ctl_softc *ctl_softc;
11353         union ctl_io *next_io, *pending_io;
11354         union ctl_io *io;
11355         int lun_idx;
11356         int i;
11357
11358         ctl_softc = control_softc;
11359
11360         mtx_lock(&ctl_softc->ctl_lock);
11361         /*
11362          * Remove any cmds from the other SC from the rtr queue.  These
11363          * will obviously only be for LUNs for which we're the primary.
11364          * We can't send status or get/send data for these commands.
11365          * Since they haven't been executed yet, we can just remove them.
11366          * We'll either abort them or delete them below, depending on
11367          * which HA mode we're in.
11368          */
11369 #ifdef notyet
11370         mtx_lock(&ctl_softc->queue_lock);
11371         for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11372              io != NULL; io = next_io) {
11373                 next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11374                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11375                         STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11376                                       ctl_io_hdr, links);
11377         }
11378         mtx_unlock(&ctl_softc->queue_lock);
11379 #endif
11380
11381         for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11382                 lun = ctl_softc->ctl_luns[lun_idx];
11383                 if (lun==NULL)
11384                         continue;
11385
11386                 /*
11387                  * Processor LUNs are primary on both sides.
11388                  * XXX will this always be true?
11389                  */
11390                 if (lun->be_lun->lun_type == T_PROCESSOR)
11391                         continue;
11392
11393                 if ((lun->flags & CTL_LUN_PRIMARY_SC)
11394                  && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11395                         printf("FAILOVER: primary lun %d\n", lun_idx);
11396                         /*
11397                          * Remove all commands from the other SC. First from the
11398                          * blocked queue then from the ooa queue. Once we have
11399                          * removed them. Call ctl_check_blocked to see if there
11400                          * is anything that can run.
11401                          */
11402                         for (io = (union ctl_io *)TAILQ_FIRST(
11403                              &lun->blocked_queue); io != NULL; io = next_io) {
11404
11405                                 next_io = (union ctl_io *)TAILQ_NEXT(
11406                                     &io->io_hdr, blocked_links);
11407
11408                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11409                                         TAILQ_REMOVE(&lun->blocked_queue,
11410                                                      &io->io_hdr,blocked_links);
11411                                         io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11412                                         TAILQ_REMOVE(&lun->ooa_queue,
11413                                                      &io->io_hdr, ooa_links);
11414
11415                                         ctl_free_io(io);
11416                                 }
11417                         }
11418
11419                         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11420                              io != NULL; io = next_io) {
11421
11422                                 next_io = (union ctl_io *)TAILQ_NEXT(
11423                                     &io->io_hdr, ooa_links);
11424
11425                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11426
11427                                         TAILQ_REMOVE(&lun->ooa_queue,
11428                                                 &io->io_hdr,
11429                                                 ooa_links);
11430
11431                                         ctl_free_io(io);
11432                                 }
11433                         }
11434                         ctl_check_blocked(lun);
11435                 } else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11436                         && (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11437
11438                         printf("FAILOVER: primary lun %d\n", lun_idx);
11439                         /*
11440                          * Abort all commands from the other SC.  We can't
11441                          * send status back for them now.  These should get
11442                          * cleaned up when they are completed or come out
11443                          * for a datamove operation.
11444                          */
11445                         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11446                              io != NULL; io = next_io) {
11447                                 next_io = (union ctl_io *)TAILQ_NEXT(
11448                                         &io->io_hdr, ooa_links);
11449
11450                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11451                                         io->io_hdr.flags |= CTL_FLAG_ABORT;
11452                         }
11453                 } else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11454                         && (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11455
11456                         printf("FAILOVER: secondary lun %d\n", lun_idx);
11457
11458                         lun->flags |= CTL_LUN_PRIMARY_SC;
11459
11460                         /*
11461                          * We send all I/O that was sent to this controller
11462                          * and redirected to the other side back with
11463                          * busy status, and have the initiator retry it.
11464                          * Figuring out how much data has been transferred,
11465                          * etc. and picking up where we left off would be
11466                          * very tricky.
11467                          *
11468                          * XXX KDM need to remove I/O from the blocked
11469                          * queue as well!
11470                          */
11471                         for (pending_io = (union ctl_io *)TAILQ_FIRST(
11472                              &lun->ooa_queue); pending_io != NULL;
11473                              pending_io = next_io) {
11474
11475                                 next_io =  (union ctl_io *)TAILQ_NEXT(
11476                                         &pending_io->io_hdr, ooa_links);
11477
11478                                 pending_io->io_hdr.flags &=
11479                                         ~CTL_FLAG_SENT_2OTHER_SC;
11480
11481                                 if (pending_io->io_hdr.flags &
11482                                     CTL_FLAG_IO_ACTIVE) {
11483                                         pending_io->io_hdr.flags |=
11484                                                 CTL_FLAG_FAILOVER;
11485                                 } else {
11486                                         ctl_set_busy(&pending_io->scsiio);
11487                                         ctl_done(pending_io);
11488                                 }
11489                         }
11490
11491                         /*
11492                          * Build Unit Attention
11493                          */
11494                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11495                                 lun->pending_ua[i] |=
11496                                                      CTL_UA_ASYM_ACC_CHANGE;
11497                         }
11498                 } else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11499                         && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11500                         printf("FAILOVER: secondary lun %d\n", lun_idx);
11501                         /*
11502                          * if the first io on the OOA is not on the RtR queue
11503                          * add it.
11504                          */
11505                         lun->flags |= CTL_LUN_PRIMARY_SC;
11506
11507                         pending_io = (union ctl_io *)TAILQ_FIRST(
11508                             &lun->ooa_queue);
11509                         if (pending_io==NULL) {
11510                                 printf("Nothing on OOA queue\n");
11511                                 continue;
11512                         }
11513
11514                         pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11515                         if ((pending_io->io_hdr.flags &
11516                              CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11517                                 pending_io->io_hdr.flags |=
11518                                     CTL_FLAG_IS_WAS_ON_RTR;
11519                                 ctl_enqueue_rtr(pending_io);
11520                         }
11521 #if 0
11522                         else
11523                         {
11524                                 printf("Tag 0x%04x is running\n",
11525                                       pending_io->scsiio.tag_num);
11526                         }
11527 #endif
11528
11529                         next_io = (union ctl_io *)TAILQ_NEXT(
11530                             &pending_io->io_hdr, ooa_links);
11531                         for (pending_io=next_io; pending_io != NULL;
11532                              pending_io = next_io) {
11533                                 pending_io->io_hdr.flags &=
11534                                     ~CTL_FLAG_SENT_2OTHER_SC;
11535                                 next_io = (union ctl_io *)TAILQ_NEXT(
11536                                         &pending_io->io_hdr, ooa_links);
11537                                 if (pending_io->io_hdr.flags &
11538                                     CTL_FLAG_IS_WAS_ON_RTR) {
11539 #if 0
11540                                         printf("Tag 0x%04x is running\n",
11541                                                 pending_io->scsiio.tag_num);
11542 #endif
11543                                         continue;
11544                                 }
11545
11546                                 switch (ctl_check_ooa(lun, pending_io,
11547                                     (union ctl_io *)TAILQ_PREV(
11548                                     &pending_io->io_hdr, ctl_ooaq,
11549                                     ooa_links))) {
11550
11551                                 case CTL_ACTION_BLOCK:
11552                                         TAILQ_INSERT_TAIL(&lun->blocked_queue,
11553                                                           &pending_io->io_hdr,
11554                                                           blocked_links);
11555                                         pending_io->io_hdr.flags |=
11556                                             CTL_FLAG_BLOCKED;
11557                                         break;
11558                                 case CTL_ACTION_PASS:
11559                                 case CTL_ACTION_SKIP:
11560                                         pending_io->io_hdr.flags |=
11561                                             CTL_FLAG_IS_WAS_ON_RTR;
11562                                         ctl_enqueue_rtr(pending_io);
11563                                         break;
11564                                 case CTL_ACTION_OVERLAP:
11565                                         ctl_set_overlapped_cmd(
11566                                             (struct ctl_scsiio *)pending_io);
11567                                         ctl_done(pending_io);
11568                                         break;
11569                                 case CTL_ACTION_OVERLAP_TAG:
11570                                         ctl_set_overlapped_tag(
11571                                             (struct ctl_scsiio *)pending_io,
11572                                             pending_io->scsiio.tag_num & 0xff);
11573                                         ctl_done(pending_io);
11574                                         break;
11575                                 case CTL_ACTION_ERROR:
11576                                 default:
11577                                         ctl_set_internal_failure(
11578                                                 (struct ctl_scsiio *)pending_io,
11579                                                 0,  // sks_valid
11580                                                 0); //retry count
11581                                         ctl_done(pending_io);
11582                                         break;
11583                                 }
11584                         }
11585
11586                         /*
11587                          * Build Unit Attention
11588                          */
11589                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11590                                 lun->pending_ua[i] |=
11591                                                      CTL_UA_ASYM_ACC_CHANGE;
11592                         }
11593                 } else {
11594                         panic("Unhandled HA mode failover, LUN flags = %#x, "
11595                               "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11596                 }
11597         }
11598         ctl_pause_rtr = 0;
11599         mtx_unlock(&ctl_softc->ctl_lock);
11600 }
11601
11602 static int
11603 ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11604 {
11605         struct ctl_lun *lun;
11606         const struct ctl_cmd_entry *entry;
11607         uint32_t initidx, targ_lun;
11608         int retval;
11609
11610         retval = 0;
11611
11612         lun = NULL;
11613
11614         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11615         if ((targ_lun < CTL_MAX_LUNS)
11616          && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11617                 lun = ctl_softc->ctl_luns[targ_lun];
11618                 /*
11619                  * If the LUN is invalid, pretend that it doesn't exist.
11620                  * It will go away as soon as all pending I/O has been
11621                  * completed.
11622                  */
11623                 if (lun->flags & CTL_LUN_DISABLED) {
11624                         lun = NULL;
11625                 } else {
11626                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11627                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11628                                 lun->be_lun;
11629                         if (lun->be_lun->lun_type == T_PROCESSOR) {
11630                                 ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11631                         }
11632
11633                         /*
11634                          * Every I/O goes into the OOA queue for a
11635                          * particular LUN, and stays there until completion.
11636                          */
11637                         mtx_lock(&lun->lun_lock);
11638                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11639                             ooa_links);
11640                 }
11641         } else {
11642                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11643                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11644         }
11645
11646         /* Get command entry and return error if it is unsuppotyed. */
11647         entry = ctl_validate_command(ctsio);
11648         if (entry == NULL) {
11649                 if (lun)
11650                         mtx_unlock(&lun->lun_lock);
11651                 return (retval);
11652         }
11653
11654         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11655         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11656
11657         /*
11658          * Check to see whether we can send this command to LUNs that don't
11659          * exist.  This should pretty much only be the case for inquiry
11660          * and request sense.  Further checks, below, really require having
11661          * a LUN, so we can't really check the command anymore.  Just put
11662          * it on the rtr queue.
11663          */
11664         if (lun == NULL) {
11665                 if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11666                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11667                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11668                         return (retval);
11669                 }
11670
11671                 ctl_set_unsupported_lun(ctsio);
11672                 ctl_done((union ctl_io *)ctsio);
11673                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11674                 return (retval);
11675         } else {
11676                 /*
11677                  * Make sure we support this particular command on this LUN.
11678                  * e.g., we don't support writes to the control LUN.
11679                  */
11680                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11681                         mtx_unlock(&lun->lun_lock);
11682                         ctl_set_invalid_opcode(ctsio);
11683                         ctl_done((union ctl_io *)ctsio);
11684                         return (retval);
11685                 }
11686         }
11687
11688         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11689
11690 #ifdef CTL_WITH_CA
11691         /*
11692          * If we've got a request sense, it'll clear the contingent
11693          * allegiance condition.  Otherwise, if we have a CA condition for
11694          * this initiator, clear it, because it sent down a command other
11695          * than request sense.
11696          */
11697         if ((ctsio->cdb[0] != REQUEST_SENSE)
11698          && (ctl_is_set(lun->have_ca, initidx)))
11699                 ctl_clear_mask(lun->have_ca, initidx);
11700 #endif
11701
11702         /*
11703          * If the command has this flag set, it handles its own unit
11704          * attention reporting, we shouldn't do anything.  Otherwise we
11705          * check for any pending unit attentions, and send them back to the
11706          * initiator.  We only do this when a command initially comes in,
11707          * not when we pull it off the blocked queue.
11708          *
11709          * According to SAM-3, section 5.3.2, the order that things get
11710          * presented back to the host is basically unit attentions caused
11711          * by some sort of reset event, busy status, reservation conflicts
11712          * or task set full, and finally any other status.
11713          *
11714          * One issue here is that some of the unit attentions we report
11715          * don't fall into the "reset" category (e.g. "reported luns data
11716          * has changed").  So reporting it here, before the reservation
11717          * check, may be technically wrong.  I guess the only thing to do
11718          * would be to check for and report the reset events here, and then
11719          * check for the other unit attention types after we check for a
11720          * reservation conflict.
11721          *
11722          * XXX KDM need to fix this
11723          */
11724         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11725                 ctl_ua_type ua_type;
11726
11727                 if (lun->pending_ua[initidx] != CTL_UA_NONE) {
11728                         scsi_sense_data_type sense_format;
11729
11730                         if (lun != NULL)
11731                                 sense_format = (lun->flags &
11732                                     CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11733                                     SSD_TYPE_FIXED;
11734                         else
11735                                 sense_format = SSD_TYPE_FIXED;
11736
11737                         ua_type = ctl_build_ua(&lun->pending_ua[initidx],
11738                             &ctsio->sense_data, sense_format);
11739                         if (ua_type != CTL_UA_NONE) {
11740                                 ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11741                                 ctsio->io_hdr.status = CTL_SCSI_ERROR |
11742                                                        CTL_AUTOSENSE;
11743                                 ctsio->sense_len = SSD_FULL_SIZE;
11744                                 mtx_unlock(&lun->lun_lock);
11745                                 ctl_done((union ctl_io *)ctsio);
11746                                 return (retval);
11747                         }
11748                 }
11749         }
11750
11751
11752         if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11753                 mtx_unlock(&lun->lun_lock);
11754                 ctl_done((union ctl_io *)ctsio);
11755                 return (retval);
11756         }
11757
11758         /*
11759          * XXX CHD this is where we want to send IO to other side if
11760          * this LUN is secondary on this SC. We will need to make a copy
11761          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11762          * the copy we send as FROM_OTHER.
11763          * We also need to stuff the address of the original IO so we can
11764          * find it easily. Something similar will need be done on the other
11765          * side so when we are done we can find the copy.
11766          */
11767         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11768                 union ctl_ha_msg msg_info;
11769                 int isc_retval;
11770
11771                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11772
11773                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11774                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11775 #if 0
11776                 printf("1. ctsio %p\n", ctsio);
11777 #endif
11778                 msg_info.hdr.serializing_sc = NULL;
11779                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11780                 msg_info.scsi.tag_num = ctsio->tag_num;
11781                 msg_info.scsi.tag_type = ctsio->tag_type;
11782                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11783
11784                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11785
11786                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11787                     (void *)&msg_info, sizeof(msg_info), 0)) >
11788                     CTL_HA_STATUS_SUCCESS) {
11789                         printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11790                                isc_retval);
11791                         printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11792                 } else {
11793 #if 0
11794                         printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11795 #endif
11796                 }
11797
11798                 /*
11799                  * XXX KDM this I/O is off the incoming queue, but hasn't
11800                  * been inserted on any other queue.  We may need to come
11801                  * up with a holding queue while we wait for serialization
11802                  * so that we have an idea of what we're waiting for from
11803                  * the other side.
11804                  */
11805                 mtx_unlock(&lun->lun_lock);
11806                 return (retval);
11807         }
11808
11809         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11810                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11811                               ctl_ooaq, ooa_links))) {
11812         case CTL_ACTION_BLOCK:
11813                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11814                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11815                                   blocked_links);
11816                 mtx_unlock(&lun->lun_lock);
11817                 return (retval);
11818         case CTL_ACTION_PASS:
11819         case CTL_ACTION_SKIP:
11820                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11821                 mtx_unlock(&lun->lun_lock);
11822                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11823                 break;
11824         case CTL_ACTION_OVERLAP:
11825                 mtx_unlock(&lun->lun_lock);
11826                 ctl_set_overlapped_cmd(ctsio);
11827                 ctl_done((union ctl_io *)ctsio);
11828                 break;
11829         case CTL_ACTION_OVERLAP_TAG:
11830                 mtx_unlock(&lun->lun_lock);
11831                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11832                 ctl_done((union ctl_io *)ctsio);
11833                 break;
11834         case CTL_ACTION_ERROR:
11835         default:
11836                 mtx_unlock(&lun->lun_lock);
11837                 ctl_set_internal_failure(ctsio,
11838                                          /*sks_valid*/ 0,
11839                                          /*retry_count*/ 0);
11840                 ctl_done((union ctl_io *)ctsio);
11841                 break;
11842         }
11843         return (retval);
11844 }
11845
11846 const struct ctl_cmd_entry *
11847 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11848 {
11849         const struct ctl_cmd_entry *entry;
11850         int service_action;
11851
11852         entry = &ctl_cmd_table[ctsio->cdb[0]];
11853         if (sa)
11854                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11855         if (entry->flags & CTL_CMD_FLAG_SA5) {
11856                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11857                 entry = &((const struct ctl_cmd_entry *)
11858                     entry->execute)[service_action];
11859         }
11860         return (entry);
11861 }
11862
11863 const struct ctl_cmd_entry *
11864 ctl_validate_command(struct ctl_scsiio *ctsio)
11865 {
11866         const struct ctl_cmd_entry *entry;
11867         int i, sa;
11868         uint8_t diff;
11869
11870         entry = ctl_get_cmd_entry(ctsio, &sa);
11871         if (entry->execute == NULL) {
11872                 if (sa)
11873                         ctl_set_invalid_field(ctsio,
11874                                               /*sks_valid*/ 1,
11875                                               /*command*/ 1,
11876                                               /*field*/ 1,
11877                                               /*bit_valid*/ 1,
11878                                               /*bit*/ 4);
11879                 else
11880                         ctl_set_invalid_opcode(ctsio);
11881                 ctl_done((union ctl_io *)ctsio);
11882                 return (NULL);
11883         }
11884         KASSERT(entry->length > 0,
11885             ("Not defined length for command 0x%02x/0x%02x",
11886              ctsio->cdb[0], ctsio->cdb[1]));
11887         for (i = 1; i < entry->length; i++) {
11888                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11889                 if (diff == 0)
11890                         continue;
11891                 ctl_set_invalid_field(ctsio,
11892                                       /*sks_valid*/ 1,
11893                                       /*command*/ 1,
11894                                       /*field*/ i,
11895                                       /*bit_valid*/ 1,
11896                                       /*bit*/ fls(diff) - 1);
11897                 ctl_done((union ctl_io *)ctsio);
11898                 return (NULL);
11899         }
11900         return (entry);
11901 }
11902
11903 static int
11904 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11905 {
11906
11907         switch (lun_type) {
11908         case T_PROCESSOR:
11909                 if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11910                     ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11911                         return (0);
11912                 break;
11913         case T_DIRECT:
11914                 if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11915                     ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11916                         return (0);
11917                 break;
11918         default:
11919                 return (0);
11920         }
11921         return (1);
11922 }
11923
11924 static int
11925 ctl_scsiio(struct ctl_scsiio *ctsio)
11926 {
11927         int retval;
11928         const struct ctl_cmd_entry *entry;
11929
11930         retval = CTL_RETVAL_COMPLETE;
11931
11932         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11933
11934         entry = ctl_get_cmd_entry(ctsio, NULL);
11935
11936         /*
11937          * If this I/O has been aborted, just send it straight to
11938          * ctl_done() without executing it.
11939          */
11940         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11941                 ctl_done((union ctl_io *)ctsio);
11942                 goto bailout;
11943         }
11944
11945         /*
11946          * All the checks should have been handled by ctl_scsiio_precheck().
11947          * We should be clear now to just execute the I/O.
11948          */
11949         retval = entry->execute(ctsio);
11950
11951 bailout:
11952         return (retval);
11953 }
11954
11955 /*
11956  * Since we only implement one target right now, a bus reset simply resets
11957  * our single target.
11958  */
11959 static int
11960 ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11961 {
11962         return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11963 }
11964
11965 static int
11966 ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11967                  ctl_ua_type ua_type)
11968 {
11969         struct ctl_lun *lun;
11970         int retval;
11971
11972         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11973                 union ctl_ha_msg msg_info;
11974
11975                 io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11976                 msg_info.hdr.nexus = io->io_hdr.nexus;
11977                 if (ua_type==CTL_UA_TARG_RESET)
11978                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11979                 else
11980                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11981                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11982                 msg_info.hdr.original_sc = NULL;
11983                 msg_info.hdr.serializing_sc = NULL;
11984                 if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11985                     (void *)&msg_info, sizeof(msg_info), 0)) {
11986                 }
11987         }
11988         retval = 0;
11989
11990         mtx_lock(&ctl_softc->ctl_lock);
11991         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11992                 retval += ctl_lun_reset(lun, io, ua_type);
11993         mtx_unlock(&ctl_softc->ctl_lock);
11994
11995         return (retval);
11996 }
11997
11998 /*
11999  * The LUN should always be set.  The I/O is optional, and is used to
12000  * distinguish between I/Os sent by this initiator, and by other
12001  * initiators.  We set unit attention for initiators other than this one.
12002  * SAM-3 is vague on this point.  It does say that a unit attention should
12003  * be established for other initiators when a LUN is reset (see section
12004  * 5.7.3), but it doesn't specifically say that the unit attention should
12005  * be established for this particular initiator when a LUN is reset.  Here
12006  * is the relevant text, from SAM-3 rev 8:
12007  *
12008  * 5.7.2 When a SCSI initiator port aborts its own tasks
12009  *
12010  * When a SCSI initiator port causes its own task(s) to be aborted, no
12011  * notification that the task(s) have been aborted shall be returned to
12012  * the SCSI initiator port other than the completion response for the
12013  * command or task management function action that caused the task(s) to
12014  * be aborted and notification(s) associated with related effects of the
12015  * action (e.g., a reset unit attention condition).
12016  *
12017  * XXX KDM for now, we're setting unit attention for all initiators.
12018  */
12019 static int
12020 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
12021 {
12022         union ctl_io *xio;
12023 #if 0
12024         uint32_t initindex;
12025 #endif
12026         int i;
12027
12028         mtx_lock(&lun->lun_lock);
12029         /*
12030          * Run through the OOA queue and abort each I/O.
12031          */
12032 #if 0
12033         TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12034 #endif
12035         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12036              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12037                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
12038         }
12039
12040         /*
12041          * This version sets unit attention for every
12042          */
12043 #if 0
12044         initindex = ctl_get_initindex(&io->io_hdr.nexus);
12045         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12046                 if (initindex == i)
12047                         continue;
12048                 lun->pending_ua[i] |= ua_type;
12049         }
12050 #endif
12051
12052         /*
12053          * A reset (any kind, really) clears reservations established with
12054          * RESERVE/RELEASE.  It does not clear reservations established
12055          * with PERSISTENT RESERVE OUT, but we don't support that at the
12056          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
12057          * reservations made with the RESERVE/RELEASE commands, because
12058          * those commands are obsolete in SPC-3.
12059          */
12060         lun->flags &= ~CTL_LUN_RESERVED;
12061
12062         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12063 #ifdef CTL_WITH_CA
12064                 ctl_clear_mask(lun->have_ca, i);
12065 #endif
12066                 lun->pending_ua[i] |= ua_type;
12067         }
12068         mtx_unlock(&lun->lun_lock);
12069
12070         return (0);
12071 }
12072
12073 static void
12074 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
12075     int other_sc)
12076 {
12077         union ctl_io *xio;
12078
12079         mtx_assert(&lun->lun_lock, MA_OWNED);
12080
12081         /*
12082          * Run through the OOA queue and attempt to find the given I/O.
12083          * The target port, initiator ID, tag type and tag number have to
12084          * match the values that we got from the initiator.  If we have an
12085          * untagged command to abort, simply abort the first untagged command
12086          * we come to.  We only allow one untagged command at a time of course.
12087          */
12088         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12089              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12090
12091                 if ((targ_port == UINT32_MAX ||
12092                      targ_port == xio->io_hdr.nexus.targ_port) &&
12093                     (init_id == UINT32_MAX ||
12094                      init_id == xio->io_hdr.nexus.initid.id)) {
12095                         if (targ_port != xio->io_hdr.nexus.targ_port ||
12096                             init_id != xio->io_hdr.nexus.initid.id)
12097                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12098                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12099                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12100                                 union ctl_ha_msg msg_info;
12101
12102                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
12103                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12104                                 msg_info.task.tag_num = xio->scsiio.tag_num;
12105                                 msg_info.task.tag_type = xio->scsiio.tag_type;
12106                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12107                                 msg_info.hdr.original_sc = NULL;
12108                                 msg_info.hdr.serializing_sc = NULL;
12109                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12110                                     (void *)&msg_info, sizeof(msg_info), 0);
12111                         }
12112                 }
12113         }
12114 }
12115
12116 static int
12117 ctl_abort_task_set(union ctl_io *io)
12118 {
12119         struct ctl_softc *softc = control_softc;
12120         struct ctl_lun *lun;
12121         uint32_t targ_lun;
12122
12123         /*
12124          * Look up the LUN.
12125          */
12126         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12127         mtx_lock(&softc->ctl_lock);
12128         if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12129                 lun = softc->ctl_luns[targ_lun];
12130         else {
12131                 mtx_unlock(&softc->ctl_lock);
12132                 return (1);
12133         }
12134
12135         mtx_lock(&lun->lun_lock);
12136         mtx_unlock(&softc->ctl_lock);
12137         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12138                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12139                     io->io_hdr.nexus.initid.id,
12140                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12141         } else { /* CTL_TASK_CLEAR_TASK_SET */
12142                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12143                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12144         }
12145         mtx_unlock(&lun->lun_lock);
12146         return (0);
12147 }
12148
12149 static int
12150 ctl_i_t_nexus_reset(union ctl_io *io)
12151 {
12152         struct ctl_softc *softc = control_softc;
12153         struct ctl_lun *lun;
12154         uint32_t initindex, residx;
12155
12156         initindex = ctl_get_initindex(&io->io_hdr.nexus);
12157         residx = ctl_get_resindex(&io->io_hdr.nexus);
12158         mtx_lock(&softc->ctl_lock);
12159         STAILQ_FOREACH(lun, &softc->lun_list, links) {
12160                 mtx_lock(&lun->lun_lock);
12161                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12162                     io->io_hdr.nexus.initid.id,
12163                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12164 #ifdef CTL_WITH_CA
12165                 ctl_clear_mask(lun->have_ca, initindex);
12166 #endif
12167                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
12168                         lun->flags &= ~CTL_LUN_RESERVED;
12169                 lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12170                 mtx_unlock(&lun->lun_lock);
12171         }
12172         mtx_unlock(&softc->ctl_lock);
12173         return (0);
12174 }
12175
12176 static int
12177 ctl_abort_task(union ctl_io *io)
12178 {
12179         union ctl_io *xio;
12180         struct ctl_lun *lun;
12181         struct ctl_softc *ctl_softc;
12182 #if 0
12183         struct sbuf sb;
12184         char printbuf[128];
12185 #endif
12186         int found;
12187         uint32_t targ_lun;
12188
12189         ctl_softc = control_softc;
12190         found = 0;
12191
12192         /*
12193          * Look up the LUN.
12194          */
12195         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12196         mtx_lock(&ctl_softc->ctl_lock);
12197         if ((targ_lun < CTL_MAX_LUNS)
12198          && (ctl_softc->ctl_luns[targ_lun] != NULL))
12199                 lun = ctl_softc->ctl_luns[targ_lun];
12200         else {
12201                 mtx_unlock(&ctl_softc->ctl_lock);
12202                 return (1);
12203         }
12204
12205 #if 0
12206         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12207                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12208 #endif
12209
12210         mtx_lock(&lun->lun_lock);
12211         mtx_unlock(&ctl_softc->ctl_lock);
12212         /*
12213          * Run through the OOA queue and attempt to find the given I/O.
12214          * The target port, initiator ID, tag type and tag number have to
12215          * match the values that we got from the initiator.  If we have an
12216          * untagged command to abort, simply abort the first untagged command
12217          * we come to.  We only allow one untagged command at a time of course.
12218          */
12219 #if 0
12220         TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12221 #endif
12222         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12223              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12224 #if 0
12225                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12226
12227                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12228                             lun->lun, xio->scsiio.tag_num,
12229                             xio->scsiio.tag_type,
12230                             (xio->io_hdr.blocked_links.tqe_prev
12231                             == NULL) ? "" : " BLOCKED",
12232                             (xio->io_hdr.flags &
12233                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12234                             (xio->io_hdr.flags &
12235                             CTL_FLAG_ABORT) ? " ABORT" : "",
12236                             (xio->io_hdr.flags &
12237                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12238                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12239                 sbuf_finish(&sb);
12240                 printf("%s\n", sbuf_data(&sb));
12241 #endif
12242
12243                 if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12244                  && (xio->io_hdr.nexus.initid.id ==
12245                      io->io_hdr.nexus.initid.id)) {
12246                         /*
12247                          * If the abort says that the task is untagged, the
12248                          * task in the queue must be untagged.  Otherwise,
12249                          * we just check to see whether the tag numbers
12250                          * match.  This is because the QLogic firmware
12251                          * doesn't pass back the tag type in an abort
12252                          * request.
12253                          */
12254 #if 0
12255                         if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12256                           && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12257                          || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12258 #endif
12259                         /*
12260                          * XXX KDM we've got problems with FC, because it
12261                          * doesn't send down a tag type with aborts.  So we
12262                          * can only really go by the tag number...
12263                          * This may cause problems with parallel SCSI.
12264                          * Need to figure that out!!
12265                          */
12266                         if (xio->scsiio.tag_num == io->taskio.tag_num) {
12267                                 xio->io_hdr.flags |= CTL_FLAG_ABORT;
12268                                 found = 1;
12269                                 if ((io->io_hdr.flags &
12270                                      CTL_FLAG_FROM_OTHER_SC) == 0 &&
12271                                     !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12272                                         union ctl_ha_msg msg_info;
12273
12274                                         io->io_hdr.flags |=
12275                                                         CTL_FLAG_SENT_2OTHER_SC;
12276                                         msg_info.hdr.nexus = io->io_hdr.nexus;
12277                                         msg_info.task.task_action =
12278                                                 CTL_TASK_ABORT_TASK;
12279                                         msg_info.task.tag_num =
12280                                                 io->taskio.tag_num;
12281                                         msg_info.task.tag_type =
12282                                                 io->taskio.tag_type;
12283                                         msg_info.hdr.msg_type =
12284                                                 CTL_MSG_MANAGE_TASKS;
12285                                         msg_info.hdr.original_sc = NULL;
12286                                         msg_info.hdr.serializing_sc = NULL;
12287 #if 0
12288                                         printf("Sent Abort to other side\n");
12289 #endif
12290                                         if (CTL_HA_STATUS_SUCCESS !=
12291                                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12292                                                 (void *)&msg_info,
12293                                                 sizeof(msg_info), 0)) {
12294                                         }
12295                                 }
12296 #if 0
12297                                 printf("ctl_abort_task: found I/O to abort\n");
12298 #endif
12299                                 break;
12300                         }
12301                 }
12302         }
12303         mtx_unlock(&lun->lun_lock);
12304
12305         if (found == 0) {
12306                 /*
12307                  * This isn't really an error.  It's entirely possible for
12308                  * the abort and command completion to cross on the wire.
12309                  * This is more of an informative/diagnostic error.
12310                  */
12311 #if 0
12312                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12313                        "%d:%d:%d:%d tag %d type %d\n",
12314                        io->io_hdr.nexus.initid.id,
12315                        io->io_hdr.nexus.targ_port,
12316                        io->io_hdr.nexus.targ_target.id,
12317                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12318                        io->taskio.tag_type);
12319 #endif
12320         }
12321         return (0);
12322 }
12323
12324 static void
12325 ctl_run_task(union ctl_io *io)
12326 {
12327         struct ctl_softc *ctl_softc = control_softc;
12328         int retval = 1;
12329         const char *task_desc;
12330
12331         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12332
12333         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12334             ("ctl_run_task: Unextected io_type %d\n",
12335              io->io_hdr.io_type));
12336
12337         task_desc = ctl_scsi_task_string(&io->taskio);
12338         if (task_desc != NULL) {
12339 #ifdef NEEDTOPORT
12340                 csevent_log(CSC_CTL | CSC_SHELF_SW |
12341                             CTL_TASK_REPORT,
12342                             csevent_LogType_Trace,
12343                             csevent_Severity_Information,
12344                             csevent_AlertLevel_Green,
12345                             csevent_FRU_Firmware,
12346                             csevent_FRU_Unknown,
12347                             "CTL: received task: %s",task_desc);
12348 #endif
12349         } else {
12350 #ifdef NEEDTOPORT
12351                 csevent_log(CSC_CTL | CSC_SHELF_SW |
12352                             CTL_TASK_REPORT,
12353                             csevent_LogType_Trace,
12354                             csevent_Severity_Information,
12355                             csevent_AlertLevel_Green,
12356                             csevent_FRU_Firmware,
12357                             csevent_FRU_Unknown,
12358                             "CTL: received unknown task "
12359                             "type: %d (%#x)",
12360                             io->taskio.task_action,
12361                             io->taskio.task_action);
12362 #endif
12363         }
12364         switch (io->taskio.task_action) {
12365         case CTL_TASK_ABORT_TASK:
12366                 retval = ctl_abort_task(io);
12367                 break;
12368         case CTL_TASK_ABORT_TASK_SET:
12369         case CTL_TASK_CLEAR_TASK_SET:
12370                 retval = ctl_abort_task_set(io);
12371                 break;
12372         case CTL_TASK_CLEAR_ACA:
12373                 break;
12374         case CTL_TASK_I_T_NEXUS_RESET:
12375                 retval = ctl_i_t_nexus_reset(io);
12376                 break;
12377         case CTL_TASK_LUN_RESET: {
12378                 struct ctl_lun *lun;
12379                 uint32_t targ_lun;
12380
12381                 targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12382                 mtx_lock(&ctl_softc->ctl_lock);
12383                 if ((targ_lun < CTL_MAX_LUNS)
12384                  && (ctl_softc->ctl_luns[targ_lun] != NULL))
12385                         lun = ctl_softc->ctl_luns[targ_lun];
12386                 else {
12387                         mtx_unlock(&ctl_softc->ctl_lock);
12388                         retval = 1;
12389                         break;
12390                 }
12391
12392                 if (!(io->io_hdr.flags &
12393                     CTL_FLAG_FROM_OTHER_SC)) {
12394                         union ctl_ha_msg msg_info;
12395
12396                         io->io_hdr.flags |=
12397                                 CTL_FLAG_SENT_2OTHER_SC;
12398                         msg_info.hdr.msg_type =
12399                                 CTL_MSG_MANAGE_TASKS;
12400                         msg_info.hdr.nexus = io->io_hdr.nexus;
12401                         msg_info.task.task_action =
12402                                 CTL_TASK_LUN_RESET;
12403                         msg_info.hdr.original_sc = NULL;
12404                         msg_info.hdr.serializing_sc = NULL;
12405                         if (CTL_HA_STATUS_SUCCESS !=
12406                             ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12407                             (void *)&msg_info,
12408                             sizeof(msg_info), 0)) {
12409                         }
12410                 }
12411
12412                 retval = ctl_lun_reset(lun, io,
12413                                        CTL_UA_LUN_RESET);
12414                 mtx_unlock(&ctl_softc->ctl_lock);
12415                 break;
12416         }
12417         case CTL_TASK_TARGET_RESET:
12418                 retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12419                 break;
12420         case CTL_TASK_BUS_RESET:
12421                 retval = ctl_bus_reset(ctl_softc, io);
12422                 break;
12423         case CTL_TASK_PORT_LOGIN:
12424                 break;
12425         case CTL_TASK_PORT_LOGOUT:
12426                 break;
12427         default:
12428                 printf("ctl_run_task: got unknown task management event %d\n",
12429                        io->taskio.task_action);
12430                 break;
12431         }
12432         if (retval == 0)
12433                 io->io_hdr.status = CTL_SUCCESS;
12434         else
12435                 io->io_hdr.status = CTL_ERROR;
12436         ctl_done(io);
12437 }
12438
12439 /*
12440  * For HA operation.  Handle commands that come in from the other
12441  * controller.
12442  */
12443 static void
12444 ctl_handle_isc(union ctl_io *io)
12445 {
12446         int free_io;
12447         struct ctl_lun *lun;
12448         struct ctl_softc *ctl_softc;
12449         uint32_t targ_lun;
12450
12451         ctl_softc = control_softc;
12452
12453         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12454         lun = ctl_softc->ctl_luns[targ_lun];
12455
12456         switch (io->io_hdr.msg_type) {
12457         case CTL_MSG_SERIALIZE:
12458                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12459                 break;
12460         case CTL_MSG_R2R: {
12461                 const struct ctl_cmd_entry *entry;
12462
12463                 /*
12464                  * This is only used in SER_ONLY mode.
12465                  */
12466                 free_io = 0;
12467                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12468                 mtx_lock(&lun->lun_lock);
12469                 if (ctl_scsiio_lun_check(ctl_softc, lun,
12470                     entry, (struct ctl_scsiio *)io) != 0) {
12471                         mtx_unlock(&lun->lun_lock);
12472                         ctl_done(io);
12473                         break;
12474                 }
12475                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12476                 mtx_unlock(&lun->lun_lock);
12477                 ctl_enqueue_rtr(io);
12478                 break;
12479         }
12480         case CTL_MSG_FINISH_IO:
12481                 if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12482                         free_io = 0;
12483                         ctl_done(io);
12484                 } else {
12485                         free_io = 1;
12486                         mtx_lock(&lun->lun_lock);
12487                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12488                                      ooa_links);
12489                         ctl_check_blocked(lun);
12490                         mtx_unlock(&lun->lun_lock);
12491                 }
12492                 break;
12493         case CTL_MSG_PERS_ACTION:
12494                 ctl_hndl_per_res_out_on_other_sc(
12495                         (union ctl_ha_msg *)&io->presio.pr_msg);
12496                 free_io = 1;
12497                 break;
12498         case CTL_MSG_BAD_JUJU:
12499                 free_io = 0;
12500                 ctl_done(io);
12501                 break;
12502         case CTL_MSG_DATAMOVE:
12503                 /* Only used in XFER mode */
12504                 free_io = 0;
12505                 ctl_datamove_remote(io);
12506                 break;
12507         case CTL_MSG_DATAMOVE_DONE:
12508                 /* Only used in XFER mode */
12509                 free_io = 0;
12510                 io->scsiio.be_move_done(io);
12511                 break;
12512         default:
12513                 free_io = 1;
12514                 printf("%s: Invalid message type %d\n",
12515                        __func__, io->io_hdr.msg_type);
12516                 break;
12517         }
12518         if (free_io)
12519                 ctl_free_io(io);
12520
12521 }
12522
12523
12524 /*
12525  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12526  * there is no match.
12527  */
12528 static ctl_lun_error_pattern
12529 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12530 {
12531         const struct ctl_cmd_entry *entry;
12532         ctl_lun_error_pattern filtered_pattern, pattern;
12533
12534         pattern = desc->error_pattern;
12535
12536         /*
12537          * XXX KDM we need more data passed into this function to match a
12538          * custom pattern, and we actually need to implement custom pattern
12539          * matching.
12540          */
12541         if (pattern & CTL_LUN_PAT_CMD)
12542                 return (CTL_LUN_PAT_CMD);
12543
12544         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12545                 return (CTL_LUN_PAT_ANY);
12546
12547         entry = ctl_get_cmd_entry(ctsio, NULL);
12548
12549         filtered_pattern = entry->pattern & pattern;
12550
12551         /*
12552          * If the user requested specific flags in the pattern (e.g.
12553          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12554          * flags.
12555          *
12556          * If the user did not specify any flags, it doesn't matter whether
12557          * or not the command supports the flags.
12558          */
12559         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12560              (pattern & ~CTL_LUN_PAT_MASK))
12561                 return (CTL_LUN_PAT_NONE);
12562
12563         /*
12564          * If the user asked for a range check, see if the requested LBA
12565          * range overlaps with this command's LBA range.
12566          */
12567         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12568                 uint64_t lba1;
12569                 uint64_t len1;
12570                 ctl_action action;
12571                 int retval;
12572
12573                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12574                 if (retval != 0)
12575                         return (CTL_LUN_PAT_NONE);
12576
12577                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12578                                               desc->lba_range.len);
12579                 /*
12580                  * A "pass" means that the LBA ranges don't overlap, so
12581                  * this doesn't match the user's range criteria.
12582                  */
12583                 if (action == CTL_ACTION_PASS)
12584                         return (CTL_LUN_PAT_NONE);
12585         }
12586
12587         return (filtered_pattern);
12588 }
12589
12590 static void
12591 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12592 {
12593         struct ctl_error_desc *desc, *desc2;
12594
12595         mtx_assert(&lun->lun_lock, MA_OWNED);
12596
12597         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12598                 ctl_lun_error_pattern pattern;
12599                 /*
12600                  * Check to see whether this particular command matches
12601                  * the pattern in the descriptor.
12602                  */
12603                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12604                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12605                         continue;
12606
12607                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12608                 case CTL_LUN_INJ_ABORTED:
12609                         ctl_set_aborted(&io->scsiio);
12610                         break;
12611                 case CTL_LUN_INJ_MEDIUM_ERR:
12612                         ctl_set_medium_error(&io->scsiio);
12613                         break;
12614                 case CTL_LUN_INJ_UA:
12615                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12616                          * OCCURRED */
12617                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12618                         break;
12619                 case CTL_LUN_INJ_CUSTOM:
12620                         /*
12621                          * We're assuming the user knows what he is doing.
12622                          * Just copy the sense information without doing
12623                          * checks.
12624                          */
12625                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12626                               ctl_min(sizeof(desc->custom_sense),
12627                                       sizeof(io->scsiio.sense_data)));
12628                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12629                         io->scsiio.sense_len = SSD_FULL_SIZE;
12630                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12631                         break;
12632                 case CTL_LUN_INJ_NONE:
12633                 default:
12634                         /*
12635                          * If this is an error injection type we don't know
12636                          * about, clear the continuous flag (if it is set)
12637                          * so it will get deleted below.
12638                          */
12639                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12640                         break;
12641                 }
12642                 /*
12643                  * By default, each error injection action is a one-shot
12644                  */
12645                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12646                         continue;
12647
12648                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12649
12650                 free(desc, M_CTL);
12651         }
12652 }
12653
12654 #ifdef CTL_IO_DELAY
12655 static void
12656 ctl_datamove_timer_wakeup(void *arg)
12657 {
12658         union ctl_io *io;
12659
12660         io = (union ctl_io *)arg;
12661
12662         ctl_datamove(io);
12663 }
12664 #endif /* CTL_IO_DELAY */
12665
12666 void
12667 ctl_datamove(union ctl_io *io)
12668 {
12669         void (*fe_datamove)(union ctl_io *io);
12670
12671         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12672
12673         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12674
12675 #ifdef CTL_TIME_IO
12676         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12677                 char str[256];
12678                 char path_str[64];
12679                 struct sbuf sb;
12680
12681                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12682                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12683
12684                 sbuf_cat(&sb, path_str);
12685                 switch (io->io_hdr.io_type) {
12686                 case CTL_IO_SCSI:
12687                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12688                         sbuf_printf(&sb, "\n");
12689                         sbuf_cat(&sb, path_str);
12690                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12691                                     io->scsiio.tag_num, io->scsiio.tag_type);
12692                         break;
12693                 case CTL_IO_TASK:
12694                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12695                                     "Tag Type: %d\n", io->taskio.task_action,
12696                                     io->taskio.tag_num, io->taskio.tag_type);
12697                         break;
12698                 default:
12699                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12700                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12701                         break;
12702                 }
12703                 sbuf_cat(&sb, path_str);
12704                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12705                             (intmax_t)time_uptime - io->io_hdr.start_time);
12706                 sbuf_finish(&sb);
12707                 printf("%s", sbuf_data(&sb));
12708         }
12709 #endif /* CTL_TIME_IO */
12710
12711 #ifdef CTL_IO_DELAY
12712         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12713                 struct ctl_lun *lun;
12714
12715                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12716
12717                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12718         } else {
12719                 struct ctl_lun *lun;
12720
12721                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12722                 if ((lun != NULL)
12723                  && (lun->delay_info.datamove_delay > 0)) {
12724                         struct callout *callout;
12725
12726                         callout = (struct callout *)&io->io_hdr.timer_bytes;
12727                         callout_init(callout, /*mpsafe*/ 1);
12728                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12729                         callout_reset(callout,
12730                                       lun->delay_info.datamove_delay * hz,
12731                                       ctl_datamove_timer_wakeup, io);
12732                         if (lun->delay_info.datamove_type ==
12733                             CTL_DELAY_TYPE_ONESHOT)
12734                                 lun->delay_info.datamove_delay = 0;
12735                         return;
12736                 }
12737         }
12738 #endif
12739
12740         /*
12741          * This command has been aborted.  Set the port status, so we fail
12742          * the data move.
12743          */
12744         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12745                 printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12746                        io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12747                        io->io_hdr.nexus.targ_port,
12748                        (uintmax_t)io->io_hdr.nexus.targ_target.id,
12749                        io->io_hdr.nexus.targ_lun);
12750                 io->io_hdr.port_status = 31337;
12751                 /*
12752                  * Note that the backend, in this case, will get the
12753                  * callback in its context.  In other cases it may get
12754                  * called in the frontend's interrupt thread context.
12755                  */
12756                 io->scsiio.be_move_done(io);
12757                 return;
12758         }
12759
12760         /*
12761          * If we're in XFER mode and this I/O is from the other shelf
12762          * controller, we need to send the DMA to the other side to
12763          * actually transfer the data to/from the host.  In serialize only
12764          * mode the transfer happens below CTL and ctl_datamove() is only
12765          * called on the machine that originally received the I/O.
12766          */
12767         if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12768          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12769                 union ctl_ha_msg msg;
12770                 uint32_t sg_entries_sent;
12771                 int do_sg_copy;
12772                 int i;
12773
12774                 memset(&msg, 0, sizeof(msg));
12775                 msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12776                 msg.hdr.original_sc = io->io_hdr.original_sc;
12777                 msg.hdr.serializing_sc = io;
12778                 msg.hdr.nexus = io->io_hdr.nexus;
12779                 msg.dt.flags = io->io_hdr.flags;
12780                 /*
12781                  * We convert everything into a S/G list here.  We can't
12782                  * pass by reference, only by value between controllers.
12783                  * So we can't pass a pointer to the S/G list, only as many
12784                  * S/G entries as we can fit in here.  If it's possible for
12785                  * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12786                  * then we need to break this up into multiple transfers.
12787                  */
12788                 if (io->scsiio.kern_sg_entries == 0) {
12789                         msg.dt.kern_sg_entries = 1;
12790                         /*
12791                          * If this is in cached memory, flush the cache
12792                          * before we send the DMA request to the other
12793                          * controller.  We want to do this in either the
12794                          * read or the write case.  The read case is
12795                          * straightforward.  In the write case, we want to
12796                          * make sure nothing is in the local cache that
12797                          * could overwrite the DMAed data.
12798                          */
12799                         if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12800                                 /*
12801                                  * XXX KDM use bus_dmamap_sync() here.
12802                                  */
12803                         }
12804
12805                         /*
12806                          * Convert to a physical address if this is a
12807                          * virtual address.
12808                          */
12809                         if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12810                                 msg.dt.sg_list[0].addr =
12811                                         io->scsiio.kern_data_ptr;
12812                         } else {
12813                                 /*
12814                                  * XXX KDM use busdma here!
12815                                  */
12816 #if 0
12817                                 msg.dt.sg_list[0].addr = (void *)
12818                                         vtophys(io->scsiio.kern_data_ptr);
12819 #endif
12820                         }
12821
12822                         msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12823                         do_sg_copy = 0;
12824                 } else {
12825                         struct ctl_sg_entry *sgl;
12826
12827                         do_sg_copy = 1;
12828                         msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12829                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12830                         if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12831                                 /*
12832                                  * XXX KDM use bus_dmamap_sync() here.
12833                                  */
12834                         }
12835                 }
12836
12837                 msg.dt.kern_data_len = io->scsiio.kern_data_len;
12838                 msg.dt.kern_total_len = io->scsiio.kern_total_len;
12839                 msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12840                 msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12841                 msg.dt.sg_sequence = 0;
12842
12843                 /*
12844                  * Loop until we've sent all of the S/G entries.  On the
12845                  * other end, we'll recompose these S/G entries into one
12846                  * contiguous list before passing it to the
12847                  */
12848                 for (sg_entries_sent = 0; sg_entries_sent <
12849                      msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12850                         msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12851                                 sizeof(msg.dt.sg_list[0])),
12852                                 msg.dt.kern_sg_entries - sg_entries_sent);
12853
12854                         if (do_sg_copy != 0) {
12855                                 struct ctl_sg_entry *sgl;
12856                                 int j;
12857
12858                                 sgl = (struct ctl_sg_entry *)
12859                                         io->scsiio.kern_data_ptr;
12860                                 /*
12861                                  * If this is in cached memory, flush the cache
12862                                  * before we send the DMA request to the other
12863                                  * controller.  We want to do this in either
12864                                  * the * read or the write case.  The read
12865                                  * case is straightforward.  In the write
12866                                  * case, we want to make sure nothing is
12867                                  * in the local cache that could overwrite
12868                                  * the DMAed data.
12869                                  */
12870
12871                                 for (i = sg_entries_sent, j = 0;
12872                                      i < msg.dt.cur_sg_entries; i++, j++) {
12873                                         if ((io->io_hdr.flags &
12874                                              CTL_FLAG_NO_DATASYNC) == 0) {
12875                                                 /*
12876                                                  * XXX KDM use bus_dmamap_sync()
12877                                                  */
12878                                         }
12879                                         if ((io->io_hdr.flags &
12880                                              CTL_FLAG_BUS_ADDR) == 0) {
12881                                                 /*
12882                                                  * XXX KDM use busdma.
12883                                                  */
12884 #if 0
12885                                                 msg.dt.sg_list[j].addr =(void *)
12886                                                        vtophys(sgl[i].addr);
12887 #endif
12888                                         } else {
12889                                                 msg.dt.sg_list[j].addr =
12890                                                         sgl[i].addr;
12891                                         }
12892                                         msg.dt.sg_list[j].len = sgl[i].len;
12893                                 }
12894                         }
12895
12896                         sg_entries_sent += msg.dt.cur_sg_entries;
12897                         if (sg_entries_sent >= msg.dt.kern_sg_entries)
12898                                 msg.dt.sg_last = 1;
12899                         else
12900                                 msg.dt.sg_last = 0;
12901
12902                         /*
12903                          * XXX KDM drop and reacquire the lock here?
12904                          */
12905                         if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12906                             sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12907                                 /*
12908                                  * XXX do something here.
12909                                  */
12910                         }
12911
12912                         msg.dt.sent_sg_entries = sg_entries_sent;
12913                 }
12914                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12915                 if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12916                         ctl_failover_io(io, /*have_lock*/ 0);
12917
12918         } else {
12919
12920                 /*
12921                  * Lookup the fe_datamove() function for this particular
12922                  * front end.
12923                  */
12924                 fe_datamove =
12925                     control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12926
12927                 fe_datamove(io);
12928         }
12929 }
12930
12931 static void
12932 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12933 {
12934         union ctl_ha_msg msg;
12935         int isc_status;
12936
12937         memset(&msg, 0, sizeof(msg));
12938
12939         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12940         msg.hdr.original_sc = io;
12941         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12942         msg.hdr.nexus = io->io_hdr.nexus;
12943         msg.hdr.status = io->io_hdr.status;
12944         msg.scsi.tag_num = io->scsiio.tag_num;
12945         msg.scsi.tag_type = io->scsiio.tag_type;
12946         msg.scsi.scsi_status = io->scsiio.scsi_status;
12947         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12948                sizeof(io->scsiio.sense_data));
12949         msg.scsi.sense_len = io->scsiio.sense_len;
12950         msg.scsi.sense_residual = io->scsiio.sense_residual;
12951         msg.scsi.fetd_status = io->io_hdr.port_status;
12952         msg.scsi.residual = io->scsiio.residual;
12953         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12954
12955         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12956                 ctl_failover_io(io, /*have_lock*/ have_lock);
12957                 return;
12958         }
12959
12960         isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12961         if (isc_status > CTL_HA_STATUS_SUCCESS) {
12962                 /* XXX do something if this fails */
12963         }
12964
12965 }
12966
12967 /*
12968  * The DMA to the remote side is done, now we need to tell the other side
12969  * we're done so it can continue with its data movement.
12970  */
12971 static void
12972 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12973 {
12974         union ctl_io *io;
12975
12976         io = rq->context;
12977
12978         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12979                 printf("%s: ISC DMA write failed with error %d", __func__,
12980                        rq->ret);
12981                 ctl_set_internal_failure(&io->scsiio,
12982                                          /*sks_valid*/ 1,
12983                                          /*retry_count*/ rq->ret);
12984         }
12985
12986         ctl_dt_req_free(rq);
12987
12988         /*
12989          * In this case, we had to malloc the memory locally.  Free it.
12990          */
12991         if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12992                 int i;
12993                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12994                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12995         }
12996         /*
12997          * The data is in local and remote memory, so now we need to send
12998          * status (good or back) back to the other side.
12999          */
13000         ctl_send_datamove_done(io, /*have_lock*/ 0);
13001 }
13002
13003 /*
13004  * We've moved the data from the host/controller into local memory.  Now we
13005  * need to push it over to the remote controller's memory.
13006  */
13007 static int
13008 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
13009 {
13010         int retval;
13011
13012         retval = 0;
13013
13014         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
13015                                           ctl_datamove_remote_write_cb);
13016
13017         return (retval);
13018 }
13019
13020 static void
13021 ctl_datamove_remote_write(union ctl_io *io)
13022 {
13023         int retval;
13024         void (*fe_datamove)(union ctl_io *io);
13025
13026         /*
13027          * - Get the data from the host/HBA into local memory.
13028          * - DMA memory from the local controller to the remote controller.
13029          * - Send status back to the remote controller.
13030          */
13031
13032         retval = ctl_datamove_remote_sgl_setup(io);
13033         if (retval != 0)
13034                 return;
13035
13036         /* Switch the pointer over so the FETD knows what to do */
13037         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13038
13039         /*
13040          * Use a custom move done callback, since we need to send completion
13041          * back to the other controller, not to the backend on this side.
13042          */
13043         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
13044
13045         fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13046
13047         fe_datamove(io);
13048
13049         return;
13050
13051 }
13052
13053 static int
13054 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
13055 {
13056 #if 0
13057         char str[256];
13058         char path_str[64];
13059         struct sbuf sb;
13060 #endif
13061
13062         /*
13063          * In this case, we had to malloc the memory locally.  Free it.
13064          */
13065         if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13066                 int i;
13067                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13068                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13069         }
13070
13071 #if 0
13072         scsi_path_string(io, path_str, sizeof(path_str));
13073         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13074         sbuf_cat(&sb, path_str);
13075         scsi_command_string(&io->scsiio, NULL, &sb);
13076         sbuf_printf(&sb, "\n");
13077         sbuf_cat(&sb, path_str);
13078         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13079                     io->scsiio.tag_num, io->scsiio.tag_type);
13080         sbuf_cat(&sb, path_str);
13081         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
13082                     io->io_hdr.flags, io->io_hdr.status);
13083         sbuf_finish(&sb);
13084         printk("%s", sbuf_data(&sb));
13085 #endif
13086
13087
13088         /*
13089          * The read is done, now we need to send status (good or bad) back
13090          * to the other side.
13091          */
13092         ctl_send_datamove_done(io, /*have_lock*/ 0);
13093
13094         return (0);
13095 }
13096
13097 static void
13098 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
13099 {
13100         union ctl_io *io;
13101         void (*fe_datamove)(union ctl_io *io);
13102
13103         io = rq->context;
13104
13105         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13106                 printf("%s: ISC DMA read failed with error %d", __func__,
13107                        rq->ret);
13108                 ctl_set_internal_failure(&io->scsiio,
13109                                          /*sks_valid*/ 1,
13110                                          /*retry_count*/ rq->ret);
13111         }
13112
13113         ctl_dt_req_free(rq);
13114
13115         /* Switch the pointer over so the FETD knows what to do */
13116         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13117
13118         /*
13119          * Use a custom move done callback, since we need to send completion
13120          * back to the other controller, not to the backend on this side.
13121          */
13122         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13123
13124         /* XXX KDM add checks like the ones in ctl_datamove? */
13125
13126         fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13127
13128         fe_datamove(io);
13129 }
13130
13131 static int
13132 ctl_datamove_remote_sgl_setup(union ctl_io *io)
13133 {
13134         struct ctl_sg_entry *local_sglist, *remote_sglist;
13135         struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13136         struct ctl_softc *softc;
13137         int retval;
13138         int i;
13139
13140         retval = 0;
13141         softc = control_softc;
13142
13143         local_sglist = io->io_hdr.local_sglist;
13144         local_dma_sglist = io->io_hdr.local_dma_sglist;
13145         remote_sglist = io->io_hdr.remote_sglist;
13146         remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13147
13148         if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13149                 for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13150                         local_sglist[i].len = remote_sglist[i].len;
13151
13152                         /*
13153                          * XXX Detect the situation where the RS-level I/O
13154                          * redirector on the other side has already read the
13155                          * data off of the AOR RS on this side, and
13156                          * transferred it to remote (mirror) memory on the
13157                          * other side.  Since we already have the data in
13158                          * memory here, we just need to use it.
13159                          *
13160                          * XXX KDM this can probably be removed once we
13161                          * get the cache device code in and take the
13162                          * current AOR implementation out.
13163                          */
13164 #ifdef NEEDTOPORT
13165                         if ((remote_sglist[i].addr >=
13166                              (void *)vtophys(softc->mirr->addr))
13167                          && (remote_sglist[i].addr <
13168                              ((void *)vtophys(softc->mirr->addr) +
13169                              CacheMirrorOffset))) {
13170                                 local_sglist[i].addr = remote_sglist[i].addr -
13171                                         CacheMirrorOffset;
13172                                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13173                                      CTL_FLAG_DATA_IN)
13174                                         io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13175                         } else {
13176                                 local_sglist[i].addr = remote_sglist[i].addr +
13177                                         CacheMirrorOffset;
13178                         }
13179 #endif
13180 #if 0
13181                         printf("%s: local %p, remote %p, len %d\n",
13182                                __func__, local_sglist[i].addr,
13183                                remote_sglist[i].addr, local_sglist[i].len);
13184 #endif
13185                 }
13186         } else {
13187                 uint32_t len_to_go;
13188
13189                 /*
13190                  * In this case, we don't have automatically allocated
13191                  * memory for this I/O on this controller.  This typically
13192                  * happens with internal CTL I/O -- e.g. inquiry, mode
13193                  * sense, etc.  Anything coming from RAIDCore will have
13194                  * a mirror area available.
13195                  */
13196                 len_to_go = io->scsiio.kern_data_len;
13197
13198                 /*
13199                  * Clear the no datasync flag, we have to use malloced
13200                  * buffers.
13201                  */
13202                 io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13203
13204                 /*
13205                  * The difficult thing here is that the size of the various
13206                  * S/G segments may be different than the size from the
13207                  * remote controller.  That'll make it harder when DMAing
13208                  * the data back to the other side.
13209                  */
13210                 for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13211                      sizeof(io->io_hdr.remote_sglist[0])) &&
13212                      (len_to_go > 0); i++) {
13213                         local_sglist[i].len = ctl_min(len_to_go, 131072);
13214                         CTL_SIZE_8B(local_dma_sglist[i].len,
13215                                     local_sglist[i].len);
13216                         local_sglist[i].addr =
13217                                 malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13218
13219                         local_dma_sglist[i].addr = local_sglist[i].addr;
13220
13221                         if (local_sglist[i].addr == NULL) {
13222                                 int j;
13223
13224                                 printf("malloc failed for %zd bytes!",
13225                                        local_dma_sglist[i].len);
13226                                 for (j = 0; j < i; j++) {
13227                                         free(local_sglist[j].addr, M_CTL);
13228                                 }
13229                                 ctl_set_internal_failure(&io->scsiio,
13230                                                          /*sks_valid*/ 1,
13231                                                          /*retry_count*/ 4857);
13232                                 retval = 1;
13233                                 goto bailout_error;
13234
13235                         }
13236                         /* XXX KDM do we need a sync here? */
13237
13238                         len_to_go -= local_sglist[i].len;
13239                 }
13240                 /*
13241                  * Reset the number of S/G entries accordingly.  The
13242                  * original number of S/G entries is available in
13243                  * rem_sg_entries.
13244                  */
13245                 io->scsiio.kern_sg_entries = i;
13246
13247 #if 0
13248                 printf("%s: kern_sg_entries = %d\n", __func__,
13249                        io->scsiio.kern_sg_entries);
13250                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13251                         printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13252                                local_sglist[i].addr, local_sglist[i].len,
13253                                local_dma_sglist[i].len);
13254 #endif
13255         }
13256
13257
13258         return (retval);
13259
13260 bailout_error:
13261
13262         ctl_send_datamove_done(io, /*have_lock*/ 0);
13263
13264         return (retval);
13265 }
13266
13267 static int
13268 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13269                          ctl_ha_dt_cb callback)
13270 {
13271         struct ctl_ha_dt_req *rq;
13272         struct ctl_sg_entry *remote_sglist, *local_sglist;
13273         struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13274         uint32_t local_used, remote_used, total_used;
13275         int retval;
13276         int i, j;
13277
13278         retval = 0;
13279
13280         rq = ctl_dt_req_alloc();
13281
13282         /*
13283          * If we failed to allocate the request, and if the DMA didn't fail
13284          * anyway, set busy status.  This is just a resource allocation
13285          * failure.
13286          */
13287         if ((rq == NULL)
13288          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13289                 ctl_set_busy(&io->scsiio);
13290
13291         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13292
13293                 if (rq != NULL)
13294                         ctl_dt_req_free(rq);
13295
13296                 /*
13297                  * The data move failed.  We need to return status back
13298                  * to the other controller.  No point in trying to DMA
13299                  * data to the remote controller.
13300                  */
13301
13302                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13303
13304                 retval = 1;
13305
13306                 goto bailout;
13307         }
13308
13309         local_sglist = io->io_hdr.local_sglist;
13310         local_dma_sglist = io->io_hdr.local_dma_sglist;
13311         remote_sglist = io->io_hdr.remote_sglist;
13312         remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13313         local_used = 0;
13314         remote_used = 0;
13315         total_used = 0;
13316
13317         if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13318                 rq->ret = CTL_HA_STATUS_SUCCESS;
13319                 rq->context = io;
13320                 callback(rq);
13321                 goto bailout;
13322         }
13323
13324         /*
13325          * Pull/push the data over the wire from/to the other controller.
13326          * This takes into account the possibility that the local and
13327          * remote sglists may not be identical in terms of the size of
13328          * the elements and the number of elements.
13329          *
13330          * One fundamental assumption here is that the length allocated for
13331          * both the local and remote sglists is identical.  Otherwise, we've
13332          * essentially got a coding error of some sort.
13333          */
13334         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13335                 int isc_ret;
13336                 uint32_t cur_len, dma_length;
13337                 uint8_t *tmp_ptr;
13338
13339                 rq->id = CTL_HA_DATA_CTL;
13340                 rq->command = command;
13341                 rq->context = io;
13342
13343                 /*
13344                  * Both pointers should be aligned.  But it is possible
13345                  * that the allocation length is not.  They should both
13346                  * also have enough slack left over at the end, though,
13347                  * to round up to the next 8 byte boundary.
13348                  */
13349                 cur_len = ctl_min(local_sglist[i].len - local_used,
13350                                   remote_sglist[j].len - remote_used);
13351
13352                 /*
13353                  * In this case, we have a size issue and need to decrease
13354                  * the size, except in the case where we actually have less
13355                  * than 8 bytes left.  In that case, we need to increase
13356                  * the DMA length to get the last bit.
13357                  */
13358                 if ((cur_len & 0x7) != 0) {
13359                         if (cur_len > 0x7) {
13360                                 cur_len = cur_len - (cur_len & 0x7);
13361                                 dma_length = cur_len;
13362                         } else {
13363                                 CTL_SIZE_8B(dma_length, cur_len);
13364                         }
13365
13366                 } else
13367                         dma_length = cur_len;
13368
13369                 /*
13370                  * If we had to allocate memory for this I/O, instead of using
13371                  * the non-cached mirror memory, we'll need to flush the cache
13372                  * before trying to DMA to the other controller.
13373                  *
13374                  * We could end up doing this multiple times for the same
13375                  * segment if we have a larger local segment than remote
13376                  * segment.  That shouldn't be an issue.
13377                  */
13378                 if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13379                         /*
13380                          * XXX KDM use bus_dmamap_sync() here.
13381                          */
13382                 }
13383
13384                 rq->size = dma_length;
13385
13386                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
13387                 tmp_ptr += local_used;
13388
13389                 /* Use physical addresses when talking to ISC hardware */
13390                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13391                         /* XXX KDM use busdma */
13392 #if 0
13393                         rq->local = vtophys(tmp_ptr);
13394 #endif
13395                 } else
13396                         rq->local = tmp_ptr;
13397
13398                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13399                 tmp_ptr += remote_used;
13400                 rq->remote = tmp_ptr;
13401
13402                 rq->callback = NULL;
13403
13404                 local_used += cur_len;
13405                 if (local_used >= local_sglist[i].len) {
13406                         i++;
13407                         local_used = 0;
13408                 }
13409
13410                 remote_used += cur_len;
13411                 if (remote_used >= remote_sglist[j].len) {
13412                         j++;
13413                         remote_used = 0;
13414                 }
13415                 total_used += cur_len;
13416
13417                 if (total_used >= io->scsiio.kern_data_len)
13418                         rq->callback = callback;
13419
13420                 if ((rq->size & 0x7) != 0) {
13421                         printf("%s: warning: size %d is not on 8b boundary\n",
13422                                __func__, rq->size);
13423                 }
13424                 if (((uintptr_t)rq->local & 0x7) != 0) {
13425                         printf("%s: warning: local %p not on 8b boundary\n",
13426                                __func__, rq->local);
13427                 }
13428                 if (((uintptr_t)rq->remote & 0x7) != 0) {
13429                         printf("%s: warning: remote %p not on 8b boundary\n",
13430                                __func__, rq->local);
13431                 }
13432 #if 0
13433                 printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13434                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13435                        rq->local, rq->remote, rq->size);
13436 #endif
13437
13438                 isc_ret = ctl_dt_single(rq);
13439                 if (isc_ret == CTL_HA_STATUS_WAIT)
13440                         continue;
13441
13442                 if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13443                         rq->ret = CTL_HA_STATUS_SUCCESS;
13444                 } else {
13445                         rq->ret = isc_ret;
13446                 }
13447                 callback(rq);
13448                 goto bailout;
13449         }
13450
13451 bailout:
13452         return (retval);
13453
13454 }
13455
13456 static void
13457 ctl_datamove_remote_read(union ctl_io *io)
13458 {
13459         int retval;
13460         int i;
13461
13462         /*
13463          * This will send an error to the other controller in the case of a
13464          * failure.
13465          */
13466         retval = ctl_datamove_remote_sgl_setup(io);
13467         if (retval != 0)
13468                 return;
13469
13470         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13471                                           ctl_datamove_remote_read_cb);
13472         if ((retval != 0)
13473          && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13474                 /*
13475                  * Make sure we free memory if there was an error..  The
13476                  * ctl_datamove_remote_xfer() function will send the
13477                  * datamove done message, or call the callback with an
13478                  * error if there is a problem.
13479                  */
13480                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13481                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13482         }
13483
13484         return;
13485 }
13486
13487 /*
13488  * Process a datamove request from the other controller.  This is used for
13489  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13490  * first.  Once that is complete, the data gets DMAed into the remote
13491  * controller's memory.  For reads, we DMA from the remote controller's
13492  * memory into our memory first, and then move it out to the FETD.
13493  */
13494 static void
13495 ctl_datamove_remote(union ctl_io *io)
13496 {
13497         struct ctl_softc *softc;
13498
13499         softc = control_softc;
13500
13501         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13502
13503         /*
13504          * Note that we look for an aborted I/O here, but don't do some of
13505          * the other checks that ctl_datamove() normally does.
13506          * We don't need to run the datamove delay code, since that should
13507          * have been done if need be on the other controller.
13508          */
13509         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13510                 printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13511                        io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13512                        io->io_hdr.nexus.targ_port,
13513                        io->io_hdr.nexus.targ_target.id,
13514                        io->io_hdr.nexus.targ_lun);
13515                 io->io_hdr.port_status = 31338;
13516                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13517                 return;
13518         }
13519
13520         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13521                 ctl_datamove_remote_write(io);
13522         } else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13523                 ctl_datamove_remote_read(io);
13524         } else {
13525                 union ctl_ha_msg msg;
13526                 struct scsi_sense_data *sense;
13527                 uint8_t sks[3];
13528                 int retry_count;
13529
13530                 memset(&msg, 0, sizeof(msg));
13531
13532                 msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13533                 msg.hdr.status = CTL_SCSI_ERROR;
13534                 msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13535
13536                 retry_count = 4243;
13537
13538                 sense = &msg.scsi.sense_data;
13539                 sks[0] = SSD_SCS_VALID;
13540                 sks[1] = (retry_count >> 8) & 0xff;
13541                 sks[2] = retry_count & 0xff;
13542
13543                 /* "Internal target failure" */
13544                 scsi_set_sense_data(sense,
13545                                     /*sense_format*/ SSD_TYPE_NONE,
13546                                     /*current_error*/ 1,
13547                                     /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13548                                     /*asc*/ 0x44,
13549                                     /*ascq*/ 0x00,
13550                                     /*type*/ SSD_ELEM_SKS,
13551                                     /*size*/ sizeof(sks),
13552                                     /*data*/ sks,
13553                                     SSD_ELEM_NONE);
13554
13555                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13556                 if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13557                         ctl_failover_io(io, /*have_lock*/ 1);
13558                         return;
13559                 }
13560
13561                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13562                     CTL_HA_STATUS_SUCCESS) {
13563                         /* XXX KDM what to do if this fails? */
13564                 }
13565                 return;
13566         }
13567
13568 }
13569
13570 static int
13571 ctl_process_done(union ctl_io *io)
13572 {
13573         struct ctl_lun *lun;
13574         struct ctl_softc *ctl_softc;
13575         void (*fe_done)(union ctl_io *io);
13576         uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13577
13578         CTL_DEBUG_PRINT(("ctl_process_done\n"));
13579
13580         fe_done =
13581             control_softc->ctl_ports[targ_port]->fe_done;
13582
13583 #ifdef CTL_TIME_IO
13584         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13585                 char str[256];
13586                 char path_str[64];
13587                 struct sbuf sb;
13588
13589                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
13590                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13591
13592                 sbuf_cat(&sb, path_str);
13593                 switch (io->io_hdr.io_type) {
13594                 case CTL_IO_SCSI:
13595                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13596                         sbuf_printf(&sb, "\n");
13597                         sbuf_cat(&sb, path_str);
13598                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13599                                     io->scsiio.tag_num, io->scsiio.tag_type);
13600                         break;
13601                 case CTL_IO_TASK:
13602                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13603                                     "Tag Type: %d\n", io->taskio.task_action,
13604                                     io->taskio.tag_num, io->taskio.tag_type);
13605                         break;
13606                 default:
13607                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13608                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13609                         break;
13610                 }
13611                 sbuf_cat(&sb, path_str);
13612                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13613                             (intmax_t)time_uptime - io->io_hdr.start_time);
13614                 sbuf_finish(&sb);
13615                 printf("%s", sbuf_data(&sb));
13616         }
13617 #endif /* CTL_TIME_IO */
13618
13619         switch (io->io_hdr.io_type) {
13620         case CTL_IO_SCSI:
13621                 break;
13622         case CTL_IO_TASK:
13623                 if (bootverbose || verbose > 0)
13624                         ctl_io_error_print(io, NULL);
13625                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13626                         ctl_free_io(io);
13627                 else
13628                         fe_done(io);
13629                 return (CTL_RETVAL_COMPLETE);
13630                 break;
13631         default:
13632                 printf("ctl_process_done: invalid io type %d\n",
13633                        io->io_hdr.io_type);
13634                 panic("ctl_process_done: invalid io type %d\n",
13635                       io->io_hdr.io_type);
13636                 break; /* NOTREACHED */
13637         }
13638
13639         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13640         if (lun == NULL) {
13641                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13642                                  io->io_hdr.nexus.targ_mapped_lun));
13643                 fe_done(io);
13644                 goto bailout;
13645         }
13646         ctl_softc = lun->ctl_softc;
13647
13648         mtx_lock(&lun->lun_lock);
13649
13650         /*
13651          * Check to see if we have any errors to inject here.  We only
13652          * inject errors for commands that don't already have errors set.
13653          */
13654         if ((STAILQ_FIRST(&lun->error_list) != NULL)
13655          && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13656                 ctl_inject_error(lun, io);
13657
13658         /*
13659          * XXX KDM how do we treat commands that aren't completed
13660          * successfully?
13661          *
13662          * XXX KDM should we also track I/O latency?
13663          */
13664         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13665             io->io_hdr.io_type == CTL_IO_SCSI) {
13666 #ifdef CTL_TIME_IO
13667                 struct bintime cur_bt;
13668 #endif
13669                 int type;
13670
13671                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13672                     CTL_FLAG_DATA_IN)
13673                         type = CTL_STATS_READ;
13674                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13675                     CTL_FLAG_DATA_OUT)
13676                         type = CTL_STATS_WRITE;
13677                 else
13678                         type = CTL_STATS_NO_IO;
13679
13680                 lun->stats.ports[targ_port].bytes[type] +=
13681                     io->scsiio.kern_total_len;
13682                 lun->stats.ports[targ_port].operations[type]++;
13683 #ifdef CTL_TIME_IO
13684                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13685                    &io->io_hdr.dma_bt);
13686                 lun->stats.ports[targ_port].num_dmas[type] +=
13687                     io->io_hdr.num_dmas;
13688                 getbintime(&cur_bt);
13689                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13690                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13691 #endif
13692         }
13693
13694         /*
13695          * Remove this from the OOA queue.
13696          */
13697         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13698
13699         /*
13700          * Run through the blocked queue on this LUN and see if anything
13701          * has become unblocked, now that this transaction is done.
13702          */
13703         ctl_check_blocked(lun);
13704
13705         /*
13706          * If the LUN has been invalidated, free it if there is nothing
13707          * left on its OOA queue.
13708          */
13709         if ((lun->flags & CTL_LUN_INVALID)
13710          && TAILQ_EMPTY(&lun->ooa_queue)) {
13711                 mtx_unlock(&lun->lun_lock);
13712                 mtx_lock(&ctl_softc->ctl_lock);
13713                 ctl_free_lun(lun);
13714                 mtx_unlock(&ctl_softc->ctl_lock);
13715         } else
13716                 mtx_unlock(&lun->lun_lock);
13717
13718         /*
13719          * If this command has been aborted, make sure we set the status
13720          * properly.  The FETD is responsible for freeing the I/O and doing
13721          * whatever it needs to do to clean up its state.
13722          */
13723         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13724                 ctl_set_task_aborted(&io->scsiio);
13725
13726         /*
13727          * We print out status for every task management command.  For SCSI
13728          * commands, we filter out any unit attention errors; they happen
13729          * on every boot, and would clutter up the log.  Note:  task
13730          * management commands aren't printed here, they are printed above,
13731          * since they should never even make it down here.
13732          */
13733         switch (io->io_hdr.io_type) {
13734         case CTL_IO_SCSI: {
13735                 int error_code, sense_key, asc, ascq;
13736
13737                 sense_key = 0;
13738
13739                 if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13740                  && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13741                         /*
13742                          * Since this is just for printing, no need to
13743                          * show errors here.
13744                          */
13745                         scsi_extract_sense_len(&io->scsiio.sense_data,
13746                                                io->scsiio.sense_len,
13747                                                &error_code,
13748                                                &sense_key,
13749                                                &asc,
13750                                                &ascq,
13751                                                /*show_errors*/ 0);
13752                 }
13753
13754                 if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13755                  && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13756                   || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13757                   || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13758
13759                         if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13760                                 ctl_softc->skipped_prints++;
13761                         } else {
13762                                 uint32_t skipped_prints;
13763
13764                                 skipped_prints = ctl_softc->skipped_prints;
13765
13766                                 ctl_softc->skipped_prints = 0;
13767                                 ctl_softc->last_print_jiffies = time_uptime;
13768
13769                                 if (skipped_prints > 0) {
13770 #ifdef NEEDTOPORT
13771                                         csevent_log(CSC_CTL | CSC_SHELF_SW |
13772                                             CTL_ERROR_REPORT,
13773                                             csevent_LogType_Trace,
13774                                             csevent_Severity_Information,
13775                                             csevent_AlertLevel_Green,
13776                                             csevent_FRU_Firmware,
13777                                             csevent_FRU_Unknown,
13778                                             "High CTL error volume, %d prints "
13779                                             "skipped", skipped_prints);
13780 #endif
13781                                 }
13782                                 if (bootverbose || verbose > 0)
13783                                         ctl_io_error_print(io, NULL);
13784                         }
13785                 }
13786                 break;
13787         }
13788         case CTL_IO_TASK:
13789                 if (bootverbose || verbose > 0)
13790                         ctl_io_error_print(io, NULL);
13791                 break;
13792         default:
13793                 break;
13794         }
13795
13796         /*
13797          * Tell the FETD or the other shelf controller we're done with this
13798          * command.  Note that only SCSI commands get to this point.  Task
13799          * management commands are completed above.
13800          *
13801          * We only send status to the other controller if we're in XFER
13802          * mode.  In SER_ONLY mode, the I/O is done on the controller that
13803          * received the I/O (from CTL's perspective), and so the status is
13804          * generated there.
13805          *
13806          * XXX KDM if we hold the lock here, we could cause a deadlock
13807          * if the frontend comes back in in this context to queue
13808          * something.
13809          */
13810         if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13811          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13812                 union ctl_ha_msg msg;
13813
13814                 memset(&msg, 0, sizeof(msg));
13815                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13816                 msg.hdr.original_sc = io->io_hdr.original_sc;
13817                 msg.hdr.nexus = io->io_hdr.nexus;
13818                 msg.hdr.status = io->io_hdr.status;
13819                 msg.scsi.scsi_status = io->scsiio.scsi_status;
13820                 msg.scsi.tag_num = io->scsiio.tag_num;
13821                 msg.scsi.tag_type = io->scsiio.tag_type;
13822                 msg.scsi.sense_len = io->scsiio.sense_len;
13823                 msg.scsi.sense_residual = io->scsiio.sense_residual;
13824                 msg.scsi.residual = io->scsiio.residual;
13825                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13826                        sizeof(io->scsiio.sense_data));
13827                 /*
13828                  * We copy this whether or not this is an I/O-related
13829                  * command.  Otherwise, we'd have to go and check to see
13830                  * whether it's a read/write command, and it really isn't
13831                  * worth it.
13832                  */
13833                 memcpy(&msg.scsi.lbalen,
13834                        &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13835                        sizeof(msg.scsi.lbalen));
13836
13837                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13838                                 sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13839                         /* XXX do something here */
13840                 }
13841
13842                 ctl_free_io(io);
13843         } else
13844                 fe_done(io);
13845
13846 bailout:
13847
13848         return (CTL_RETVAL_COMPLETE);
13849 }
13850
13851 #ifdef CTL_WITH_CA
13852 /*
13853  * Front end should call this if it doesn't do autosense.  When the request
13854  * sense comes back in from the initiator, we'll dequeue this and send it.
13855  */
13856 int
13857 ctl_queue_sense(union ctl_io *io)
13858 {
13859         struct ctl_lun *lun;
13860         struct ctl_softc *ctl_softc;
13861         uint32_t initidx, targ_lun;
13862
13863         ctl_softc = control_softc;
13864
13865         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13866
13867         /*
13868          * LUN lookup will likely move to the ctl_work_thread() once we
13869          * have our new queueing infrastructure (that doesn't put things on
13870          * a per-LUN queue initially).  That is so that we can handle
13871          * things like an INQUIRY to a LUN that we don't have enabled.  We
13872          * can't deal with that right now.
13873          */
13874         mtx_lock(&ctl_softc->ctl_lock);
13875
13876         /*
13877          * If we don't have a LUN for this, just toss the sense
13878          * information.
13879          */
13880         targ_lun = io->io_hdr.nexus.targ_lun;
13881         targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13882         if ((targ_lun < CTL_MAX_LUNS)
13883          && (ctl_softc->ctl_luns[targ_lun] != NULL))
13884                 lun = ctl_softc->ctl_luns[targ_lun];
13885         else
13886                 goto bailout;
13887
13888         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13889
13890         mtx_lock(&lun->lun_lock);
13891         /*
13892          * Already have CA set for this LUN...toss the sense information.
13893          */
13894         if (ctl_is_set(lun->have_ca, initidx)) {
13895                 mtx_unlock(&lun->lun_lock);
13896                 goto bailout;
13897         }
13898
13899         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13900                ctl_min(sizeof(lun->pending_sense[initidx]),
13901                sizeof(io->scsiio.sense_data)));
13902         ctl_set_mask(lun->have_ca, initidx);
13903         mtx_unlock(&lun->lun_lock);
13904
13905 bailout:
13906         mtx_unlock(&ctl_softc->ctl_lock);
13907
13908         ctl_free_io(io);
13909
13910         return (CTL_RETVAL_COMPLETE);
13911 }
13912 #endif
13913
13914 /*
13915  * Primary command inlet from frontend ports.  All SCSI and task I/O
13916  * requests must go through this function.
13917  */
13918 int
13919 ctl_queue(union ctl_io *io)
13920 {
13921         struct ctl_softc *ctl_softc;
13922
13923         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13924
13925         ctl_softc = control_softc;
13926
13927 #ifdef CTL_TIME_IO
13928         io->io_hdr.start_time = time_uptime;
13929         getbintime(&io->io_hdr.start_bt);
13930 #endif /* CTL_TIME_IO */
13931
13932         /* Map FE-specific LUN ID into global one. */
13933         io->io_hdr.nexus.targ_mapped_lun =
13934             ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13935
13936         switch (io->io_hdr.io_type) {
13937         case CTL_IO_SCSI:
13938         case CTL_IO_TASK:
13939                 ctl_enqueue_incoming(io);
13940                 break;
13941         default:
13942                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13943                 return (EINVAL);
13944         }
13945
13946         return (CTL_RETVAL_COMPLETE);
13947 }
13948
13949 #ifdef CTL_IO_DELAY
13950 static void
13951 ctl_done_timer_wakeup(void *arg)
13952 {
13953         union ctl_io *io;
13954
13955         io = (union ctl_io *)arg;
13956         ctl_done(io);
13957 }
13958 #endif /* CTL_IO_DELAY */
13959
13960 void
13961 ctl_done(union ctl_io *io)
13962 {
13963         struct ctl_softc *ctl_softc;
13964
13965         ctl_softc = control_softc;
13966
13967         /*
13968          * Enable this to catch duplicate completion issues.
13969          */
13970 #if 0
13971         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13972                 printf("%s: type %d msg %d cdb %x iptl: "
13973                        "%d:%d:%d:%d tag 0x%04x "
13974                        "flag %#x status %x\n",
13975                         __func__,
13976                         io->io_hdr.io_type,
13977                         io->io_hdr.msg_type,
13978                         io->scsiio.cdb[0],
13979                         io->io_hdr.nexus.initid.id,
13980                         io->io_hdr.nexus.targ_port,
13981                         io->io_hdr.nexus.targ_target.id,
13982                         io->io_hdr.nexus.targ_lun,
13983                         (io->io_hdr.io_type ==
13984                         CTL_IO_TASK) ?
13985                         io->taskio.tag_num :
13986                         io->scsiio.tag_num,
13987                         io->io_hdr.flags,
13988                         io->io_hdr.status);
13989         } else
13990                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13991 #endif
13992
13993         /*
13994          * This is an internal copy of an I/O, and should not go through
13995          * the normal done processing logic.
13996          */
13997         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13998                 return;
13999
14000         /*
14001          * We need to send a msg to the serializing shelf to finish the IO
14002          * as well.  We don't send a finish message to the other shelf if
14003          * this is a task management command.  Task management commands
14004          * aren't serialized in the OOA queue, but rather just executed on
14005          * both shelf controllers for commands that originated on that
14006          * controller.
14007          */
14008         if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
14009          && (io->io_hdr.io_type != CTL_IO_TASK)) {
14010                 union ctl_ha_msg msg_io;
14011
14012                 msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
14013                 msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
14014                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
14015                     sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
14016                 }
14017                 /* continue on to finish IO */
14018         }
14019 #ifdef CTL_IO_DELAY
14020         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
14021                 struct ctl_lun *lun;
14022
14023                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14024
14025                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
14026         } else {
14027                 struct ctl_lun *lun;
14028
14029                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14030
14031                 if ((lun != NULL)
14032                  && (lun->delay_info.done_delay > 0)) {
14033                         struct callout *callout;
14034
14035                         callout = (struct callout *)&io->io_hdr.timer_bytes;
14036                         callout_init(callout, /*mpsafe*/ 1);
14037                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
14038                         callout_reset(callout,
14039                                       lun->delay_info.done_delay * hz,
14040                                       ctl_done_timer_wakeup, io);
14041                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
14042                                 lun->delay_info.done_delay = 0;
14043                         return;
14044                 }
14045         }
14046 #endif /* CTL_IO_DELAY */
14047
14048         ctl_enqueue_done(io);
14049 }
14050
14051 int
14052 ctl_isc(struct ctl_scsiio *ctsio)
14053 {
14054         struct ctl_lun *lun;
14055         int retval;
14056
14057         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14058
14059         CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
14060
14061         CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
14062
14063         retval = lun->backend->data_submit((union ctl_io *)ctsio);
14064
14065         return (retval);
14066 }
14067
14068
14069 static void
14070 ctl_work_thread(void *arg)
14071 {
14072         struct ctl_thread *thr = (struct ctl_thread *)arg;
14073         struct ctl_softc *softc = thr->ctl_softc;
14074         union ctl_io *io;
14075         int retval;
14076
14077         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
14078
14079         for (;;) {
14080                 retval = 0;
14081
14082                 /*
14083                  * We handle the queues in this order:
14084                  * - ISC
14085                  * - done queue (to free up resources, unblock other commands)
14086                  * - RtR queue
14087                  * - incoming queue
14088                  *
14089                  * If those queues are empty, we break out of the loop and
14090                  * go to sleep.
14091                  */
14092                 mtx_lock(&thr->queue_lock);
14093                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
14094                 if (io != NULL) {
14095                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
14096                         mtx_unlock(&thr->queue_lock);
14097                         ctl_handle_isc(io);
14098                         continue;
14099                 }
14100                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
14101                 if (io != NULL) {
14102                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
14103                         /* clear any blocked commands, call fe_done */
14104                         mtx_unlock(&thr->queue_lock);
14105                         retval = ctl_process_done(io);
14106                         continue;
14107                 }
14108                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
14109                 if (io != NULL) {
14110                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
14111                         mtx_unlock(&thr->queue_lock);
14112                         if (io->io_hdr.io_type == CTL_IO_TASK)
14113                                 ctl_run_task(io);
14114                         else
14115                                 ctl_scsiio_precheck(softc, &io->scsiio);
14116                         continue;
14117                 }
14118                 if (!ctl_pause_rtr) {
14119                         io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
14120                         if (io != NULL) {
14121                                 STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
14122                                 mtx_unlock(&thr->queue_lock);
14123                                 retval = ctl_scsiio(&io->scsiio);
14124                                 if (retval != CTL_RETVAL_COMPLETE)
14125                                         CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
14126                                 continue;
14127                         }
14128                 }
14129
14130                 /* Sleep until we have something to do. */
14131                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
14132         }
14133 }
14134
14135 static void
14136 ctl_lun_thread(void *arg)
14137 {
14138         struct ctl_softc *softc = (struct ctl_softc *)arg;
14139         struct ctl_be_lun *be_lun;
14140         int retval;
14141
14142         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14143
14144         for (;;) {
14145                 retval = 0;
14146                 mtx_lock(&softc->ctl_lock);
14147                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14148                 if (be_lun != NULL) {
14149                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14150                         mtx_unlock(&softc->ctl_lock);
14151                         ctl_create_lun(be_lun);
14152                         continue;
14153                 }
14154
14155                 /* Sleep until we have something to do. */
14156                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14157                     PDROP | PRIBIO, "-", 0);
14158         }
14159 }
14160
14161 static void
14162 ctl_enqueue_incoming(union ctl_io *io)
14163 {
14164         struct ctl_softc *softc = control_softc;
14165         struct ctl_thread *thr;
14166         u_int idx;
14167
14168         idx = (io->io_hdr.nexus.targ_port * 127 +
14169                io->io_hdr.nexus.initid.id) % worker_threads;
14170         thr = &softc->threads[idx];
14171         mtx_lock(&thr->queue_lock);
14172         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14173         mtx_unlock(&thr->queue_lock);
14174         wakeup(thr);
14175 }
14176
14177 static void
14178 ctl_enqueue_rtr(union ctl_io *io)
14179 {
14180         struct ctl_softc *softc = control_softc;
14181         struct ctl_thread *thr;
14182
14183         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14184         mtx_lock(&thr->queue_lock);
14185         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14186         mtx_unlock(&thr->queue_lock);
14187         wakeup(thr);
14188 }
14189
14190 static void
14191 ctl_enqueue_done(union ctl_io *io)
14192 {
14193         struct ctl_softc *softc = control_softc;
14194         struct ctl_thread *thr;
14195
14196         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14197         mtx_lock(&thr->queue_lock);
14198         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14199         mtx_unlock(&thr->queue_lock);
14200         wakeup(thr);
14201 }
14202
14203 static void
14204 ctl_enqueue_isc(union ctl_io *io)
14205 {
14206         struct ctl_softc *softc = control_softc;
14207         struct ctl_thread *thr;
14208
14209         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14210         mtx_lock(&thr->queue_lock);
14211         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14212         mtx_unlock(&thr->queue_lock);
14213         wakeup(thr);
14214 }
14215
14216 /* Initialization and failover */
14217
14218 void
14219 ctl_init_isc_msg(void)
14220 {
14221         printf("CTL: Still calling this thing\n");
14222 }
14223
14224 /*
14225  * Init component
14226  *      Initializes component into configuration defined by bootMode
14227  *      (see hasc-sv.c)
14228  *      returns hasc_Status:
14229  *              OK
14230  *              ERROR - fatal error
14231  */
14232 static ctl_ha_comp_status
14233 ctl_isc_init(struct ctl_ha_component *c)
14234 {
14235         ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14236
14237         c->status = ret;
14238         return ret;
14239 }
14240
14241 /* Start component
14242  *      Starts component in state requested. If component starts successfully,
14243  *      it must set its own state to the requestrd state
14244  *      When requested state is HASC_STATE_HA, the component may refine it
14245  *      by adding _SLAVE or _MASTER flags.
14246  *      Currently allowed state transitions are:
14247  *      UNKNOWN->HA             - initial startup
14248  *      UNKNOWN->SINGLE - initial startup when no parter detected
14249  *      HA->SINGLE              - failover
14250  * returns ctl_ha_comp_status:
14251  *              OK      - component successfully started in requested state
14252  *              FAILED  - could not start the requested state, failover may
14253  *                        be possible
14254  *              ERROR   - fatal error detected, no future startup possible
14255  */
14256 static ctl_ha_comp_status
14257 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14258 {
14259         ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14260
14261         printf("%s: go\n", __func__);
14262
14263         // UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14264         if (c->state == CTL_HA_STATE_UNKNOWN ) {
14265                 ctl_is_single = 0;
14266                 if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14267                     != CTL_HA_STATUS_SUCCESS) {
14268                         printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14269                         ret = CTL_HA_COMP_STATUS_ERROR;
14270                 }
14271         } else if (CTL_HA_STATE_IS_HA(c->state)
14272                 && CTL_HA_STATE_IS_SINGLE(state)){
14273                 // HA->SINGLE transition
14274                 ctl_failover();
14275                 ctl_is_single = 1;
14276         } else {
14277                 printf("ctl_isc_start:Invalid state transition %X->%X\n",
14278                        c->state, state);
14279                 ret = CTL_HA_COMP_STATUS_ERROR;
14280         }
14281         if (CTL_HA_STATE_IS_SINGLE(state))
14282                 ctl_is_single = 1;
14283
14284         c->state = state;
14285         c->status = ret;
14286         return ret;
14287 }
14288
14289 /*
14290  * Quiesce component
14291  * The component must clear any error conditions (set status to OK) and
14292  * prepare itself to another Start call
14293  * returns ctl_ha_comp_status:
14294  *      OK
14295  *      ERROR
14296  */
14297 static ctl_ha_comp_status
14298 ctl_isc_quiesce(struct ctl_ha_component *c)
14299 {
14300         int ret = CTL_HA_COMP_STATUS_OK;
14301
14302         ctl_pause_rtr = 1;
14303         c->status = ret;
14304         return ret;
14305 }
14306
14307 struct ctl_ha_component ctl_ha_component_ctlisc =
14308 {
14309         .name = "CTL ISC",
14310         .state = CTL_HA_STATE_UNKNOWN,
14311         .init = ctl_isc_init,
14312         .start = ctl_isc_start,
14313         .quiesce = ctl_isc_quiesce
14314 };
14315
14316 /*
14317  *  vim: ts=8
14318  */