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*/ 0,
 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*/0,
 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*/0,
 296         /*eca_and_aen*/0,
 297         /*flags4*/0,
 298         /*aen_holdoff_period*/{0, 0},
 299         /*busy_timeout_period*/{0, 0},
 300         /*extended_selftest_completion_time*/{0, 0}
 301 };
 302
 303
 304 /*
 305  * XXX KDM move these into the softc.
 306  */
 307 static int rcv_sync_msg;
 308 static int persis_offset;
 309 static uint8_t ctl_pause_rtr;
 310 static int     ctl_is_single = 1;
 311 static int     index_to_aps_page;
 312
 313 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
 314 static int worker_threads = -1;
 315 TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
 316 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 317     &worker_threads, 1, "Number of worker threads");
 318 static int verbose = 0;
 319 TUNABLE_INT("kern.cam.ctl.verbose", &verbose);
 320 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
 321     &verbose, 0, "Show SCSI errors returned to initiator");
 322
 323 /*
 324  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 325  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0) and
 326  * Logical Block Provisioning (0xB2)
 327  */
 328 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   7
 329
 330 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 331                                   int param);
 332 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 333 static int ctl_init(void);
 334 void ctl_shutdown(void);
 335 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 336 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 337 static void ctl_ioctl_online(void *arg);
 338 static void ctl_ioctl_offline(void *arg);
 339 static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
 340 static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
 341 static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
 342 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 343 static int ctl_ioctl_submit_wait(union ctl_io *io);
 344 static void ctl_ioctl_datamove(union ctl_io *io);
 345 static void ctl_ioctl_done(union ctl_io *io);
 346 static void ctl_ioctl_hard_startstop_callback(void *arg,
 347                                               struct cfi_metatask *metatask);
 348 static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
 349 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 350                               struct ctl_ooa *ooa_hdr,
 351                               struct ctl_ooa_entry *kern_entries);
 352 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 353                      struct thread *td);
 354 static uint32_t ctl_map_lun(int port_num, uint32_t lun);
 355 static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
 356 #ifdef unused
 357 static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
 358                                    uint32_t targ_target, uint32_t targ_lun,
 359                                    int can_wait);
 360 static void ctl_kfree_io(union ctl_io *io);
 361 #endif /* unused */
 362 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 363                          struct ctl_be_lun *be_lun, struct ctl_id target_id);
 364 static int ctl_free_lun(struct ctl_lun *lun);
 365 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 366 /**
 367 static void ctl_failover_change_pages(struct ctl_softc *softc,
 368                                       struct ctl_scsiio *ctsio, int master);
 369 **/
 370
 371 static int ctl_do_mode_select(union ctl_io *io);
 372 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 373                            uint64_t res_key, uint64_t sa_res_key,
 374                            uint8_t type, uint32_t residx,
 375                            struct ctl_scsiio *ctsio,
 376                            struct scsi_per_res_out *cdb,
 377                            struct scsi_per_res_out_parms* param);
 378 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 379                                   union ctl_ha_msg *msg);
 380 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
 381 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 382 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 383 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 384 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 385                                          int alloc_len);
 386 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 387                                          int alloc_len);
 388 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 389 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 390 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 391 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
 392 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
 393 static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
 394                                          union ctl_io *ooa_io);
 395 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 396                                 union ctl_io *starting_io);
 397 static int ctl_check_blocked(struct ctl_lun *lun);
 398 static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
 399                                 struct ctl_lun *lun,
 400                                 const struct ctl_cmd_entry *entry,
 401                                 struct ctl_scsiio *ctsio);
 402 //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
 403 static void ctl_failover(void);
 404 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 405                                struct ctl_scsiio *ctsio);
 406 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 407
 408 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 409 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
 410                             ctl_ua_type ua_type);
 411 static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
 412                          ctl_ua_type ua_type);
 413 static int ctl_abort_task(union ctl_io *io);
 414 static int ctl_abort_task_set(union ctl_io *io);
 415 static int ctl_i_t_nexus_reset(union ctl_io *io);
 416 static void ctl_run_task(union ctl_io *io);
 417 #ifdef CTL_IO_DELAY
 418 static void ctl_datamove_timer_wakeup(void *arg);
 419 static void ctl_done_timer_wakeup(void *arg);
 420 #endif /* CTL_IO_DELAY */
 421
 422 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 423 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 424 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 425 static void ctl_datamove_remote_write(union ctl_io *io);
 426 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 427 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 428 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 429 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 430                                     ctl_ha_dt_cb callback);
 431 static void ctl_datamove_remote_read(union ctl_io *io);
 432 static void ctl_datamove_remote(union ctl_io *io);
 433 static int ctl_process_done(union ctl_io *io);
 434 static void ctl_lun_thread(void *arg);
 435 static void ctl_work_thread(void *arg);
 436 static void ctl_enqueue_incoming(union ctl_io *io);
 437 static void ctl_enqueue_rtr(union ctl_io *io);
 438 static void ctl_enqueue_done(union ctl_io *io);
 439 static void ctl_enqueue_isc(union ctl_io *io);
 440 static const struct ctl_cmd_entry *
 441     ctl_get_cmd_entry(struct ctl_scsiio *ctsio);
 442 static const struct ctl_cmd_entry *
 443     ctl_validate_command(struct ctl_scsiio *ctsio);
 444 static int ctl_cmd_applicable(uint8_t lun_type,
 445     const struct ctl_cmd_entry *entry);
 446
 447 /*
 448  * Load the serialization table.  This isn't very pretty, but is probably
 449  * the easiest way to do it.
 450  */
 451 #include "ctl_ser_table.c"
 452
 453 /*
 454  * We only need to define open, close and ioctl routines for this driver.
 455  */
 456 static struct cdevsw ctl_cdevsw = {
 457         .d_version =    D_VERSION,
 458         .d_flags =      0,
 459         .d_open =       ctl_open,
 460         .d_close =      ctl_close,
 461         .d_ioctl =      ctl_ioctl,
 462         .d_name =       "ctl",
 463 };
 464
 465
 466 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 467 MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
 468
 469 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 470
 471 static moduledata_t ctl_moduledata = {
 472         "ctl",
 473         ctl_module_event_handler,
 474         NULL
 475 };
 476
 477 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 478 MODULE_VERSION(ctl, 1);
 479
 480 static struct ctl_frontend ioctl_frontend =
 481 {
 482         .name = "ioctl",
 483 };
 484
 485 static void
 486 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 487                             union ctl_ha_msg *msg_info)
 488 {
 489         struct ctl_scsiio *ctsio;
 490
 491         if (msg_info->hdr.original_sc == NULL) {
 492                 printf("%s: original_sc == NULL!\n", __func__);
 493                 /* XXX KDM now what? */
 494                 return;
 495         }
 496
 497         ctsio = &msg_info->hdr.original_sc->scsiio;
 498         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 499         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 500         ctsio->io_hdr.status = msg_info->hdr.status;
 501         ctsio->scsi_status = msg_info->scsi.scsi_status;
 502         ctsio->sense_len = msg_info->scsi.sense_len;
 503         ctsio->sense_residual = msg_info->scsi.sense_residual;
 504         ctsio->residual = msg_info->scsi.residual;
 505         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 506                sizeof(ctsio->sense_data));
 507         memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
 508                &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
 509         ctl_enqueue_isc((union ctl_io *)ctsio);
 510 }
 511
 512 static void
 513 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 514                                 union ctl_ha_msg *msg_info)
 515 {
 516         struct ctl_scsiio *ctsio;
 517
 518         if (msg_info->hdr.serializing_sc == NULL) {
 519                 printf("%s: serializing_sc == NULL!\n", __func__);
 520                 /* XXX KDM now what? */
 521                 return;
 522         }
 523
 524         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 525 #if 0
 526         /*
 527          * Attempt to catch the situation where an I/O has
 528          * been freed, and we're using it again.
 529          */
 530         if (ctsio->io_hdr.io_type == 0xff) {
 531                 union ctl_io *tmp_io;
 532                 tmp_io = (union ctl_io *)ctsio;
 533                 printf("%s: %p use after free!\n", __func__,
 534                        ctsio);
 535                 printf("%s: type %d msg %d cdb %x iptl: "
 536                        "%d:%d:%d:%d tag 0x%04x "
 537                        "flag %#x status %x\n",
 538                         __func__,
 539                         tmp_io->io_hdr.io_type,
 540                         tmp_io->io_hdr.msg_type,
 541                         tmp_io->scsiio.cdb[0],
 542                         tmp_io->io_hdr.nexus.initid.id,
 543                         tmp_io->io_hdr.nexus.targ_port,
 544                         tmp_io->io_hdr.nexus.targ_target.id,
 545                         tmp_io->io_hdr.nexus.targ_lun,
 546                         (tmp_io->io_hdr.io_type ==
 547                         CTL_IO_TASK) ?
 548                         tmp_io->taskio.tag_num :
 549                         tmp_io->scsiio.tag_num,
 550                         tmp_io->io_hdr.flags,
 551                         tmp_io->io_hdr.status);
 552         }
 553 #endif
 554         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 555         ctl_enqueue_isc((union ctl_io *)ctsio);
 556 }
 557
 558 /*
 559  * ISC (Inter Shelf Communication) event handler.  Events from the HA
 560  * subsystem come in here.
 561  */
 562 static void
 563 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
 564 {
 565         struct ctl_softc *ctl_softc;
 566         union ctl_io *io;
 567         struct ctl_prio *presio;
 568         ctl_ha_status isc_status;
 569
 570         ctl_softc = control_softc;
 571         io = NULL;
 572
 573
 574 #if 0
 575         printf("CTL: Isc Msg event %d\n", event);
 576 #endif
 577         if (event == CTL_HA_EVT_MSG_RECV) {
 578                 union ctl_ha_msg msg_info;
 579
 580                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
 581                                              sizeof(msg_info), /*wait*/ 0);
 582 #if 0
 583                 printf("CTL: msg_type %d\n", msg_info.msg_type);
 584 #endif
 585                 if (isc_status != 0) {
 586                         printf("Error receiving message, status = %d\n",
 587                                isc_status);
 588                         return;
 589                 }
 590
 591                 switch (msg_info.hdr.msg_type) {
 592                 case CTL_MSG_SERIALIZE:
 593 #if 0
 594                         printf("Serialize\n");
 595 #endif
 596                         io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
 597                         if (io == NULL) {
 598                                 printf("ctl_isc_event_handler: can't allocate "
 599                                        "ctl_io!\n");
 600                                 /* Bad Juju */
 601                                 /* Need to set busy and send msg back */
 602                                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
 603                                 msg_info.hdr.status = CTL_SCSI_ERROR;
 604                                 msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
 605                                 msg_info.scsi.sense_len = 0;
 606                                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
 607                                     sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
 608                                 }
 609                                 goto bailout;
 610                         }
 611                         ctl_zero_io(io);
 612                         // populate ctsio from msg_info
 613                         io->io_hdr.io_type = CTL_IO_SCSI;
 614                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
 615                         io->io_hdr.original_sc = msg_info.hdr.original_sc;
 616 #if 0
 617                         printf("pOrig %x\n", (int)msg_info.original_sc);
 618 #endif
 619                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
 620                                             CTL_FLAG_IO_ACTIVE;
 621                         /*
 622                          * If we're in serialization-only mode, we don't
 623                          * want to go through full done processing.  Thus
 624                          * the COPY flag.
 625                          *
 626                          * XXX KDM add another flag that is more specific.
 627                          */
 628                         if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
 629                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
 630                         io->io_hdr.nexus = msg_info.hdr.nexus;
 631 #if 0
 632                         printf("targ %d, port %d, iid %d, lun %d\n",
 633                                io->io_hdr.nexus.targ_target.id,
 634                                io->io_hdr.nexus.targ_port,
 635                                io->io_hdr.nexus.initid.id,
 636                                io->io_hdr.nexus.targ_lun);
 637 #endif
 638                         io->scsiio.tag_num = msg_info.scsi.tag_num;
 639                         io->scsiio.tag_type = msg_info.scsi.tag_type;
 640                         memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
 641                                CTL_MAX_CDBLEN);
 642                         if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
 643                                 const struct ctl_cmd_entry *entry;
 644
 645                                 entry = ctl_get_cmd_entry(&io->scsiio);
 646                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
 647                                 io->io_hdr.flags |=
 648                                         entry->flags & CTL_FLAG_DATA_MASK;
 649                         }
 650                         ctl_enqueue_isc(io);
 651                         break;
 652
 653                 /* Performed on the Originating SC, XFER mode only */
 654                 case CTL_MSG_DATAMOVE: {
 655                         struct ctl_sg_entry *sgl;
 656                         int i, j;
 657
 658                         io = msg_info.hdr.original_sc;
 659                         if (io == NULL) {
 660                                 printf("%s: original_sc == NULL!\n", __func__);
 661                                 /* XXX KDM do something here */
 662                                 break;
 663                         }
 664                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
 665                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 666                         /*
 667                          * Keep track of this, we need to send it back over
 668                          * when the datamove is complete.
 669                          */
 670                         io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
 671
 672                         if (msg_info.dt.sg_sequence == 0) {
 673                                 /*
 674                                  * XXX KDM we use the preallocated S/G list
 675                                  * here, but we'll need to change this to
 676                                  * dynamic allocation if we need larger S/G
 677                                  * lists.
 678                                  */
 679                                 if (msg_info.dt.kern_sg_entries >
 680                                     sizeof(io->io_hdr.remote_sglist) /
 681                                     sizeof(io->io_hdr.remote_sglist[0])) {
 682                                         printf("%s: number of S/G entries "
 683                                             "needed %u > allocated num %zd\n",
 684                                             __func__,
 685                                             msg_info.dt.kern_sg_entries,
 686                                             sizeof(io->io_hdr.remote_sglist)/
 687                                             sizeof(io->io_hdr.remote_sglist[0]));
 688
 689                                         /*
 690                                          * XXX KDM send a message back to
 691                                          * the other side to shut down the
 692                                          * DMA.  The error will come back
 693                                          * through via the normal channel.
 694                                          */
 695                                         break;
 696                                 }
 697                                 sgl = io->io_hdr.remote_sglist;
 698                                 memset(sgl, 0,
 699                                        sizeof(io->io_hdr.remote_sglist));
 700
 701                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
 702
 703                                 io->scsiio.kern_sg_entries =
 704                                         msg_info.dt.kern_sg_entries;
 705                                 io->scsiio.rem_sg_entries =
 706                                         msg_info.dt.kern_sg_entries;
 707                                 io->scsiio.kern_data_len =
 708                                         msg_info.dt.kern_data_len;
 709                                 io->scsiio.kern_total_len =
 710                                         msg_info.dt.kern_total_len;
 711                                 io->scsiio.kern_data_resid =
 712                                         msg_info.dt.kern_data_resid;
 713                                 io->scsiio.kern_rel_offset =
 714                                         msg_info.dt.kern_rel_offset;
 715                                 /*
 716                                  * Clear out per-DMA flags.
 717                                  */
 718                                 io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
 719                                 /*
 720                                  * Add per-DMA flags that are set for this
 721                                  * particular DMA request.
 722                                  */
 723                                 io->io_hdr.flags |= msg_info.dt.flags &
 724                                                     CTL_FLAG_RDMA_MASK;
 725                         } else
 726                                 sgl = (struct ctl_sg_entry *)
 727                                         io->scsiio.kern_data_ptr;
 728
 729                         for (i = msg_info.dt.sent_sg_entries, j = 0;
 730                              i < (msg_info.dt.sent_sg_entries +
 731                              msg_info.dt.cur_sg_entries); i++, j++) {
 732                                 sgl[i].addr = msg_info.dt.sg_list[j].addr;
 733                                 sgl[i].len = msg_info.dt.sg_list[j].len;
 734
 735 #if 0
 736                                 printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
 737                                        __func__,
 738                                        msg_info.dt.sg_list[j].addr,
 739                                        msg_info.dt.sg_list[j].len,
 740                                        sgl[i].addr, sgl[i].len, j, i);
 741 #endif
 742                         }
 743 #if 0
 744                         memcpy(&sgl[msg_info.dt.sent_sg_entries],
 745                                msg_info.dt.sg_list,
 746                                sizeof(*sgl) * msg_info.dt.cur_sg_entries);
 747 #endif
 748
 749                         /*
 750                          * If this is the last piece of the I/O, we've got
 751                          * the full S/G list.  Queue processing in the thread.
 752                          * Otherwise wait for the next piece.
 753                          */
 754                         if (msg_info.dt.sg_last != 0)
 755                                 ctl_enqueue_isc(io);
 756                         break;
 757                 }
 758                 /* Performed on the Serializing (primary) SC, XFER mode only */
 759                 case CTL_MSG_DATAMOVE_DONE: {
 760                         if (msg_info.hdr.serializing_sc == NULL) {
 761                                 printf("%s: serializing_sc == NULL!\n",
 762                                        __func__);
 763                                 /* XXX KDM now what? */
 764                                 break;
 765                         }
 766                         /*
 767                          * We grab the sense information here in case
 768                          * there was a failure, so we can return status
 769                          * back to the initiator.
 770                          */
 771                         io = msg_info.hdr.serializing_sc;
 772                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
 773                         io->io_hdr.status = msg_info.hdr.status;
 774                         io->scsiio.scsi_status = msg_info.scsi.scsi_status;
 775                         io->scsiio.sense_len = msg_info.scsi.sense_len;
 776                         io->scsiio.sense_residual =msg_info.scsi.sense_residual;
 777                         io->io_hdr.port_status = msg_info.scsi.fetd_status;
 778                         io->scsiio.residual = msg_info.scsi.residual;
 779                         memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
 780                                sizeof(io->scsiio.sense_data));
 781                         ctl_enqueue_isc(io);
 782                         break;
 783                 }
 784
 785                 /* Preformed on Originating SC, SER_ONLY mode */
 786                 case CTL_MSG_R2R:
 787                         io = msg_info.hdr.original_sc;
 788                         if (io == NULL) {
 789                                 printf("%s: Major Bummer\n", __func__);
 790                                 return;
 791                         } else {
 792 #if 0
 793                                 printf("pOrig %x\n",(int) ctsio);
 794 #endif
 795                         }
 796                         io->io_hdr.msg_type = CTL_MSG_R2R;
 797                         io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
 798                         ctl_enqueue_isc(io);
 799                         break;
 800
 801                 /*
 802                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
 803                  * mode.
 804                  * Performed on the Originating (i.e. secondary) SC in XFER
 805                  * mode
 806                  */
 807                 case CTL_MSG_FINISH_IO:
 808                         if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
 809                                 ctl_isc_handler_finish_xfer(ctl_softc,
 810                                                             &msg_info);
 811                         else
 812                                 ctl_isc_handler_finish_ser_only(ctl_softc,
 813                                                                 &msg_info);
 814                         break;
 815
 816                 /* Preformed on Originating SC */
 817                 case CTL_MSG_BAD_JUJU:
 818                         io = msg_info.hdr.original_sc;
 819                         if (io == NULL) {
 820                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
 821                                        __func__);
 822                                 break;
 823                         }
 824                         ctl_copy_sense_data(&msg_info, io);
 825                         /*
 826                          * IO should have already been cleaned up on other
 827                          * SC so clear this flag so we won't send a message
 828                          * back to finish the IO there.
 829                          */
 830                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
 831                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 832
 833                         /* io = msg_info.hdr.serializing_sc; */
 834                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
 835                         ctl_enqueue_isc(io);
 836                         break;
 837
 838                 /* Handle resets sent from the other side */
 839                 case CTL_MSG_MANAGE_TASKS: {
 840                         struct ctl_taskio *taskio;
 841                         taskio = (struct ctl_taskio *)ctl_alloc_io(
 842                                 (void *)ctl_softc->othersc_pool);
 843                         if (taskio == NULL) {
 844                                 printf("ctl_isc_event_handler: can't allocate "
 845                                        "ctl_io!\n");
 846                                 /* Bad Juju */
 847                                 /* should I just call the proper reset func
 848                                    here??? */
 849                                 goto bailout;
 850                         }
 851                         ctl_zero_io((union ctl_io *)taskio);
 852                         taskio->io_hdr.io_type = CTL_IO_TASK;
 853                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
 854                         taskio->io_hdr.nexus = msg_info.hdr.nexus;
 855                         taskio->task_action = msg_info.task.task_action;
 856                         taskio->tag_num = msg_info.task.tag_num;
 857                         taskio->tag_type = msg_info.task.tag_type;
 858 #ifdef CTL_TIME_IO
 859                         taskio->io_hdr.start_time = time_uptime;
 860                         getbintime(&taskio->io_hdr.start_bt);
 861 #if 0
 862                         cs_prof_gettime(&taskio->io_hdr.start_ticks);
 863 #endif
 864 #endif /* CTL_TIME_IO */
 865                         ctl_run_task((union ctl_io *)taskio);
 866                         break;
 867                 }
 868                 /* Persistent Reserve action which needs attention */
 869                 case CTL_MSG_PERS_ACTION:
 870                         presio = (struct ctl_prio *)ctl_alloc_io(
 871                                 (void *)ctl_softc->othersc_pool);
 872                         if (presio == NULL) {
 873                                 printf("ctl_isc_event_handler: can't allocate "
 874                                        "ctl_io!\n");
 875                                 /* Bad Juju */
 876                                 /* Need to set busy and send msg back */
 877                                 goto bailout;
 878                         }
 879                         ctl_zero_io((union ctl_io *)presio);
 880                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
 881                         presio->pr_msg = msg_info.pr;
 882                         ctl_enqueue_isc((union ctl_io *)presio);
 883                         break;
 884                 case CTL_MSG_SYNC_FE:
 885                         rcv_sync_msg = 1;
 886                         break;
 887                 case CTL_MSG_APS_LOCK: {
 888                         // It's quicker to execute this then to
 889                         // queue it.
 890                         struct ctl_lun *lun;
 891                         struct ctl_page_index *page_index;
 892                         struct copan_aps_subpage *current_sp;
 893                         uint32_t targ_lun;
 894
 895                         targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
 896                         lun = ctl_softc->ctl_luns[targ_lun];
 897                         mtx_lock(&lun->lun_lock);
 898                         page_index = &lun->mode_pages.index[index_to_aps_page];
 899                         current_sp = (struct copan_aps_subpage *)
 900                                      (page_index->page_data +
 901                                      (page_index->page_len * CTL_PAGE_CURRENT));
 902
 903                         current_sp->lock_active = msg_info.aps.lock_flag;
 904                         mtx_unlock(&lun->lun_lock);
 905                         break;
 906                 }
 907                 default:
 908                         printf("How did I get here?\n");
 909                 }
 910         } else if (event == CTL_HA_EVT_MSG_SENT) {
 911                 if (param != CTL_HA_STATUS_SUCCESS) {
 912                         printf("Bad status from ctl_ha_msg_send status %d\n",
 913                                param);
 914                 }
 915                 return;
 916         } else if (event == CTL_HA_EVT_DISCONNECT) {
 917                 printf("CTL: Got a disconnect from Isc\n");
 918                 return;
 919         } else {
 920                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
 921                 return;
 922         }
 923
 924 bailout:
 925         return;
 926 }
 927
 928 static void
 929 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
 930 {
 931         struct scsi_sense_data *sense;
 932
 933         sense = &dest->scsiio.sense_data;
 934         bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
 935         dest->scsiio.scsi_status = src->scsi.scsi_status;
 936         dest->scsiio.sense_len = src->scsi.sense_len;
 937         dest->io_hdr.status = src->hdr.status;
 938 }
 939
 940 static int
 941 ctl_init(void)
 942 {
 943         struct ctl_softc *softc;
 944         struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
 945         struct ctl_port *port;
 946         uint8_t sc_id =0;
 947         int i, error, retval;
 948         //int isc_retval;
 949
 950         retval = 0;
 951         ctl_pause_rtr = 0;
 952         rcv_sync_msg = 0;
 953
 954         control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
 955                                M_WAITOK | M_ZERO);
 956         softc = control_softc;
 957
 958         softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
 959                               "cam/ctl");
 960
 961         softc->dev->si_drv1 = softc;
 962
 963         /*
 964          * By default, return a "bad LUN" peripheral qualifier for unknown
 965          * LUNs.  The user can override this default using the tunable or
 966          * sysctl.  See the comment in ctl_inquiry_std() for more details.
 967          */
 968         softc->inquiry_pq_no_lun = 1;
 969         TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
 970                           &softc->inquiry_pq_no_lun);
 971         sysctl_ctx_init(&softc->sysctl_ctx);
 972         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
 973                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
 974                 CTLFLAG_RD, 0, "CAM Target Layer");
 975
 976         if (softc->sysctl_tree == NULL) {
 977                 printf("%s: unable to allocate sysctl tree\n", __func__);
 978                 destroy_dev(softc->dev);
 979                 free(control_softc, M_DEVBUF);
 980                 control_softc = NULL;
 981                 return (ENOMEM);
 982         }
 983
 984         SYSCTL_ADD_INT(&softc->sysctl_ctx,
 985                        SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
 986                        "inquiry_pq_no_lun", CTLFLAG_RW,
 987                        &softc->inquiry_pq_no_lun, 0,
 988                        "Report no lun possible for invalid LUNs");
 989
 990         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
 991         mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
 992         softc->open_count = 0;
 993
 994         /*
 995          * Default to actually sending a SYNCHRONIZE CACHE command down to
 996          * the drive.
 997          */
 998         softc->flags = CTL_FLAG_REAL_SYNC;
 999
1000         /*
1001          * In Copan's HA scheme, the "master" and "slave" roles are
1002          * figured out through the slot the controller is in.  Although it
1003          * is an active/active system, someone has to be in charge.
1004          */
1005 #ifdef NEEDTOPORT
1006         scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1007 #endif
1008
1009         if (sc_id == 0) {
1010                 softc->flags |= CTL_FLAG_MASTER_SHELF;
1011                 persis_offset = 0;
1012         } else
1013                 persis_offset = CTL_MAX_INITIATORS;
1014
1015         /*
1016          * XXX KDM need to figure out where we want to get our target ID
1017          * and WWID.  Is it different on each port?
1018          */
1019         softc->target.id = 0;
1020         softc->target.wwid[0] = 0x12345678;
1021         softc->target.wwid[1] = 0x87654321;
1022         STAILQ_INIT(&softc->lun_list);
1023         STAILQ_INIT(&softc->pending_lun_queue);
1024         STAILQ_INIT(&softc->fe_list);
1025         STAILQ_INIT(&softc->port_list);
1026         STAILQ_INIT(&softc->be_list);
1027         STAILQ_INIT(&softc->io_pools);
1028         ctl_tpc_init(softc);
1029
1030         if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1031                             &internal_pool)!= 0){
1032                 printf("ctl: can't allocate %d entry internal pool, "
1033                        "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1034                 return (ENOMEM);
1035         }
1036
1037         if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1038                             CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1039                 printf("ctl: can't allocate %d entry emergency pool, "
1040                        "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1041                 ctl_pool_free(internal_pool);
1042                 return (ENOMEM);
1043         }
1044
1045         if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1046                             &other_pool) != 0)
1047         {
1048                 printf("ctl: can't allocate %d entry other SC pool, "
1049                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1050                 ctl_pool_free(internal_pool);
1051                 ctl_pool_free(emergency_pool);
1052                 return (ENOMEM);
1053         }
1054
1055         softc->internal_pool = internal_pool;
1056         softc->emergency_pool = emergency_pool;
1057         softc->othersc_pool = other_pool;
1058
1059         if (worker_threads <= 0)
1060                 worker_threads = max(1, mp_ncpus / 4);
1061         if (worker_threads > CTL_MAX_THREADS)
1062                 worker_threads = CTL_MAX_THREADS;
1063
1064         for (i = 0; i < worker_threads; i++) {
1065                 struct ctl_thread *thr = &softc->threads[i];
1066
1067                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1068                 thr->ctl_softc = softc;
1069                 STAILQ_INIT(&thr->incoming_queue);
1070                 STAILQ_INIT(&thr->rtr_queue);
1071                 STAILQ_INIT(&thr->done_queue);
1072                 STAILQ_INIT(&thr->isc_queue);
1073
1074                 error = kproc_kthread_add(ctl_work_thread, thr,
1075                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1076                 if (error != 0) {
1077                         printf("error creating CTL work thread!\n");
1078                         ctl_pool_free(internal_pool);
1079                         ctl_pool_free(emergency_pool);
1080                         ctl_pool_free(other_pool);
1081                         return (error);
1082                 }
1083         }
1084         error = kproc_kthread_add(ctl_lun_thread, softc,
1085             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1086         if (error != 0) {
1087                 printf("error creating CTL lun thread!\n");
1088                 ctl_pool_free(internal_pool);
1089                 ctl_pool_free(emergency_pool);
1090                 ctl_pool_free(other_pool);
1091                 return (error);
1092         }
1093         if (bootverbose)
1094                 printf("ctl: CAM Target Layer loaded\n");
1095
1096         /*
1097          * Initialize the ioctl front end.
1098          */
1099         ctl_frontend_register(&ioctl_frontend);
1100         port = &softc->ioctl_info.port;
1101         port->frontend = &ioctl_frontend;
1102         sprintf(softc->ioctl_info.port_name, "ioctl");
1103         port->port_type = CTL_PORT_IOCTL;
1104         port->num_requested_ctl_io = 100;
1105         port->port_name = softc->ioctl_info.port_name;
1106         port->port_online = ctl_ioctl_online;
1107         port->port_offline = ctl_ioctl_offline;
1108         port->onoff_arg = &softc->ioctl_info;
1109         port->lun_enable = ctl_ioctl_lun_enable;
1110         port->lun_disable = ctl_ioctl_lun_disable;
1111         port->targ_lun_arg = &softc->ioctl_info;
1112         port->fe_datamove = ctl_ioctl_datamove;
1113         port->fe_done = ctl_ioctl_done;
1114         port->max_targets = 15;
1115         port->max_target_id = 15;
1116
1117         if (ctl_port_register(&softc->ioctl_info.port,
1118                           (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1119                 printf("ctl: ioctl front end registration failed, will "
1120                        "continue anyway\n");
1121         }
1122
1123 #ifdef CTL_IO_DELAY
1124         if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1125                 printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1126                        sizeof(struct callout), CTL_TIMER_BYTES);
1127                 return (EINVAL);
1128         }
1129 #endif /* CTL_IO_DELAY */
1130
1131         return (0);
1132 }
1133
1134 void
1135 ctl_shutdown(void)
1136 {
1137         struct ctl_softc *softc;
1138         struct ctl_lun *lun, *next_lun;
1139         struct ctl_io_pool *pool;
1140
1141         softc = (struct ctl_softc *)control_softc;
1142
1143         if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1144                 printf("ctl: ioctl front end deregistration failed\n");
1145
1146         mtx_lock(&softc->ctl_lock);
1147
1148         /*
1149          * Free up each LUN.
1150          */
1151         for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1152                 next_lun = STAILQ_NEXT(lun, links);
1153                 ctl_free_lun(lun);
1154         }
1155
1156         mtx_unlock(&softc->ctl_lock);
1157
1158         ctl_frontend_deregister(&ioctl_frontend);
1159
1160         /*
1161          * This will rip the rug out from under any FETDs or anyone else
1162          * that has a pool allocated.  Since we increment our module
1163          * refcount any time someone outside the main CTL module allocates
1164          * a pool, we shouldn't have any problems here.  The user won't be
1165          * able to unload the CTL module until client modules have
1166          * successfully unloaded.
1167          */
1168         while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1169                 ctl_pool_free(pool);
1170
1171 #if 0
1172         ctl_shutdown_thread(softc->work_thread);
1173         mtx_destroy(&softc->queue_lock);
1174 #endif
1175
1176         ctl_tpc_shutdown(softc);
1177         mtx_destroy(&softc->pool_lock);
1178         mtx_destroy(&softc->ctl_lock);
1179
1180         destroy_dev(softc->dev);
1181
1182         sysctl_ctx_free(&softc->sysctl_ctx);
1183
1184         free(control_softc, M_DEVBUF);
1185         control_softc = NULL;
1186
1187         if (bootverbose)
1188                 printf("ctl: CAM Target Layer unloaded\n");
1189 }
1190
1191 static int
1192 ctl_module_event_handler(module_t mod, int what, void *arg)
1193 {
1194
1195         switch (what) {
1196         case MOD_LOAD:
1197                 return (ctl_init());
1198         case MOD_UNLOAD:
1199                 return (EBUSY);
1200         default:
1201                 return (EOPNOTSUPP);
1202         }
1203 }
1204
1205 /*
1206  * XXX KDM should we do some access checks here?  Bump a reference count to
1207  * prevent a CTL module from being unloaded while someone has it open?
1208  */
1209 static int
1210 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1211 {
1212         return (0);
1213 }
1214
1215 static int
1216 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1217 {
1218         return (0);
1219 }
1220
1221 int
1222 ctl_port_enable(ctl_port_type port_type)
1223 {
1224         struct ctl_softc *softc;
1225         struct ctl_port *port;
1226
1227         if (ctl_is_single == 0) {
1228                 union ctl_ha_msg msg_info;
1229                 int isc_retval;
1230
1231 #if 0
1232                 printf("%s: HA mode, synchronizing frontend enable\n",
1233                         __func__);
1234 #endif
1235                 msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1236                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1237                         sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1238                         printf("Sync msg send error retval %d\n", isc_retval);
1239                 }
1240                 if (!rcv_sync_msg) {
1241                         isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1242                                 sizeof(msg_info), 1);
1243                 }
1244 #if 0
1245                 printf("CTL:Frontend Enable\n");
1246         } else {
1247                 printf("%s: single mode, skipping frontend synchronization\n",
1248                         __func__);
1249 #endif
1250         }
1251
1252         softc = control_softc;
1253
1254         STAILQ_FOREACH(port, &softc->port_list, links) {
1255                 if (port_type & port->port_type)
1256                 {
1257 #if 0
1258                         printf("port %d\n", port->targ_port);
1259 #endif
1260                         ctl_port_online(port);
1261                 }
1262         }
1263
1264         return (0);
1265 }
1266
1267 int
1268 ctl_port_disable(ctl_port_type port_type)
1269 {
1270         struct ctl_softc *softc;
1271         struct ctl_port *port;
1272
1273         softc = control_softc;
1274
1275         STAILQ_FOREACH(port, &softc->port_list, links) {
1276                 if (port_type & port->port_type)
1277                         ctl_port_offline(port);
1278         }
1279
1280         return (0);
1281 }
1282
1283 /*
1284  * Returns 0 for success, 1 for failure.
1285  * Currently the only failure mode is if there aren't enough entries
1286  * allocated.  So, in case of a failure, look at num_entries_dropped,
1287  * reallocate and try again.
1288  */
1289 int
1290 ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1291               int *num_entries_filled, int *num_entries_dropped,
1292               ctl_port_type port_type, int no_virtual)
1293 {
1294         struct ctl_softc *softc;
1295         struct ctl_port *port;
1296         int entries_dropped, entries_filled;
1297         int retval;
1298         int i;
1299
1300         softc = control_softc;
1301
1302         retval = 0;
1303         entries_filled = 0;
1304         entries_dropped = 0;
1305
1306         i = 0;
1307         mtx_lock(&softc->ctl_lock);
1308         STAILQ_FOREACH(port, &softc->port_list, links) {
1309                 struct ctl_port_entry *entry;
1310
1311                 if ((port->port_type & port_type) == 0)
1312                         continue;
1313
1314                 if ((no_virtual != 0)
1315                  && (port->virtual_port != 0))
1316                         continue;
1317
1318                 if (entries_filled >= num_entries_alloced) {
1319                         entries_dropped++;
1320                         continue;
1321                 }
1322                 entry = &entries[i];
1323
1324                 entry->port_type = port->port_type;
1325                 strlcpy(entry->port_name, port->port_name,
1326                         sizeof(entry->port_name));
1327                 entry->physical_port = port->physical_port;
1328                 entry->virtual_port = port->virtual_port;
1329                 entry->wwnn = port->wwnn;
1330                 entry->wwpn = port->wwpn;
1331
1332                 i++;
1333                 entries_filled++;
1334         }
1335
1336         mtx_unlock(&softc->ctl_lock);
1337
1338         if (entries_dropped > 0)
1339                 retval = 1;
1340
1341         *num_entries_dropped = entries_dropped;
1342         *num_entries_filled = entries_filled;
1343
1344         return (retval);
1345 }
1346
1347 static void
1348 ctl_ioctl_online(void *arg)
1349 {
1350         struct ctl_ioctl_info *ioctl_info;
1351
1352         ioctl_info = (struct ctl_ioctl_info *)arg;
1353
1354         ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1355 }
1356
1357 static void
1358 ctl_ioctl_offline(void *arg)
1359 {
1360         struct ctl_ioctl_info *ioctl_info;
1361
1362         ioctl_info = (struct ctl_ioctl_info *)arg;
1363
1364         ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1365 }
1366
1367 /*
1368  * Remove an initiator by port number and initiator ID.
1369  * Returns 0 for success, -1 for failure.
1370  */
1371 int
1372 ctl_remove_initiator(struct ctl_port *port, int iid)
1373 {
1374         struct ctl_softc *softc = control_softc;
1375
1376         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1377
1378         if (iid > CTL_MAX_INIT_PER_PORT) {
1379                 printf("%s: initiator ID %u > maximun %u!\n",
1380                        __func__, iid, CTL_MAX_INIT_PER_PORT);
1381                 return (-1);
1382         }
1383
1384         mtx_lock(&softc->ctl_lock);
1385         port->wwpn_iid[iid].in_use--;
1386         port->wwpn_iid[iid].last_use = time_uptime;
1387         mtx_unlock(&softc->ctl_lock);
1388
1389         return (0);
1390 }
1391
1392 /*
1393  * Add an initiator to the initiator map.
1394  * Returns iid for success, < 0 for failure.
1395  */
1396 int
1397 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1398 {
1399         struct ctl_softc *softc = control_softc;
1400         time_t best_time;
1401         int i, best;
1402
1403         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1404
1405         if (iid >= CTL_MAX_INIT_PER_PORT) {
1406                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1407                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1408                 free(name, M_CTL);
1409                 return (-1);
1410         }
1411
1412         mtx_lock(&softc->ctl_lock);
1413
1414         if (iid < 0 && (wwpn != 0 || name != NULL)) {
1415                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1416                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1417                                 iid = i;
1418                                 break;
1419                         }
1420                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
1421                             strcmp(name, port->wwpn_iid[i].name) == 0) {
1422                                 iid = i;
1423                                 break;
1424                         }
1425                 }
1426         }
1427
1428         if (iid < 0) {
1429                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1430                         if (port->wwpn_iid[i].in_use == 0 &&
1431                             port->wwpn_iid[i].wwpn == 0 &&
1432                             port->wwpn_iid[i].name == NULL) {
1433                                 iid = i;
1434                                 break;
1435                         }
1436                 }
1437         }
1438
1439         if (iid < 0) {
1440                 best = -1;
1441                 best_time = INT32_MAX;
1442                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1443                         if (port->wwpn_iid[i].in_use == 0) {
1444                                 if (port->wwpn_iid[i].last_use < best_time) {
1445                                         best = i;
1446                                         best_time = port->wwpn_iid[i].last_use;
1447                                 }
1448                         }
1449                 }
1450                 iid = best;
1451         }
1452
1453         if (iid < 0) {
1454                 mtx_unlock(&softc->ctl_lock);
1455                 free(name, M_CTL);
1456                 return (-2);
1457         }
1458
1459         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1460                 /*
1461                  * This is not an error yet.
1462                  */
1463                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1464 #if 0
1465                         printf("%s: port %d iid %u WWPN %#jx arrived"
1466                             " again\n", __func__, port->targ_port,
1467                             iid, (uintmax_t)wwpn);
1468 #endif
1469                         goto take;
1470                 }
1471                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1472                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
1473 #if 0
1474                         printf("%s: port %d iid %u name '%s' arrived"
1475                             " again\n", __func__, port->targ_port,
1476                             iid, name);
1477 #endif
1478                         goto take;
1479                 }
1480
1481                 /*
1482                  * This is an error, but what do we do about it?  The
1483                  * driver is telling us we have a new WWPN for this
1484                  * initiator ID, so we pretty much need to use it.
1485                  */
1486                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1487                     " but WWPN %#jx '%s' is still at that address\n",
1488                     __func__, port->targ_port, iid, wwpn, name,
1489                     (uintmax_t)port->wwpn_iid[iid].wwpn,
1490                     port->wwpn_iid[iid].name);
1491
1492                 /*
1493                  * XXX KDM clear have_ca and ua_pending on each LUN for
1494                  * this initiator.
1495                  */
1496         }
1497 take:
1498         free(port->wwpn_iid[iid].name, M_CTL);
1499         port->wwpn_iid[iid].name = name;
1500         port->wwpn_iid[iid].wwpn = wwpn;
1501         port->wwpn_iid[iid].in_use++;
1502         mtx_unlock(&softc->ctl_lock);
1503
1504         return (iid);
1505 }
1506
1507 static int
1508 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1509 {
1510         int len;
1511
1512         switch (port->port_type) {
1513         case CTL_PORT_FC:
1514         {
1515                 struct scsi_transportid_fcp *id =
1516                     (struct scsi_transportid_fcp *)buf;
1517                 if (port->wwpn_iid[iid].wwpn == 0)
1518                         return (0);
1519                 memset(id, 0, sizeof(*id));
1520                 id->format_protocol = SCSI_PROTO_FC;
1521                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1522                 return (sizeof(*id));
1523         }
1524         case CTL_PORT_ISCSI:
1525         {
1526                 struct scsi_transportid_iscsi_port *id =
1527                     (struct scsi_transportid_iscsi_port *)buf;
1528                 if (port->wwpn_iid[iid].name == NULL)
1529                         return (0);
1530                 memset(id, 0, 256);
1531                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1532                     SCSI_PROTO_ISCSI;
1533                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1534                 len = roundup2(min(len, 252), 4);
1535                 scsi_ulto2b(len, id->additional_length);
1536                 return (sizeof(*id) + len);
1537         }
1538         case CTL_PORT_SAS:
1539         {
1540                 struct scsi_transportid_sas *id =
1541                     (struct scsi_transportid_sas *)buf;
1542                 if (port->wwpn_iid[iid].wwpn == 0)
1543                         return (0);
1544                 memset(id, 0, sizeof(*id));
1545                 id->format_protocol = SCSI_PROTO_SAS;
1546                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1547                 return (sizeof(*id));
1548         }
1549         default:
1550         {
1551                 struct scsi_transportid_spi *id =
1552                     (struct scsi_transportid_spi *)buf;
1553                 memset(id, 0, sizeof(*id));
1554                 id->format_protocol = SCSI_PROTO_SPI;
1555                 scsi_ulto2b(iid, id->scsi_addr);
1556                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1557                 return (sizeof(*id));
1558         }
1559         }
1560 }
1561
1562 static int
1563 ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1564 {
1565         return (0);
1566 }
1567
1568 static int
1569 ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1570 {
1571         return (0);
1572 }
1573
1574 /*
1575  * Data movement routine for the CTL ioctl frontend port.
1576  */
1577 static int
1578 ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1579 {
1580         struct ctl_sg_entry *ext_sglist, *kern_sglist;
1581         struct ctl_sg_entry ext_entry, kern_entry;
1582         int ext_sglen, ext_sg_entries, kern_sg_entries;
1583         int ext_sg_start, ext_offset;
1584         int len_to_copy, len_copied;
1585         int kern_watermark, ext_watermark;
1586         int ext_sglist_malloced;
1587         int i, j;
1588
1589         ext_sglist_malloced = 0;
1590         ext_sg_start = 0;
1591         ext_offset = 0;
1592
1593         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1594
1595         /*
1596          * If this flag is set, fake the data transfer.
1597          */
1598         if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1599                 ctsio->ext_data_filled = ctsio->ext_data_len;
1600                 goto bailout;
1601         }
1602
1603         /*
1604          * To simplify things here, if we have a single buffer, stick it in
1605          * a S/G entry and just make it a single entry S/G list.
1606          */
1607         if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1608                 int len_seen;
1609
1610                 ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1611
1612                 ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1613                                                            M_WAITOK);
1614                 ext_sglist_malloced = 1;
1615                 if (copyin(ctsio->ext_data_ptr, ext_sglist,
1616                                    ext_sglen) != 0) {
1617                         ctl_set_internal_failure(ctsio,
1618                                                  /*sks_valid*/ 0,
1619                                                  /*retry_count*/ 0);
1620                         goto bailout;
1621                 }
1622                 ext_sg_entries = ctsio->ext_sg_entries;
1623                 len_seen = 0;
1624                 for (i = 0; i < ext_sg_entries; i++) {
1625                         if ((len_seen + ext_sglist[i].len) >=
1626                              ctsio->ext_data_filled) {
1627                                 ext_sg_start = i;
1628                                 ext_offset = ctsio->ext_data_filled - len_seen;
1629                                 break;
1630                         }
1631                         len_seen += ext_sglist[i].len;
1632                 }
1633         } else {
1634                 ext_sglist = &ext_entry;
1635                 ext_sglist->addr = ctsio->ext_data_ptr;
1636                 ext_sglist->len = ctsio->ext_data_len;
1637                 ext_sg_entries = 1;
1638                 ext_sg_start = 0;
1639                 ext_offset = ctsio->ext_data_filled;
1640         }
1641
1642         if (ctsio->kern_sg_entries > 0) {
1643                 kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1644                 kern_sg_entries = ctsio->kern_sg_entries;
1645         } else {
1646                 kern_sglist = &kern_entry;
1647                 kern_sglist->addr = ctsio->kern_data_ptr;
1648                 kern_sglist->len = ctsio->kern_data_len;
1649                 kern_sg_entries = 1;
1650         }
1651
1652
1653         kern_watermark = 0;
1654         ext_watermark = ext_offset;
1655         len_copied = 0;
1656         for (i = ext_sg_start, j = 0;
1657              i < ext_sg_entries && j < kern_sg_entries;) {
1658                 uint8_t *ext_ptr, *kern_ptr;
1659
1660                 len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1661                                       kern_sglist[j].len - kern_watermark);
1662
1663                 ext_ptr = (uint8_t *)ext_sglist[i].addr;
1664                 ext_ptr = ext_ptr + ext_watermark;
1665                 if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1666                         /*
1667                          * XXX KDM fix this!
1668                          */
1669                         panic("need to implement bus address support");
1670 #if 0
1671                         kern_ptr = bus_to_virt(kern_sglist[j].addr);
1672 #endif
1673                 } else
1674                         kern_ptr = (uint8_t *)kern_sglist[j].addr;
1675                 kern_ptr = kern_ptr + kern_watermark;
1676
1677                 kern_watermark += len_to_copy;
1678                 ext_watermark += len_to_copy;
1679
1680                 if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1681                      CTL_FLAG_DATA_IN) {
1682                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1683                                          "bytes to user\n", len_to_copy));
1684                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1685                                          "to %p\n", kern_ptr, ext_ptr));
1686                         if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1687                                 ctl_set_internal_failure(ctsio,
1688                                                          /*sks_valid*/ 0,
1689                                                          /*retry_count*/ 0);
1690                                 goto bailout;
1691                         }
1692                 } else {
1693                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1694                                          "bytes from user\n", len_to_copy));
1695                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1696                                          "to %p\n", ext_ptr, kern_ptr));
1697                         if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1698                                 ctl_set_internal_failure(ctsio,
1699                                                          /*sks_valid*/ 0,
1700                                                          /*retry_count*/0);
1701                                 goto bailout;
1702                         }
1703                 }
1704
1705                 len_copied += len_to_copy;
1706
1707                 if (ext_sglist[i].len == ext_watermark) {
1708                         i++;
1709                         ext_watermark = 0;
1710                 }
1711
1712                 if (kern_sglist[j].len == kern_watermark) {
1713                         j++;
1714                         kern_watermark = 0;
1715                 }
1716         }
1717
1718         ctsio->ext_data_filled += len_copied;
1719
1720         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1721                          "kern_sg_entries: %d\n", ext_sg_entries,
1722                          kern_sg_entries));
1723         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1724                          "kern_data_len = %d\n", ctsio->ext_data_len,
1725                          ctsio->kern_data_len));
1726
1727
1728         /* XXX KDM set residual?? */
1729 bailout:
1730
1731         if (ext_sglist_malloced != 0)
1732                 free(ext_sglist, M_CTL);
1733
1734         return (CTL_RETVAL_COMPLETE);
1735 }
1736
1737 /*
1738  * Serialize a command that went down the "wrong" side, and so was sent to
1739  * this controller for execution.  The logic is a little different than the
1740  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1741  * sent back to the other side, but in the success case, we execute the
1742  * command on this side (XFER mode) or tell the other side to execute it
1743  * (SER_ONLY mode).
1744  */
1745 static int
1746 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1747 {
1748         struct ctl_softc *ctl_softc;
1749         union ctl_ha_msg msg_info;
1750         struct ctl_lun *lun;
1751         int retval = 0;
1752         uint32_t targ_lun;
1753
1754         ctl_softc = control_softc;
1755
1756         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1757         lun = ctl_softc->ctl_luns[targ_lun];
1758         if (lun==NULL)
1759         {
1760                 /*
1761                  * Why isn't LUN defined? The other side wouldn't
1762                  * send a cmd if the LUN is undefined.
1763                  */
1764                 printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1765
1766                 /* "Logical unit not supported" */
1767                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1768                                    lun,
1769                                    /*sense_format*/SSD_TYPE_NONE,
1770                                    /*current_error*/ 1,
1771                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1772                                    /*asc*/ 0x25,
1773                                    /*ascq*/ 0x00,
1774                                    SSD_ELEM_NONE);
1775
1776                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1777                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1778                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1779                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1780                 msg_info.hdr.serializing_sc = NULL;
1781                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1782                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1783                                 sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1784                 }
1785                 return(1);
1786
1787         }
1788
1789         mtx_lock(&lun->lun_lock);
1790         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1791
1792         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1793                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1794                  ooa_links))) {
1795         case CTL_ACTION_BLOCK:
1796                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1797                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1798                                   blocked_links);
1799                 break;
1800         case CTL_ACTION_PASS:
1801         case CTL_ACTION_SKIP:
1802                 if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1803                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1804                         ctl_enqueue_rtr((union ctl_io *)ctsio);
1805                 } else {
1806
1807                         /* send msg back to other side */
1808                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1809                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1810                         msg_info.hdr.msg_type = CTL_MSG_R2R;
1811 #if 0
1812                         printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1813 #endif
1814                         if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1815                             sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1816                         }
1817                 }
1818                 break;
1819         case CTL_ACTION_OVERLAP:
1820                 /* OVERLAPPED COMMANDS ATTEMPTED */
1821                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1822                                    lun,
1823                                    /*sense_format*/SSD_TYPE_NONE,
1824                                    /*current_error*/ 1,
1825                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1826                                    /*asc*/ 0x4E,
1827                                    /*ascq*/ 0x00,
1828                                    SSD_ELEM_NONE);
1829
1830                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1831                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1832                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1833                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1834                 msg_info.hdr.serializing_sc = NULL;
1835                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1836 #if 0
1837                 printf("BAD JUJU:Major Bummer Overlap\n");
1838 #endif
1839                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1840                 retval = 1;
1841                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1842                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1843                 }
1844                 break;
1845         case CTL_ACTION_OVERLAP_TAG:
1846                 /* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1847                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1848                                    lun,
1849                                    /*sense_format*/SSD_TYPE_NONE,
1850                                    /*current_error*/ 1,
1851                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1852                                    /*asc*/ 0x4D,
1853                                    /*ascq*/ ctsio->tag_num & 0xff,
1854                                    SSD_ELEM_NONE);
1855
1856                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1857                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1858                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1859                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1860                 msg_info.hdr.serializing_sc = NULL;
1861                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1862 #if 0
1863                 printf("BAD JUJU:Major Bummer Overlap Tag\n");
1864 #endif
1865                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1866                 retval = 1;
1867                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1868                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1869                 }
1870                 break;
1871         case CTL_ACTION_ERROR:
1872         default:
1873                 /* "Internal target failure" */
1874                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1875                                    lun,
1876                                    /*sense_format*/SSD_TYPE_NONE,
1877                                    /*current_error*/ 1,
1878                                    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1879                                    /*asc*/ 0x44,
1880                                    /*ascq*/ 0x00,
1881                                    SSD_ELEM_NONE);
1882
1883                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1884                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1885                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1886                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1887                 msg_info.hdr.serializing_sc = NULL;
1888                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1889 #if 0
1890                 printf("BAD JUJU:Major Bummer HW Error\n");
1891 #endif
1892                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1893                 retval = 1;
1894                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1895                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1896                 }
1897                 break;
1898         }
1899         mtx_unlock(&lun->lun_lock);
1900         return (retval);
1901 }
1902
1903 static int
1904 ctl_ioctl_submit_wait(union ctl_io *io)
1905 {
1906         struct ctl_fe_ioctl_params params;
1907         ctl_fe_ioctl_state last_state;
1908         int done, retval;
1909
1910         retval = 0;
1911
1912         bzero(&params, sizeof(params));
1913
1914         mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1915         cv_init(&params.sem, "ctlioccv");
1916         params.state = CTL_IOCTL_INPROG;
1917         last_state = params.state;
1918
1919         io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1920
1921         CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1922
1923         /* This shouldn't happen */
1924         if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1925                 return (retval);
1926
1927         done = 0;
1928
1929         do {
1930                 mtx_lock(&params.ioctl_mtx);
1931                 /*
1932                  * Check the state here, and don't sleep if the state has
1933                  * already changed (i.e. wakeup has already occured, but we
1934                  * weren't waiting yet).
1935                  */
1936                 if (params.state == last_state) {
1937                         /* XXX KDM cv_wait_sig instead? */
1938                         cv_wait(&params.sem, &params.ioctl_mtx);
1939                 }
1940                 last_state = params.state;
1941
1942                 switch (params.state) {
1943                 case CTL_IOCTL_INPROG:
1944                         /* Why did we wake up? */
1945                         /* XXX KDM error here? */
1946                         mtx_unlock(&params.ioctl_mtx);
1947                         break;
1948                 case CTL_IOCTL_DATAMOVE:
1949                         CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1950
1951                         /*
1952                          * change last_state back to INPROG to avoid
1953                          * deadlock on subsequent data moves.
1954                          */
1955                         params.state = last_state = CTL_IOCTL_INPROG;
1956
1957                         mtx_unlock(&params.ioctl_mtx);
1958                         ctl_ioctl_do_datamove(&io->scsiio);
1959                         /*
1960                          * Note that in some cases, most notably writes,
1961                          * this will queue the I/O and call us back later.
1962                          * In other cases, generally reads, this routine
1963                          * will immediately call back and wake us up,
1964                          * probably using our own context.
1965                          */
1966                         io->scsiio.be_move_done(io);
1967                         break;
1968                 case CTL_IOCTL_DONE:
1969                         mtx_unlock(&params.ioctl_mtx);
1970                         CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1971                         done = 1;
1972                         break;
1973                 default:
1974                         mtx_unlock(&params.ioctl_mtx);
1975                         /* XXX KDM error here? */
1976                         break;
1977                 }
1978         } while (done == 0);
1979
1980         mtx_destroy(&params.ioctl_mtx);
1981         cv_destroy(&params.sem);
1982
1983         return (CTL_RETVAL_COMPLETE);
1984 }
1985
1986 static void
1987 ctl_ioctl_datamove(union ctl_io *io)
1988 {
1989         struct ctl_fe_ioctl_params *params;
1990
1991         params = (struct ctl_fe_ioctl_params *)
1992                 io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1993
1994         mtx_lock(&params->ioctl_mtx);
1995         params->state = CTL_IOCTL_DATAMOVE;
1996         cv_broadcast(&params->sem);
1997         mtx_unlock(&params->ioctl_mtx);
1998 }
1999
2000 static void
2001 ctl_ioctl_done(union ctl_io *io)
2002 {
2003         struct ctl_fe_ioctl_params *params;
2004
2005         params = (struct ctl_fe_ioctl_params *)
2006                 io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2007
2008         mtx_lock(&params->ioctl_mtx);
2009         params->state = CTL_IOCTL_DONE;
2010         cv_broadcast(&params->sem);
2011         mtx_unlock(&params->ioctl_mtx);
2012 }
2013
2014 static void
2015 ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2016 {
2017         struct ctl_fe_ioctl_startstop_info *sd_info;
2018
2019         sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2020
2021         sd_info->hs_info.status = metatask->status;
2022         sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2023         sd_info->hs_info.luns_complete =
2024                 metatask->taskinfo.startstop.luns_complete;
2025         sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2026
2027         cv_broadcast(&sd_info->sem);
2028 }
2029
2030 static void
2031 ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2032 {
2033         struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2034
2035         fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2036
2037         mtx_lock(fe_bbr_info->lock);
2038         fe_bbr_info->bbr_info->status = metatask->status;
2039         fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2040         fe_bbr_info->wakeup_done = 1;
2041         mtx_unlock(fe_bbr_info->lock);
2042
2043         cv_broadcast(&fe_bbr_info->sem);
2044 }
2045
2046 /*
2047  * Returns 0 for success, errno for failure.
2048  */
2049 static int
2050 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2051                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2052 {
2053         union ctl_io *io;
2054         int retval;
2055
2056         retval = 0;
2057
2058         mtx_lock(&lun->lun_lock);
2059         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2060              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2061              ooa_links)) {
2062                 struct ctl_ooa_entry *entry;
2063
2064                 /*
2065                  * If we've got more than we can fit, just count the
2066                  * remaining entries.
2067                  */
2068                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2069                         continue;
2070
2071                 entry = &kern_entries[*cur_fill_num];
2072
2073                 entry->tag_num = io->scsiio.tag_num;
2074                 entry->lun_num = lun->lun;
2075 #ifdef CTL_TIME_IO
2076                 entry->start_bt = io->io_hdr.start_bt;
2077 #endif
2078                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2079                 entry->cdb_len = io->scsiio.cdb_len;
2080                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2081                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2082
2083                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2084                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2085
2086                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2087                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2088
2089                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2090                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2091
2092                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2093                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2094         }
2095         mtx_unlock(&lun->lun_lock);
2096
2097         return (retval);
2098 }
2099
2100 static void *
2101 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2102                  size_t error_str_len)
2103 {
2104         void *kptr;
2105
2106         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2107
2108         if (copyin(user_addr, kptr, len) != 0) {
2109                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2110                          "from user address %p to kernel address %p", len,
2111                          user_addr, kptr);
2112                 free(kptr, M_CTL);
2113                 return (NULL);
2114         }
2115
2116         return (kptr);
2117 }
2118
2119 static void
2120 ctl_free_args(int num_args, struct ctl_be_arg *args)
2121 {
2122         int i;
2123
2124         if (args == NULL)
2125                 return;
2126
2127         for (i = 0; i < num_args; i++) {
2128                 free(args[i].kname, M_CTL);
2129                 free(args[i].kvalue, M_CTL);
2130         }
2131
2132         free(args, M_CTL);
2133 }
2134
2135 static struct ctl_be_arg *
2136 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2137                 char *error_str, size_t error_str_len)
2138 {
2139         struct ctl_be_arg *args;
2140         int i;
2141
2142         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2143                                 error_str, error_str_len);
2144
2145         if (args == NULL)
2146                 goto bailout;
2147
2148         for (i = 0; i < num_args; i++) {
2149                 args[i].kname = NULL;
2150                 args[i].kvalue = NULL;
2151         }
2152
2153         for (i = 0; i < num_args; i++) {
2154                 uint8_t *tmpptr;
2155
2156                 args[i].kname = ctl_copyin_alloc(args[i].name,
2157                         args[i].namelen, error_str, error_str_len);
2158                 if (args[i].kname == NULL)
2159                         goto bailout;
2160
2161                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2162                         snprintf(error_str, error_str_len, "Argument %d "
2163                                  "name is not NUL-terminated", i);
2164                         goto bailout;
2165                 }
2166
2167                 if (args[i].flags & CTL_BEARG_RD) {
2168                         tmpptr = ctl_copyin_alloc(args[i].value,
2169                                 args[i].vallen, error_str, error_str_len);
2170                         if (tmpptr == NULL)
2171                                 goto bailout;
2172                         if ((args[i].flags & CTL_BEARG_ASCII)
2173                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2174                                 snprintf(error_str, error_str_len, "Argument "
2175                                     "%d value is not NUL-terminated", i);
2176                                 goto bailout;
2177                         }
2178                         args[i].kvalue = tmpptr;
2179                 } else {
2180                         args[i].kvalue = malloc(args[i].vallen,
2181                             M_CTL, M_WAITOK | M_ZERO);
2182                 }
2183         }
2184
2185         return (args);
2186 bailout:
2187
2188         ctl_free_args(num_args, args);
2189
2190         return (NULL);
2191 }
2192
2193 static void
2194 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2195 {
2196         int i;
2197
2198         for (i = 0; i < num_args; i++) {
2199                 if (args[i].flags & CTL_BEARG_WR)
2200                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2201         }
2202 }
2203
2204 /*
2205  * Escape characters that are illegal or not recommended in XML.
2206  */
2207 int
2208 ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2209 {
2210         int retval;
2211
2212         retval = 0;
2213
2214         for (; *str; str++) {
2215                 switch (*str) {
2216                 case '&':
2217                         retval = sbuf_printf(sb, "&amp;");
2218                         break;
2219                 case '>':
2220                         retval = sbuf_printf(sb, "&gt;");
2221                         break;
2222                 case '<':
2223                         retval = sbuf_printf(sb, "&lt;");
2224                         break;
2225                 default:
2226                         retval = sbuf_putc(sb, *str);
2227                         break;
2228                 }
2229
2230                 if (retval != 0)
2231                         break;
2232
2233         }
2234
2235         return (retval);
2236 }
2237
2238 static int
2239 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2240           struct thread *td)
2241 {
2242         struct ctl_softc *softc;
2243         int retval;
2244
2245         softc = control_softc;
2246
2247         retval = 0;
2248
2249         switch (cmd) {
2250         case CTL_IO: {
2251                 union ctl_io *io;
2252                 void *pool_tmp;
2253
2254                 /*
2255                  * If we haven't been "enabled", don't allow any SCSI I/O
2256                  * to this FETD.
2257                  */
2258                 if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2259                         retval = EPERM;
2260                         break;
2261                 }
2262
2263                 io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2264                 if (io == NULL) {
2265                         printf("ctl_ioctl: can't allocate ctl_io!\n");
2266                         retval = ENOSPC;
2267                         break;
2268                 }
2269
2270                 /*
2271                  * Need to save the pool reference so it doesn't get
2272                  * spammed by the user's ctl_io.
2273                  */
2274                 pool_tmp = io->io_hdr.pool;
2275
2276                 memcpy(io, (void *)addr, sizeof(*io));
2277
2278                 io->io_hdr.pool = pool_tmp;
2279                 /*
2280                  * No status yet, so make sure the status is set properly.
2281                  */
2282                 io->io_hdr.status = CTL_STATUS_NONE;
2283
2284                 /*
2285                  * The user sets the initiator ID, target and LUN IDs.
2286                  */
2287                 io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2288                 io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2289                 if ((io->io_hdr.io_type == CTL_IO_SCSI)
2290                  && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2291                         io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2292
2293                 retval = ctl_ioctl_submit_wait(io);
2294
2295                 if (retval != 0) {
2296                         ctl_free_io(io);
2297                         break;
2298                 }
2299
2300                 memcpy((void *)addr, io, sizeof(*io));
2301
2302                 /* return this to our pool */
2303                 ctl_free_io(io);
2304
2305                 break;
2306         }
2307         case CTL_ENABLE_PORT:
2308         case CTL_DISABLE_PORT:
2309         case CTL_SET_PORT_WWNS: {
2310                 struct ctl_port *port;
2311                 struct ctl_port_entry *entry;
2312
2313                 entry = (struct ctl_port_entry *)addr;
2314
2315                 mtx_lock(&softc->ctl_lock);
2316                 STAILQ_FOREACH(port, &softc->port_list, links) {
2317                         int action, done;
2318
2319                         action = 0;
2320                         done = 0;
2321
2322                         if ((entry->port_type == CTL_PORT_NONE)
2323                          && (entry->targ_port == port->targ_port)) {
2324                                 /*
2325                                  * If the user only wants to enable or
2326                                  * disable or set WWNs on a specific port,
2327                                  * do the operation and we're done.
2328                                  */
2329                                 action = 1;
2330                                 done = 1;
2331                         } else if (entry->port_type & port->port_type) {
2332                                 /*
2333                                  * Compare the user's type mask with the
2334                                  * particular frontend type to see if we
2335                                  * have a match.
2336                                  */
2337                                 action = 1;
2338                                 done = 0;
2339
2340                                 /*
2341                                  * Make sure the user isn't trying to set
2342                                  * WWNs on multiple ports at the same time.
2343                                  */
2344                                 if (cmd == CTL_SET_PORT_WWNS) {
2345                                         printf("%s: Can't set WWNs on "
2346                                                "multiple ports\n", __func__);
2347                                         retval = EINVAL;
2348                                         break;
2349                                 }
2350                         }
2351                         if (action != 0) {
2352                                 /*
2353                                  * XXX KDM we have to drop the lock here,
2354                                  * because the online/offline operations
2355                                  * can potentially block.  We need to
2356                                  * reference count the frontends so they
2357                                  * can't go away,
2358                                  */
2359                                 mtx_unlock(&softc->ctl_lock);
2360
2361                                 if (cmd == CTL_ENABLE_PORT) {
2362                                         struct ctl_lun *lun;
2363
2364                                         STAILQ_FOREACH(lun, &softc->lun_list,
2365                                                        links) {
2366                                                 port->lun_enable(port->targ_lun_arg,
2367                                                     lun->target,
2368                                                     lun->lun);
2369                                         }
2370
2371                                         ctl_port_online(port);
2372                                 } else if (cmd == CTL_DISABLE_PORT) {
2373                                         struct ctl_lun *lun;
2374
2375                                         ctl_port_offline(port);
2376
2377                                         STAILQ_FOREACH(lun, &softc->lun_list,
2378                                                        links) {
2379                                                 port->lun_disable(
2380                                                     port->targ_lun_arg,
2381                                                     lun->target,
2382                                                     lun->lun);
2383                                         }
2384                                 }
2385
2386                                 mtx_lock(&softc->ctl_lock);
2387
2388                                 if (cmd == CTL_SET_PORT_WWNS)
2389                                         ctl_port_set_wwns(port,
2390                                             (entry->flags & CTL_PORT_WWNN_VALID) ?
2391                                             1 : 0, entry->wwnn,
2392                                             (entry->flags & CTL_PORT_WWPN_VALID) ?
2393                                             1 : 0, entry->wwpn);
2394                         }
2395                         if (done != 0)
2396                                 break;
2397                 }
2398                 mtx_unlock(&softc->ctl_lock);
2399                 break;
2400         }
2401         case CTL_GET_PORT_LIST: {
2402                 struct ctl_port *port;
2403                 struct ctl_port_list *list;
2404                 int i;
2405
2406                 list = (struct ctl_port_list *)addr;
2407
2408                 if (list->alloc_len != (list->alloc_num *
2409                     sizeof(struct ctl_port_entry))) {
2410                         printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2411                                "alloc_num %u * sizeof(struct ctl_port_entry) "
2412                                "%zu\n", __func__, list->alloc_len,
2413                                list->alloc_num, sizeof(struct ctl_port_entry));
2414                         retval = EINVAL;
2415                         break;
2416                 }
2417                 list->fill_len = 0;
2418                 list->fill_num = 0;
2419                 list->dropped_num = 0;
2420                 i = 0;
2421                 mtx_lock(&softc->ctl_lock);
2422                 STAILQ_FOREACH(port, &softc->port_list, links) {
2423                         struct ctl_port_entry entry, *list_entry;
2424
2425                         if (list->fill_num >= list->alloc_num) {
2426                                 list->dropped_num++;
2427                                 continue;
2428                         }
2429
2430                         entry.port_type = port->port_type;
2431                         strlcpy(entry.port_name, port->port_name,
2432                                 sizeof(entry.port_name));
2433                         entry.targ_port = port->targ_port;
2434                         entry.physical_port = port->physical_port;
2435                         entry.virtual_port = port->virtual_port;
2436                         entry.wwnn = port->wwnn;
2437                         entry.wwpn = port->wwpn;
2438                         if (port->status & CTL_PORT_STATUS_ONLINE)
2439                                 entry.online = 1;
2440                         else
2441                                 entry.online = 0;
2442
2443                         list_entry = &list->entries[i];
2444
2445                         retval = copyout(&entry, list_entry, sizeof(entry));
2446                         if (retval != 0) {
2447                                 printf("%s: CTL_GET_PORT_LIST: copyout "
2448                                        "returned %d\n", __func__, retval);
2449                                 break;
2450                         }
2451                         i++;
2452                         list->fill_num++;
2453                         list->fill_len += sizeof(entry);
2454                 }
2455                 mtx_unlock(&softc->ctl_lock);
2456
2457                 /*
2458                  * If this is non-zero, we had a copyout fault, so there's
2459                  * probably no point in attempting to set the status inside
2460                  * the structure.
2461                  */
2462                 if (retval != 0)
2463                         break;
2464
2465                 if (list->dropped_num > 0)
2466                         list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2467                 else
2468                         list->status = CTL_PORT_LIST_OK;
2469                 break;
2470         }
2471         case CTL_DUMP_OOA: {
2472                 struct ctl_lun *lun;
2473                 union ctl_io *io;
2474                 char printbuf[128];
2475                 struct sbuf sb;
2476
2477                 mtx_lock(&softc->ctl_lock);
2478                 printf("Dumping OOA queues:\n");
2479                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2480                         mtx_lock(&lun->lun_lock);
2481                         for (io = (union ctl_io *)TAILQ_FIRST(
2482                              &lun->ooa_queue); io != NULL;
2483                              io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2484                              ooa_links)) {
2485                                 sbuf_new(&sb, printbuf, sizeof(printbuf),
2486                                          SBUF_FIXEDLEN);
2487                                 sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2488                                             (intmax_t)lun->lun,
2489                                             io->scsiio.tag_num,
2490                                             (io->io_hdr.flags &
2491                                             CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2492                                             (io->io_hdr.flags &
2493                                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2494                                             (io->io_hdr.flags &
2495                                             CTL_FLAG_ABORT) ? " ABORT" : "",
2496                                             (io->io_hdr.flags &
2497                                         CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2498                                 ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2499                                 sbuf_finish(&sb);
2500                                 printf("%s\n", sbuf_data(&sb));
2501                         }
2502                         mtx_unlock(&lun->lun_lock);
2503                 }
2504                 printf("OOA queues dump done\n");
2505                 mtx_unlock(&softc->ctl_lock);
2506                 break;
2507         }
2508         case CTL_GET_OOA: {
2509                 struct ctl_lun *lun;
2510                 struct ctl_ooa *ooa_hdr;
2511                 struct ctl_ooa_entry *entries;
2512                 uint32_t cur_fill_num;
2513
2514                 ooa_hdr = (struct ctl_ooa *)addr;
2515
2516                 if ((ooa_hdr->alloc_len == 0)
2517                  || (ooa_hdr->alloc_num == 0)) {
2518                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2519                                "must be non-zero\n", __func__,
2520                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2521                         retval = EINVAL;
2522                         break;
2523                 }
2524
2525                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2526                     sizeof(struct ctl_ooa_entry))) {
2527                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2528                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2529                                __func__, ooa_hdr->alloc_len,
2530                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2531                         retval = EINVAL;
2532                         break;
2533                 }
2534
2535                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2536                 if (entries == NULL) {
2537                         printf("%s: could not allocate %d bytes for OOA "
2538                                "dump\n", __func__, ooa_hdr->alloc_len);
2539                         retval = ENOMEM;
2540                         break;
2541                 }
2542
2543                 mtx_lock(&softc->ctl_lock);
2544                 if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2545                  && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2546                   || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2547                         mtx_unlock(&softc->ctl_lock);
2548                         free(entries, M_CTL);
2549                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2550                                __func__, (uintmax_t)ooa_hdr->lun_num);
2551                         retval = EINVAL;
2552                         break;
2553                 }
2554
2555                 cur_fill_num = 0;
2556
2557                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2558                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2559                                 retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2560                                         ooa_hdr, entries);
2561                                 if (retval != 0)
2562                                         break;
2563                         }
2564                         if (retval != 0) {
2565                                 mtx_unlock(&softc->ctl_lock);
2566                                 free(entries, M_CTL);
2567                                 break;
2568                         }
2569                 } else {
2570                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2571
2572                         retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2573                                                     entries);
2574                 }
2575                 mtx_unlock(&softc->ctl_lock);
2576
2577                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2578                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2579                         sizeof(struct ctl_ooa_entry);
2580                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2581                 if (retval != 0) {
2582                         printf("%s: error copying out %d bytes for OOA dump\n",
2583                                __func__, ooa_hdr->fill_len);
2584                 }
2585
2586                 getbintime(&ooa_hdr->cur_bt);
2587
2588                 if (cur_fill_num > ooa_hdr->alloc_num) {
2589                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2590                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2591                 } else {
2592                         ooa_hdr->dropped_num = 0;
2593                         ooa_hdr->status = CTL_OOA_OK;
2594                 }
2595
2596                 free(entries, M_CTL);
2597                 break;
2598         }
2599         case CTL_CHECK_OOA: {
2600                 union ctl_io *io;
2601                 struct ctl_lun *lun;
2602                 struct ctl_ooa_info *ooa_info;
2603
2604
2605                 ooa_info = (struct ctl_ooa_info *)addr;
2606
2607                 if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2608                         ooa_info->status = CTL_OOA_INVALID_LUN;
2609                         break;
2610                 }
2611                 mtx_lock(&softc->ctl_lock);
2612                 lun = softc->ctl_luns[ooa_info->lun_id];
2613                 if (lun == NULL) {
2614                         mtx_unlock(&softc->ctl_lock);
2615                         ooa_info->status = CTL_OOA_INVALID_LUN;
2616                         break;
2617                 }
2618                 mtx_lock(&lun->lun_lock);
2619                 mtx_unlock(&softc->ctl_lock);
2620                 ooa_info->num_entries = 0;
2621                 for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2622                      io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2623                      &io->io_hdr, ooa_links)) {
2624                         ooa_info->num_entries++;
2625                 }
2626                 mtx_unlock(&lun->lun_lock);
2627
2628                 ooa_info->status = CTL_OOA_SUCCESS;
2629
2630                 break;
2631         }
2632         case CTL_HARD_START:
2633         case CTL_HARD_STOP: {
2634                 struct ctl_fe_ioctl_startstop_info ss_info;
2635                 struct cfi_metatask *metatask;
2636                 struct mtx hs_mtx;
2637
2638                 mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2639
2640                 cv_init(&ss_info.sem, "hard start/stop cv" );
2641
2642                 metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2643                 if (metatask == NULL) {
2644                         retval = ENOMEM;
2645                         mtx_destroy(&hs_mtx);
2646                         break;
2647                 }
2648
2649                 if (cmd == CTL_HARD_START)
2650                         metatask->tasktype = CFI_TASK_STARTUP;
2651                 else
2652                         metatask->tasktype = CFI_TASK_SHUTDOWN;
2653
2654                 metatask->callback = ctl_ioctl_hard_startstop_callback;
2655                 metatask->callback_arg = &ss_info;
2656
2657                 cfi_action(metatask);
2658
2659                 /* Wait for the callback */
2660                 mtx_lock(&hs_mtx);
2661                 cv_wait_sig(&ss_info.sem, &hs_mtx);
2662                 mtx_unlock(&hs_mtx);
2663
2664                 /*
2665                  * All information has been copied from the metatask by the
2666                  * time cv_broadcast() is called, so we free the metatask here.
2667                  */
2668                 cfi_free_metatask(metatask);
2669
2670                 memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2671
2672                 mtx_destroy(&hs_mtx);
2673                 break;
2674         }
2675         case CTL_BBRREAD: {
2676                 struct ctl_bbrread_info *bbr_info;
2677                 struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2678                 struct mtx bbr_mtx;
2679                 struct cfi_metatask *metatask;
2680
2681                 bbr_info = (struct ctl_bbrread_info *)addr;
2682
2683                 bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2684
2685                 bzero(&bbr_mtx, sizeof(bbr_mtx));
2686                 mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2687
2688                 fe_bbr_info.bbr_info = bbr_info;
2689                 fe_bbr_info.lock = &bbr_mtx;
2690
2691                 cv_init(&fe_bbr_info.sem, "BBR read cv");
2692                 metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2693
2694                 if (metatask == NULL) {
2695                         mtx_destroy(&bbr_mtx);
2696                         cv_destroy(&fe_bbr_info.sem);
2697                         retval = ENOMEM;
2698                         break;
2699                 }
2700                 metatask->tasktype = CFI_TASK_BBRREAD;
2701                 metatask->callback = ctl_ioctl_bbrread_callback;
2702                 metatask->callback_arg = &fe_bbr_info;
2703                 metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2704                 metatask->taskinfo.bbrread.lba = bbr_info->lba;
2705                 metatask->taskinfo.bbrread.len = bbr_info->len;
2706
2707                 cfi_action(metatask);
2708
2709                 mtx_lock(&bbr_mtx);
2710                 while (fe_bbr_info.wakeup_done == 0)
2711                         cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2712                 mtx_unlock(&bbr_mtx);
2713
2714                 bbr_info->status = metatask->status;
2715                 bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2716                 bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2717                 memcpy(&bbr_info->sense_data,
2718                        &metatask->taskinfo.bbrread.sense_data,
2719                        ctl_min(sizeof(bbr_info->sense_data),
2720                                sizeof(metatask->taskinfo.bbrread.sense_data)));
2721
2722                 cfi_free_metatask(metatask);
2723
2724                 mtx_destroy(&bbr_mtx);
2725                 cv_destroy(&fe_bbr_info.sem);
2726
2727                 break;
2728         }
2729         case CTL_DELAY_IO: {
2730                 struct ctl_io_delay_info *delay_info;
2731 #ifdef CTL_IO_DELAY
2732                 struct ctl_lun *lun;
2733 #endif /* CTL_IO_DELAY */
2734
2735                 delay_info = (struct ctl_io_delay_info *)addr;
2736
2737 #ifdef CTL_IO_DELAY
2738                 mtx_lock(&softc->ctl_lock);
2739
2740                 if ((delay_info->lun_id > CTL_MAX_LUNS)
2741                  || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2742                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2743                 } else {
2744                         lun = softc->ctl_luns[delay_info->lun_id];
2745                         mtx_lock(&lun->lun_lock);
2746
2747                         delay_info->status = CTL_DELAY_STATUS_OK;
2748
2749                         switch (delay_info->delay_type) {
2750                         case CTL_DELAY_TYPE_CONT:
2751                                 break;
2752                         case CTL_DELAY_TYPE_ONESHOT:
2753                                 break;
2754                         default:
2755                                 delay_info->status =
2756                                         CTL_DELAY_STATUS_INVALID_TYPE;
2757                                 break;
2758                         }
2759
2760                         switch (delay_info->delay_loc) {
2761                         case CTL_DELAY_LOC_DATAMOVE:
2762                                 lun->delay_info.datamove_type =
2763                                         delay_info->delay_type;
2764                                 lun->delay_info.datamove_delay =
2765                                         delay_info->delay_secs;
2766                                 break;
2767                         case CTL_DELAY_LOC_DONE:
2768                                 lun->delay_info.done_type =
2769                                         delay_info->delay_type;
2770                                 lun->delay_info.done_delay =
2771                                         delay_info->delay_secs;
2772                                 break;
2773                         default:
2774                                 delay_info->status =
2775                                         CTL_DELAY_STATUS_INVALID_LOC;
2776                                 break;
2777                         }
2778                         mtx_unlock(&lun->lun_lock);
2779                 }
2780
2781                 mtx_unlock(&softc->ctl_lock);
2782 #else
2783                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2784 #endif /* CTL_IO_DELAY */
2785                 break;
2786         }
2787         case CTL_REALSYNC_SET: {
2788                 int *syncstate;
2789
2790                 syncstate = (int *)addr;
2791
2792                 mtx_lock(&softc->ctl_lock);
2793                 switch (*syncstate) {
2794                 case 0:
2795                         softc->flags &= ~CTL_FLAG_REAL_SYNC;
2796                         break;
2797                 case 1:
2798                         softc->flags |= CTL_FLAG_REAL_SYNC;
2799                         break;
2800                 default:
2801                         retval = EINVAL;
2802                         break;
2803                 }
2804                 mtx_unlock(&softc->ctl_lock);
2805                 break;
2806         }
2807         case CTL_REALSYNC_GET: {
2808                 int *syncstate;
2809
2810                 syncstate = (int*)addr;
2811
2812                 mtx_lock(&softc->ctl_lock);
2813                 if (softc->flags & CTL_FLAG_REAL_SYNC)
2814                         *syncstate = 1;
2815                 else
2816                         *syncstate = 0;
2817                 mtx_unlock(&softc->ctl_lock);
2818
2819                 break;
2820         }
2821         case CTL_SETSYNC:
2822         case CTL_GETSYNC: {
2823                 struct ctl_sync_info *sync_info;
2824                 struct ctl_lun *lun;
2825
2826                 sync_info = (struct ctl_sync_info *)addr;
2827
2828                 mtx_lock(&softc->ctl_lock);
2829                 lun = softc->ctl_luns[sync_info->lun_id];
2830                 if (lun == NULL) {
2831                         mtx_unlock(&softc->ctl_lock);
2832                         sync_info->status = CTL_GS_SYNC_NO_LUN;
2833                 }
2834                 /*
2835                  * Get or set the sync interval.  We're not bounds checking
2836                  * in the set case, hopefully the user won't do something
2837                  * silly.
2838                  */
2839                 mtx_lock(&lun->lun_lock);
2840                 mtx_unlock(&softc->ctl_lock);
2841                 if (cmd == CTL_GETSYNC)
2842                         sync_info->sync_interval = lun->sync_interval;
2843                 else
2844                         lun->sync_interval = sync_info->sync_interval;
2845                 mtx_unlock(&lun->lun_lock);
2846
2847                 sync_info->status = CTL_GS_SYNC_OK;
2848
2849                 break;
2850         }
2851         case CTL_GETSTATS: {
2852                 struct ctl_stats *stats;
2853                 struct ctl_lun *lun;
2854                 int i;
2855
2856                 stats = (struct ctl_stats *)addr;
2857
2858                 if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2859                      stats->alloc_len) {
2860                         stats->status = CTL_SS_NEED_MORE_SPACE;
2861                         stats->num_luns = softc->num_luns;
2862                         break;
2863                 }
2864                 /*
2865                  * XXX KDM no locking here.  If the LUN list changes,
2866                  * things can blow up.
2867                  */
2868                 for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2869                      i++, lun = STAILQ_NEXT(lun, links)) {
2870                         retval = copyout(&lun->stats, &stats->lun_stats[i],
2871                                          sizeof(lun->stats));
2872                         if (retval != 0)
2873                                 break;
2874                 }
2875                 stats->num_luns = softc->num_luns;
2876                 stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2877                                  softc->num_luns;
2878                 stats->status = CTL_SS_OK;
2879 #ifdef CTL_TIME_IO
2880                 stats->flags = CTL_STATS_FLAG_TIME_VALID;
2881 #else
2882                 stats->flags = CTL_STATS_FLAG_NONE;
2883 #endif
2884                 getnanouptime(&stats->timestamp);
2885                 break;
2886         }
2887         case CTL_ERROR_INJECT: {
2888                 struct ctl_error_desc *err_desc, *new_err_desc;
2889                 struct ctl_lun *lun;
2890
2891                 err_desc = (struct ctl_error_desc *)addr;
2892
2893                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2894                                       M_WAITOK | M_ZERO);
2895                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2896
2897                 mtx_lock(&softc->ctl_lock);
2898                 lun = softc->ctl_luns[err_desc->lun_id];
2899                 if (lun == NULL) {
2900                         mtx_unlock(&softc->ctl_lock);
2901                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2902                                __func__, (uintmax_t)err_desc->lun_id);
2903                         retval = EINVAL;
2904                         break;
2905                 }
2906                 mtx_lock(&lun->lun_lock);
2907                 mtx_unlock(&softc->ctl_lock);
2908
2909                 /*
2910                  * We could do some checking here to verify the validity
2911                  * of the request, but given the complexity of error
2912                  * injection requests, the checking logic would be fairly
2913                  * complex.
2914                  *
2915                  * For now, if the request is invalid, it just won't get
2916                  * executed and might get deleted.
2917                  */
2918                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2919
2920                 /*
2921                  * XXX KDM check to make sure the serial number is unique,
2922                  * in case we somehow manage to wrap.  That shouldn't
2923                  * happen for a very long time, but it's the right thing to
2924                  * do.
2925                  */
2926                 new_err_desc->serial = lun->error_serial;
2927                 err_desc->serial = lun->error_serial;
2928                 lun->error_serial++;
2929
2930                 mtx_unlock(&lun->lun_lock);
2931                 break;
2932         }
2933         case CTL_ERROR_INJECT_DELETE: {
2934                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2935                 struct ctl_lun *lun;
2936                 int delete_done;
2937
2938                 delete_desc = (struct ctl_error_desc *)addr;
2939                 delete_done = 0;
2940
2941                 mtx_lock(&softc->ctl_lock);
2942                 lun = softc->ctl_luns[delete_desc->lun_id];
2943                 if (lun == NULL) {
2944                         mtx_unlock(&softc->ctl_lock);
2945                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2946                                __func__, (uintmax_t)delete_desc->lun_id);
2947                         retval = EINVAL;
2948                         break;
2949                 }
2950                 mtx_lock(&lun->lun_lock);
2951                 mtx_unlock(&softc->ctl_lock);
2952                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2953                         if (desc->serial != delete_desc->serial)
2954                                 continue;
2955
2956                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2957                                       links);
2958                         free(desc, M_CTL);
2959                         delete_done = 1;
2960                 }
2961                 mtx_unlock(&lun->lun_lock);
2962                 if (delete_done == 0) {
2963                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2964                                "error serial %ju on LUN %u\n", __func__,
2965                                delete_desc->serial, delete_desc->lun_id);
2966                         retval = EINVAL;
2967                         break;
2968                 }
2969                 break;
2970         }
2971         case CTL_DUMP_STRUCTS: {
2972                 int i, j, k, idx;
2973                 struct ctl_port *port;
2974                 struct ctl_frontend *fe;
2975
2976                 mtx_lock(&softc->ctl_lock);
2977                 printf("CTL Persistent Reservation information start:\n");
2978                 for (i = 0; i < CTL_MAX_LUNS; i++) {
2979                         struct ctl_lun *lun;
2980
2981                         lun = softc->ctl_luns[i];
2982
2983                         if ((lun == NULL)
2984                          || ((lun->flags & CTL_LUN_DISABLED) != 0))
2985                                 continue;
2986
2987                         for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2988                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2989                                         idx = j * CTL_MAX_INIT_PER_PORT + k;
2990                                         if (lun->per_res[idx].registered == 0)
2991                                                 continue;
2992                                         printf("  LUN %d port %d iid %d key "
2993                                                "%#jx\n", i, j, k,
2994                                                (uintmax_t)scsi_8btou64(
2995                                                lun->per_res[idx].res_key.key));
2996                                 }
2997                         }
2998                 }
2999                 printf("CTL Persistent Reservation information end\n");
3000                 printf("CTL Ports:\n");
3001                 STAILQ_FOREACH(port, &softc->port_list, links) {
3002                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3003                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3004                                port->frontend->name, port->port_type,
3005                                port->physical_port, port->virtual_port,
3006                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3007                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3008                                 if (port->wwpn_iid[j].in_use == 0 &&
3009                                     port->wwpn_iid[j].wwpn == 0 &&
3010                                     port->wwpn_iid[j].name == NULL)
3011                                         continue;
3012
3013                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
3014                                     j, port->wwpn_iid[j].in_use,
3015                                     (uintmax_t)port->wwpn_iid[j].wwpn,
3016                                     port->wwpn_iid[j].name);
3017                         }
3018                 }
3019                 printf("CTL Port information end\n");
3020                 mtx_unlock(&softc->ctl_lock);
3021                 /*
3022                  * XXX KDM calling this without a lock.  We'd likely want
3023                  * to drop the lock before calling the frontend's dump
3024                  * routine anyway.
3025                  */
3026                 printf("CTL Frontends:\n");
3027                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
3028                         printf("  Frontend '%s'\n", fe->name);
3029                         if (fe->fe_dump != NULL)
3030                                 fe->fe_dump();
3031                 }
3032                 printf("CTL Frontend information end\n");
3033                 break;
3034         }
3035         case CTL_LUN_REQ: {
3036                 struct ctl_lun_req *lun_req;
3037                 struct ctl_backend_driver *backend;
3038
3039                 lun_req = (struct ctl_lun_req *)addr;
3040
3041                 backend = ctl_backend_find(lun_req->backend);
3042                 if (backend == NULL) {
3043                         lun_req->status = CTL_LUN_ERROR;
3044                         snprintf(lun_req->error_str,
3045                                  sizeof(lun_req->error_str),
3046                                  "Backend \"%s\" not found.",
3047                                  lun_req->backend);
3048                         break;
3049                 }
3050                 if (lun_req->num_be_args > 0) {
3051                         lun_req->kern_be_args = ctl_copyin_args(
3052                                 lun_req->num_be_args,
3053                                 lun_req->be_args,
3054                                 lun_req->error_str,
3055                                 sizeof(lun_req->error_str));
3056                         if (lun_req->kern_be_args == NULL) {
3057                                 lun_req->status = CTL_LUN_ERROR;
3058                                 break;
3059                         }
3060                 }
3061
3062                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3063
3064                 if (lun_req->num_be_args > 0) {
3065                         ctl_copyout_args(lun_req->num_be_args,
3066                                       lun_req->kern_be_args);
3067                         ctl_free_args(lun_req->num_be_args,
3068                                       lun_req->kern_be_args);
3069                 }
3070                 break;
3071         }
3072         case CTL_LUN_LIST: {
3073                 struct sbuf *sb;
3074                 struct ctl_lun *lun;
3075                 struct ctl_lun_list *list;
3076                 struct ctl_option *opt;
3077
3078                 list = (struct ctl_lun_list *)addr;
3079
3080                 /*
3081                  * Allocate a fixed length sbuf here, based on the length
3082                  * of the user's buffer.  We could allocate an auto-extending
3083                  * buffer, and then tell the user how much larger our
3084                  * amount of data is than his buffer, but that presents
3085                  * some problems:
3086                  *
3087                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3088                  *     we can't hold a lock while calling them with an
3089                  *     auto-extending buffer.
3090                  *
3091                  * 2.  There is not currently a LUN reference counting
3092                  *     mechanism, outside of outstanding transactions on
3093                  *     the LUN's OOA queue.  So a LUN could go away on us
3094                  *     while we're getting the LUN number, backend-specific
3095                  *     information, etc.  Thus, given the way things
3096                  *     currently work, we need to hold the CTL lock while
3097                  *     grabbing LUN information.
3098                  *
3099                  * So, from the user's standpoint, the best thing to do is
3100                  * allocate what he thinks is a reasonable buffer length,
3101                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3102                  * double the buffer length and try again.  (And repeat
3103                  * that until he succeeds.)
3104                  */
3105                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3106                 if (sb == NULL) {
3107                         list->status = CTL_LUN_LIST_ERROR;
3108                         snprintf(list->error_str, sizeof(list->error_str),
3109                                  "Unable to allocate %d bytes for LUN list",
3110                                  list->alloc_len);
3111                         break;
3112                 }
3113
3114                 sbuf_printf(sb, "<ctllunlist>\n");
3115
3116                 mtx_lock(&softc->ctl_lock);
3117                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3118                         mtx_lock(&lun->lun_lock);
3119                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3120                                              (uintmax_t)lun->lun);
3121
3122                         /*
3123                          * Bail out as soon as we see that we've overfilled
3124                          * the buffer.
3125                          */
3126                         if (retval != 0)
3127                                 break;
3128
3129                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3130                                              "</backend_type>\n",
3131                                              (lun->backend == NULL) ?  "none" :
3132                                              lun->backend->name);
3133
3134                         if (retval != 0)
3135                                 break;
3136
3137                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3138                                              lun->be_lun->lun_type);
3139
3140                         if (retval != 0)
3141                                 break;
3142
3143                         if (lun->backend == NULL) {
3144                                 retval = sbuf_printf(sb, "</lun>\n");
3145                                 if (retval != 0)
3146                                         break;
3147                                 continue;
3148                         }
3149
3150                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3151                                              (lun->be_lun->maxlba > 0) ?
3152                                              lun->be_lun->maxlba + 1 : 0);
3153
3154                         if (retval != 0)
3155                                 break;
3156
3157                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3158                                              lun->be_lun->blocksize);
3159
3160                         if (retval != 0)
3161                                 break;
3162
3163                         retval = sbuf_printf(sb, "\t<serial_number>");
3164
3165                         if (retval != 0)
3166                                 break;
3167
3168                         retval = ctl_sbuf_printf_esc(sb,
3169                                                      lun->be_lun->serial_num);
3170
3171                         if (retval != 0)
3172                                 break;
3173
3174                         retval = sbuf_printf(sb, "</serial_number>\n");
3175
3176                         if (retval != 0)
3177                                 break;
3178
3179                         retval = sbuf_printf(sb, "\t<device_id>");
3180
3181                         if (retval != 0)
3182                                 break;
3183
3184                         retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3185
3186                         if (retval != 0)
3187                                 break;
3188
3189                         retval = sbuf_printf(sb, "</device_id>\n");
3190
3191                         if (retval != 0)
3192                                 break;
3193
3194                         if (lun->backend->lun_info != NULL) {
3195                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3196                                 if (retval != 0)
3197                                         break;
3198                         }
3199                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3200                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3201                                     opt->name, opt->value, opt->name);
3202                                 if (retval != 0)
3203                                         break;
3204                         }
3205
3206                         retval = sbuf_printf(sb, "</lun>\n");
3207
3208                         if (retval != 0)
3209                                 break;
3210                         mtx_unlock(&lun->lun_lock);
3211                 }
3212                 if (lun != NULL)
3213                         mtx_unlock(&lun->lun_lock);
3214                 mtx_unlock(&softc->ctl_lock);
3215
3216                 if ((retval != 0)
3217                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3218                         retval = 0;
3219                         sbuf_delete(sb);
3220                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3221                         snprintf(list->error_str, sizeof(list->error_str),
3222                                  "Out of space, %d bytes is too small",
3223                                  list->alloc_len);
3224                         break;
3225                 }
3226
3227                 sbuf_finish(sb);
3228
3229                 retval = copyout(sbuf_data(sb), list->lun_xml,
3230                                  sbuf_len(sb) + 1);
3231
3232                 list->fill_len = sbuf_len(sb) + 1;
3233                 list->status = CTL_LUN_LIST_OK;
3234                 sbuf_delete(sb);
3235                 break;
3236         }
3237         case CTL_ISCSI: {
3238                 struct ctl_iscsi *ci;
3239                 struct ctl_frontend *fe;
3240
3241                 ci = (struct ctl_iscsi *)addr;
3242
3243                 fe = ctl_frontend_find("iscsi");
3244                 if (fe == NULL) {
3245                         ci->status = CTL_ISCSI_ERROR;
3246                         snprintf(ci->error_str, sizeof(ci->error_str),
3247                             "Frontend \"iscsi\" not found.");
3248                         break;
3249                 }
3250
3251                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3252                 break;
3253         }
3254         case CTL_PORT_REQ: {
3255                 struct ctl_req *req;
3256                 struct ctl_frontend *fe;
3257
3258                 req = (struct ctl_req *)addr;
3259
3260                 fe = ctl_frontend_find(req->driver);
3261                 if (fe == NULL) {
3262                         req->status = CTL_LUN_ERROR;
3263                         snprintf(req->error_str, sizeof(req->error_str),
3264                             "Frontend \"%s\" not found.", req->driver);
3265                         break;
3266                 }
3267                 if (req->num_args > 0) {
3268                         req->kern_args = ctl_copyin_args(req->num_args,
3269                             req->args, req->error_str, sizeof(req->error_str));
3270                         if (req->kern_args == NULL) {
3271                                 req->status = CTL_LUN_ERROR;
3272                                 break;
3273                         }
3274                 }
3275
3276                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3277
3278                 if (req->num_args > 0) {
3279                         ctl_copyout_args(req->num_args, req->kern_args);
3280                         ctl_free_args(req->num_args, req->kern_args);
3281                 }
3282                 break;
3283         }
3284         case CTL_PORT_LIST: {
3285                 struct sbuf *sb;
3286                 struct ctl_port *port;
3287                 struct ctl_lun_list *list;
3288                 struct ctl_option *opt;
3289
3290                 list = (struct ctl_lun_list *)addr;
3291
3292                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3293                 if (sb == NULL) {
3294                         list->status = CTL_LUN_LIST_ERROR;
3295                         snprintf(list->error_str, sizeof(list->error_str),
3296                                  "Unable to allocate %d bytes for LUN list",
3297                                  list->alloc_len);
3298                         break;
3299                 }
3300
3301                 sbuf_printf(sb, "<ctlportlist>\n");
3302
3303                 mtx_lock(&softc->ctl_lock);
3304                 STAILQ_FOREACH(port, &softc->port_list, links) {
3305                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3306                                              (uintmax_t)port->targ_port);
3307
3308                         /*
3309                          * Bail out as soon as we see that we've overfilled
3310                          * the buffer.
3311                          */
3312                         if (retval != 0)
3313                                 break;
3314
3315                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3316                             "</frontend_type>\n", port->frontend->name);
3317                         if (retval != 0)
3318                                 break;
3319
3320                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3321                                              port->port_type);
3322                         if (retval != 0)
3323                                 break;
3324
3325                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3326                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3327                         if (retval != 0)
3328                                 break;
3329
3330                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3331                             port->port_name);
3332                         if (retval != 0)
3333                                 break;
3334
3335                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3336                             port->physical_port);
3337                         if (retval != 0)
3338                                 break;
3339
3340                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3341                             port->virtual_port);
3342                         if (retval != 0)
3343                                 break;
3344
3345                         retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3346                             (uintmax_t)port->wwnn);
3347                         if (retval != 0)
3348                                 break;
3349
3350                         retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3351                             (uintmax_t)port->wwpn);
3352                         if (retval != 0)
3353                                 break;
3354
3355                         if (port->port_info != NULL) {
3356                                 retval = port->port_info(port->onoff_arg, sb);
3357                                 if (retval != 0)
3358                                         break;
3359                         }
3360                         STAILQ_FOREACH(opt, &port->options, links) {
3361                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3362                                     opt->name, opt->value, opt->name);
3363                                 if (retval != 0)
3364                                         break;
3365                         }
3366
3367                         retval = sbuf_printf(sb, "</targ_port>\n");
3368                         if (retval != 0)
3369                                 break;
3370                 }
3371                 mtx_unlock(&softc->ctl_lock);
3372
3373                 if ((retval != 0)
3374                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3375                         retval = 0;
3376                         sbuf_delete(sb);
3377                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3378                         snprintf(list->error_str, sizeof(list->error_str),
3379                                  "Out of space, %d bytes is too small",
3380                                  list->alloc_len);
3381                         break;
3382                 }
3383
3384                 sbuf_finish(sb);
3385
3386                 retval = copyout(sbuf_data(sb), list->lun_xml,
3387                                  sbuf_len(sb) + 1);
3388
3389                 list->fill_len = sbuf_len(sb) + 1;
3390                 list->status = CTL_LUN_LIST_OK;
3391                 sbuf_delete(sb);
3392                 break;
3393         }
3394         default: {
3395                 /* XXX KDM should we fix this? */
3396 #if 0
3397                 struct ctl_backend_driver *backend;
3398                 unsigned int type;
3399                 int found;
3400
3401                 found = 0;
3402
3403                 /*
3404                  * We encode the backend type as the ioctl type for backend
3405                  * ioctls.  So parse it out here, and then search for a
3406                  * backend of this type.
3407                  */
3408                 type = _IOC_TYPE(cmd);
3409
3410                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3411                         if (backend->type == type) {
3412                                 found = 1;
3413                                 break;
3414                         }
3415                 }
3416                 if (found == 0) {
3417                         printf("ctl: unknown ioctl command %#lx or backend "
3418                                "%d\n", cmd, type);
3419                         retval = EINVAL;
3420                         break;
3421                 }
3422                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3423 #endif
3424                 retval = ENOTTY;
3425                 break;
3426         }
3427         }
3428         return (retval);
3429 }
3430
3431 uint32_t
3432 ctl_get_initindex(struct ctl_nexus *nexus)
3433 {
3434         if (nexus->targ_port < CTL_MAX_PORTS)
3435                 return (nexus->initid.id +
3436                         (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3437         else
3438                 return (nexus->initid.id +
3439                        ((nexus->targ_port - CTL_MAX_PORTS) *
3440                         CTL_MAX_INIT_PER_PORT));
3441 }
3442
3443 uint32_t
3444 ctl_get_resindex(struct ctl_nexus *nexus)
3445 {
3446         return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3447 }
3448
3449 uint32_t
3450 ctl_port_idx(int port_num)
3451 {
3452         if (port_num < CTL_MAX_PORTS)
3453                 return(port_num);
3454         else
3455                 return(port_num - CTL_MAX_PORTS);
3456 }
3457
3458 static uint32_t
3459 ctl_map_lun(int port_num, uint32_t lun_id)
3460 {
3461         struct ctl_port *port;
3462
3463         port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3464         if (port == NULL)
3465                 return (UINT32_MAX);
3466         if (port->lun_map == NULL)
3467                 return (lun_id);
3468         return (port->lun_map(port->targ_lun_arg, lun_id));
3469 }
3470
3471 static uint32_t
3472 ctl_map_lun_back(int port_num, uint32_t lun_id)
3473 {
3474         struct ctl_port *port;
3475         uint32_t i;
3476
3477         port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3478         if (port->lun_map == NULL)
3479                 return (lun_id);
3480         for (i = 0; i < CTL_MAX_LUNS; i++) {
3481                 if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3482                         return (i);
3483         }
3484         return (UINT32_MAX);
3485 }
3486
3487 /*
3488  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3489  * that are a power of 2.
3490  */
3491 int
3492 ctl_ffz(uint32_t *mask, uint32_t size)
3493 {
3494         uint32_t num_chunks, num_pieces;
3495         int i, j;
3496
3497         num_chunks = (size >> 5);
3498         if (num_chunks == 0)
3499                 num_chunks++;
3500         num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3501
3502         for (i = 0; i < num_chunks; i++) {
3503                 for (j = 0; j < num_pieces; j++) {
3504                         if ((mask[i] & (1 << j)) == 0)
3505                                 return ((i << 5) + j);
3506                 }
3507         }
3508
3509         return (-1);
3510 }
3511
3512 int
3513 ctl_set_mask(uint32_t *mask, uint32_t bit)
3514 {
3515         uint32_t chunk, piece;
3516
3517         chunk = bit >> 5;
3518         piece = bit % (sizeof(uint32_t) * 8);
3519
3520         if ((mask[chunk] & (1 << piece)) != 0)
3521                 return (-1);
3522         else
3523                 mask[chunk] |= (1 << piece);
3524
3525         return (0);
3526 }
3527
3528 int
3529 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3530 {
3531         uint32_t chunk, piece;
3532
3533         chunk = bit >> 5;
3534         piece = bit % (sizeof(uint32_t) * 8);
3535
3536         if ((mask[chunk] & (1 << piece)) == 0)
3537                 return (-1);
3538         else
3539                 mask[chunk] &= ~(1 << piece);
3540
3541         return (0);
3542 }
3543
3544 int
3545 ctl_is_set(uint32_t *mask, uint32_t bit)
3546 {
3547         uint32_t chunk, piece;
3548
3549         chunk = bit >> 5;
3550         piece = bit % (sizeof(uint32_t) * 8);
3551
3552         if ((mask[chunk] & (1 << piece)) == 0)
3553                 return (0);
3554         else
3555                 return (1);
3556 }
3557
3558 #ifdef unused
3559 /*
3560  * The bus, target and lun are optional, they can be filled in later.
3561  * can_wait is used to determine whether we can wait on the malloc or not.
3562  */
3563 union ctl_io*
3564 ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3565               uint32_t targ_lun, int can_wait)
3566 {
3567         union ctl_io *io;
3568
3569         if (can_wait)
3570                 io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3571         else
3572                 io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3573
3574         if (io != NULL) {
3575                 io->io_hdr.io_type = io_type;
3576                 io->io_hdr.targ_port = targ_port;
3577                 /*
3578                  * XXX KDM this needs to change/go away.  We need to move
3579                  * to a preallocated pool of ctl_scsiio structures.
3580                  */
3581                 io->io_hdr.nexus.targ_target.id = targ_target;
3582                 io->io_hdr.nexus.targ_lun = targ_lun;
3583         }
3584
3585         return (io);
3586 }
3587
3588 void
3589 ctl_kfree_io(union ctl_io *io)
3590 {
3591         free(io, M_CTL);
3592 }
3593 #endif /* unused */
3594
3595 /*
3596  * ctl_softc, pool_type, total_ctl_io are passed in.
3597  * npool is passed out.
3598  */
3599 int
3600 ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3601                 uint32_t total_ctl_io, struct ctl_io_pool **npool)
3602 {
3603         uint32_t i;
3604         union ctl_io *cur_io, *next_io;
3605         struct ctl_io_pool *pool;
3606         int retval;
3607
3608         retval = 0;
3609
3610         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3611                                             M_NOWAIT | M_ZERO);
3612         if (pool == NULL) {
3613                 retval = ENOMEM;
3614                 goto bailout;
3615         }
3616
3617         pool->type = pool_type;
3618         pool->ctl_softc = ctl_softc;
3619
3620         mtx_lock(&ctl_softc->pool_lock);
3621         pool->id = ctl_softc->cur_pool_id++;
3622         mtx_unlock(&ctl_softc->pool_lock);
3623
3624         pool->flags = CTL_POOL_FLAG_NONE;
3625         pool->refcount = 1;             /* Reference for validity. */
3626         STAILQ_INIT(&pool->free_queue);
3627
3628         /*
3629          * XXX KDM other options here:
3630          * - allocate a page at a time
3631          * - allocate one big chunk of memory.
3632          * Page allocation might work well, but would take a little more
3633          * tracking.
3634          */
3635         for (i = 0; i < total_ctl_io; i++) {
3636                 cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3637                                                 M_NOWAIT);
3638                 if (cur_io == NULL) {
3639                         retval = ENOMEM;
3640                         break;
3641                 }
3642                 cur_io->io_hdr.pool = pool;
3643                 STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3644                 pool->total_ctl_io++;
3645                 pool->free_ctl_io++;
3646         }
3647
3648         if (retval != 0) {
3649                 for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3650                      cur_io != NULL; cur_io = next_io) {
3651                         next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3652                                                               links);
3653                         STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3654                                       ctl_io_hdr, links);
3655                         free(cur_io, M_CTLIO);
3656                 }
3657
3658                 free(pool, M_CTL);
3659                 goto bailout;
3660         }
3661         mtx_lock(&ctl_softc->pool_lock);
3662         ctl_softc->num_pools++;
3663         STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3664         /*
3665          * Increment our usage count if this is an external consumer, so we
3666          * can't get unloaded until the external consumer (most likely a
3667          * FETD) unloads and frees his pool.
3668          *
3669          * XXX KDM will this increment the caller's module use count, or
3670          * mine?
3671          */
3672 #if 0
3673         if ((pool_type != CTL_POOL_EMERGENCY)
3674          && (pool_type != CTL_POOL_INTERNAL)
3675          && (pool_type != CTL_POOL_4OTHERSC))
3676                 MOD_INC_USE_COUNT;
3677 #endif
3678
3679         mtx_unlock(&ctl_softc->pool_lock);
3680
3681         *npool = pool;
3682
3683 bailout:
3684
3685         return (retval);
3686 }
3687
3688 static int
3689 ctl_pool_acquire(struct ctl_io_pool *pool)
3690 {
3691
3692         mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3693
3694         if (pool->flags & CTL_POOL_FLAG_INVALID)
3695                 return (EINVAL);
3696
3697         pool->refcount++;
3698
3699         return (0);
3700 }
3701
3702 static void
3703 ctl_pool_release(struct ctl_io_pool *pool)
3704 {
3705         struct ctl_softc *ctl_softc = pool->ctl_softc;
3706         union ctl_io *io;
3707
3708         mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3709
3710         if (--pool->refcount != 0)
3711                 return;
3712
3713         while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3714                 STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3715                               links);
3716                 free(io, M_CTLIO);
3717         }
3718
3719         STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3720         ctl_softc->num_pools--;
3721
3722         /*
3723          * XXX KDM will this decrement the caller's usage count or mine?
3724          */
3725 #if 0
3726         if ((pool->type != CTL_POOL_EMERGENCY)
3727          && (pool->type != CTL_POOL_INTERNAL)
3728          && (pool->type != CTL_POOL_4OTHERSC))
3729                 MOD_DEC_USE_COUNT;
3730 #endif
3731
3732         free(pool, M_CTL);
3733 }
3734
3735 void
3736 ctl_pool_free(struct ctl_io_pool *pool)
3737 {
3738         struct ctl_softc *ctl_softc;
3739
3740         if (pool == NULL)
3741                 return;
3742
3743         ctl_softc = pool->ctl_softc;
3744         mtx_lock(&ctl_softc->pool_lock);
3745         pool->flags |= CTL_POOL_FLAG_INVALID;
3746         ctl_pool_release(pool);
3747         mtx_unlock(&ctl_softc->pool_lock);
3748 }
3749
3750 /*
3751  * This routine does not block (except for spinlocks of course).
3752  * It tries to allocate a ctl_io union from the caller's pool as quickly as
3753  * possible.
3754  */
3755 union ctl_io *
3756 ctl_alloc_io(void *pool_ref)
3757 {
3758         union ctl_io *io;
3759         struct ctl_softc *ctl_softc;
3760         struct ctl_io_pool *pool, *npool;
3761         struct ctl_io_pool *emergency_pool;
3762
3763         pool = (struct ctl_io_pool *)pool_ref;
3764
3765         if (pool == NULL) {
3766                 printf("%s: pool is NULL\n", __func__);
3767                 return (NULL);
3768         }
3769
3770         emergency_pool = NULL;
3771
3772         ctl_softc = pool->ctl_softc;
3773
3774         mtx_lock(&ctl_softc->pool_lock);
3775         /*
3776          * First, try to get the io structure from the user's pool.
3777          */
3778         if (ctl_pool_acquire(pool) == 0) {
3779                 io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3780                 if (io != NULL) {
3781                         STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3782                         pool->total_allocated++;
3783                         pool->free_ctl_io--;
3784                         mtx_unlock(&ctl_softc->pool_lock);
3785                         return (io);
3786                 } else
3787                         ctl_pool_release(pool);
3788         }
3789         /*
3790          * If he doesn't have any io structures left, search for an
3791          * emergency pool and grab one from there.
3792          */
3793         STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3794                 if (npool->type != CTL_POOL_EMERGENCY)
3795                         continue;
3796
3797                 if (ctl_pool_acquire(npool) != 0)
3798                         continue;
3799
3800                 emergency_pool = npool;
3801
3802                 io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3803                 if (io != NULL) {
3804                         STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3805                         npool->total_allocated++;
3806                         npool->free_ctl_io--;
3807                         mtx_unlock(&ctl_softc->pool_lock);
3808                         return (io);
3809                 } else
3810                         ctl_pool_release(npool);
3811         }
3812
3813         /* Drop the spinlock before we malloc */
3814         mtx_unlock(&ctl_softc->pool_lock);
3815
3816         /*
3817          * The emergency pool (if it exists) didn't have one, so try an
3818          * atomic (i.e. nonblocking) malloc and see if we get lucky.
3819          */
3820         io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3821         if (io != NULL) {
3822                 /*
3823                  * If the emergency pool exists but is empty, add this
3824                  * ctl_io to its list when it gets freed.
3825                  */
3826                 if (emergency_pool != NULL) {
3827                         mtx_lock(&ctl_softc->pool_lock);
3828                         if (ctl_pool_acquire(emergency_pool) == 0) {
3829                                 io->io_hdr.pool = emergency_pool;
3830                                 emergency_pool->total_ctl_io++;
3831                                 /*
3832                                  * Need to bump this, otherwise
3833                                  * total_allocated and total_freed won't
3834                                  * match when we no longer have anything
3835                                  * outstanding.
3836                                  */
3837                                 emergency_pool->total_allocated++;
3838                         }
3839                         mtx_unlock(&ctl_softc->pool_lock);
3840                 } else
3841                         io->io_hdr.pool = NULL;
3842         }
3843
3844         return (io);
3845 }
3846
3847 void
3848 ctl_free_io(union ctl_io *io)
3849 {
3850         if (io == NULL)
3851                 return;
3852
3853         /*
3854          * If this ctl_io has a pool, return it to that pool.
3855          */
3856         if (io->io_hdr.pool != NULL) {
3857                 struct ctl_io_pool *pool;
3858
3859                 pool = (struct ctl_io_pool *)io->io_hdr.pool;
3860                 mtx_lock(&pool->ctl_softc->pool_lock);
3861                 io->io_hdr.io_type = 0xff;
3862                 STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3863                 pool->total_freed++;
3864                 pool->free_ctl_io++;
3865                 ctl_pool_release(pool);
3866                 mtx_unlock(&pool->ctl_softc->pool_lock);
3867         } else {
3868                 /*
3869                  * Otherwise, just free it.  We probably malloced it and
3870                  * the emergency pool wasn't available.
3871                  */
3872                 free(io, M_CTLIO);
3873         }
3874
3875 }
3876
3877 void
3878 ctl_zero_io(union ctl_io *io)
3879 {
3880         void *pool_ref;
3881
3882         if (io == NULL)
3883                 return;
3884
3885         /*
3886          * May need to preserve linked list pointers at some point too.
3887          */
3888         pool_ref = io->io_hdr.pool;
3889
3890         memset(io, 0, sizeof(*io));
3891
3892         io->io_hdr.pool = pool_ref;
3893 }
3894
3895 /*
3896  * This routine is currently used for internal copies of ctl_ios that need
3897  * to persist for some reason after we've already returned status to the
3898  * FETD.  (Thus the flag set.)
3899  *
3900  * XXX XXX
3901  * Note that this makes a blind copy of all fields in the ctl_io, except
3902  * for the pool reference.  This includes any memory that has been
3903  * allocated!  That memory will no longer be valid after done has been
3904  * called, so this would be VERY DANGEROUS for command that actually does
3905  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3906  * start and stop commands, which don't transfer any data, so this is not a
3907  * problem.  If it is used for anything else, the caller would also need to
3908  * allocate data buffer space and this routine would need to be modified to
3909  * copy the data buffer(s) as well.
3910  */
3911 void
3912 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3913 {
3914         void *pool_ref;
3915
3916         if ((src == NULL)
3917          || (dest == NULL))
3918                 return;
3919
3920         /*
3921          * May need to preserve linked list pointers at some point too.
3922          */
3923         pool_ref = dest->io_hdr.pool;
3924
3925         memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3926
3927         dest->io_hdr.pool = pool_ref;
3928         /*
3929          * We need to know that this is an internal copy, and doesn't need
3930          * to get passed back to the FETD that allocated it.
3931          */
3932         dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3933 }
3934
3935 #ifdef NEEDTOPORT
3936 static void
3937 ctl_update_power_subpage(struct copan_power_subpage *page)
3938 {
3939         int num_luns, num_partitions, config_type;
3940         struct ctl_softc *softc;
3941         cs_BOOL_t aor_present, shelf_50pct_power;
3942         cs_raidset_personality_t rs_type;
3943         int max_active_luns;
3944
3945         softc = control_softc;
3946
3947         /* subtract out the processor LUN */
3948         num_luns = softc->num_luns - 1;
3949         /*
3950          * Default to 7 LUNs active, which was the only number we allowed
3951          * in the past.
3952          */
3953         max_active_luns = 7;
3954
3955         num_partitions = config_GetRsPartitionInfo();
3956         config_type = config_GetConfigType();
3957         shelf_50pct_power = config_GetShelfPowerMode();
3958         aor_present = config_IsAorRsPresent();
3959
3960         rs_type = ddb_GetRsRaidType(1);
3961         if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3962          && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3963                 EPRINT(0, "Unsupported RS type %d!", rs_type);
3964         }
3965
3966
3967         page->total_luns = num_luns;
3968
3969         switch (config_type) {
3970         case 40:
3971                 /*
3972                  * In a 40 drive configuration, it doesn't matter what DC
3973                  * cards we have, whether we have AOR enabled or not,
3974                  * partitioning or not, or what type of RAIDset we have.
3975                  * In that scenario, we can power up every LUN we present
3976                  * to the user.
3977                  */
3978                 max_active_luns = num_luns;
3979
3980                 break;
3981         case 64:
3982                 if (shelf_50pct_power == CS_FALSE) {
3983                         /* 25% power */
3984                         if (aor_present == CS_TRUE) {
3985                                 if (rs_type ==
3986                                      CS_RAIDSET_PERSONALITY_RAID5) {
3987                                         max_active_luns = 7;
3988                                 } else if (rs_type ==
3989                                          CS_RAIDSET_PERSONALITY_RAID1){
3990                                         max_active_luns = 14;
3991                                 } else {
3992                                         /* XXX KDM now what?? */
3993                                 }
3994                         } else {
3995                                 if (rs_type ==
3996                                      CS_RAIDSET_PERSONALITY_RAID5) {
3997                                         max_active_luns = 8;
3998                                 } else if (rs_type ==
3999                                          CS_RAIDSET_PERSONALITY_RAID1){
4000                                         max_active_luns = 16;
4001                                 } else {
4002                                         /* XXX KDM now what?? */
4003                                 }
4004                         }
4005                 } else {
4006                         /* 50% power */
4007                         /*
4008                          * With 50% power in a 64 drive configuration, we
4009                          * can power all LUNs we present.
4010                          */
4011                         max_active_luns = num_luns;
4012                 }
4013                 break;
4014         case 112:
4015                 if (shelf_50pct_power == CS_FALSE) {
4016                         /* 25% power */
4017                         if (aor_present == CS_TRUE) {
4018                                 if (rs_type ==
4019                                      CS_RAIDSET_PERSONALITY_RAID5) {
4020                                         max_active_luns = 7;
4021                                 } else if (rs_type ==
4022                                          CS_RAIDSET_PERSONALITY_RAID1){
4023                                         max_active_luns = 14;
4024                                 } else {
4025                                         /* XXX KDM now what?? */
4026                                 }
4027                         } else {
4028                                 if (rs_type ==
4029                                      CS_RAIDSET_PERSONALITY_RAID5) {
4030                                         max_active_luns = 8;
4031                                 } else if (rs_type ==
4032                                          CS_RAIDSET_PERSONALITY_RAID1){
4033                                         max_active_luns = 16;
4034                                 } else {
4035                                         /* XXX KDM now what?? */
4036                                 }
4037                         }
4038                 } else {
4039                         /* 50% power */
4040                         if (aor_present == CS_TRUE) {
4041                                 if (rs_type ==
4042                                      CS_RAIDSET_PERSONALITY_RAID5) {
4043                                         max_active_luns = 14;
4044                                 } else if (rs_type ==
4045                                          CS_RAIDSET_PERSONALITY_RAID1){
4046                                         /*
4047                                          * We're assuming here that disk
4048                                          * caching is enabled, and so we're
4049                                          * able to power up half of each
4050                                          * LUN, and cache all writes.
4051                                          */
4052                                         max_active_luns = num_luns;
4053                                 } else {
4054                                         /* XXX KDM now what?? */
4055                                 }
4056                         } else {
4057                                 if (rs_type ==
4058                                      CS_RAIDSET_PERSONALITY_RAID5) {
4059                                         max_active_luns = 15;
4060                                 } else if (rs_type ==
4061                                          CS_RAIDSET_PERSONALITY_RAID1){
4062                                         max_active_luns = 30;
4063                                 } else {
4064                                         /* XXX KDM now what?? */
4065                                 }
4066                         }
4067                 }
4068                 break;
4069         default:
4070                 /*
4071                  * In this case, we have an unknown configuration, so we
4072                  * just use the default from above.
4073                  */
4074                 break;
4075         }
4076
4077         page->max_active_luns = max_active_luns;
4078 #if 0
4079         printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4080                page->total_luns, page->max_active_luns);
4081 #endif
4082 }
4083 #endif /* NEEDTOPORT */
4084
4085 /*
4086  * This routine could be used in the future to load default and/or saved
4087  * mode page parameters for a particuar lun.
4088  */
4089 static int
4090 ctl_init_page_index(struct ctl_lun *lun)
4091 {
4092         int i;
4093         struct ctl_page_index *page_index;
4094         struct ctl_softc *softc;
4095
4096         memcpy(&lun->mode_pages.index, page_index_template,
4097                sizeof(page_index_template));
4098
4099         softc = lun->ctl_softc;
4100
4101         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4102
4103                 page_index = &lun->mode_pages.index[i];
4104                 /*
4105                  * If this is a disk-only mode page, there's no point in
4106                  * setting it up.  For some pages, we have to have some
4107                  * basic information about the disk in order to calculate the
4108                  * mode page data.
4109                  */
4110                 if ((lun->be_lun->lun_type != T_DIRECT)
4111                  && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4112                         continue;
4113
4114                 switch (page_index->page_code & SMPH_PC_MASK) {
4115                 case SMS_FORMAT_DEVICE_PAGE: {
4116                         struct scsi_format_page *format_page;
4117
4118                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4119                                 panic("subpage is incorrect!");
4120
4121                         /*
4122                          * Sectors per track are set above.  Bytes per
4123                          * sector need to be set here on a per-LUN basis.
4124                          */
4125                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4126                                &format_page_default,
4127                                sizeof(format_page_default));
4128                         memcpy(&lun->mode_pages.format_page[
4129                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4130                                sizeof(format_page_changeable));
4131                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4132                                &format_page_default,
4133                                sizeof(format_page_default));
4134                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4135                                &format_page_default,
4136                                sizeof(format_page_default));
4137
4138                         format_page = &lun->mode_pages.format_page[
4139                                 CTL_PAGE_CURRENT];
4140                         scsi_ulto2b(lun->be_lun->blocksize,
4141                                     format_page->bytes_per_sector);
4142
4143                         format_page = &lun->mode_pages.format_page[
4144                                 CTL_PAGE_DEFAULT];
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_SAVED];
4150                         scsi_ulto2b(lun->be_lun->blocksize,
4151                                     format_page->bytes_per_sector);
4152
4153                         page_index->page_data =
4154                                 (uint8_t *)lun->mode_pages.format_page;
4155                         break;
4156                 }
4157                 case SMS_RIGID_DISK_PAGE: {
4158                         struct scsi_rigid_disk_page *rigid_disk_page;
4159                         uint32_t sectors_per_cylinder;
4160                         uint64_t cylinders;
4161 #ifndef __XSCALE__
4162                         int shift;
4163 #endif /* !__XSCALE__ */
4164
4165                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4166                                 panic("invalid subpage value %d",
4167                                       page_index->subpage);
4168
4169                         /*
4170                          * Rotation rate and sectors per track are set
4171                          * above.  We calculate the cylinders here based on
4172                          * capacity.  Due to the number of heads and
4173                          * sectors per track we're using, smaller arrays
4174                          * may turn out to have 0 cylinders.  Linux and
4175                          * FreeBSD don't pay attention to these mode pages
4176                          * to figure out capacity, but Solaris does.  It
4177                          * seems to deal with 0 cylinders just fine, and
4178                          * works out a fake geometry based on the capacity.
4179                          */
4180                         memcpy(&lun->mode_pages.rigid_disk_page[
4181                                CTL_PAGE_CURRENT], &rigid_disk_page_default,
4182                                sizeof(rigid_disk_page_default));
4183                         memcpy(&lun->mode_pages.rigid_disk_page[
4184                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4185                                sizeof(rigid_disk_page_changeable));
4186                         memcpy(&lun->mode_pages.rigid_disk_page[
4187                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4188                                sizeof(rigid_disk_page_default));
4189                         memcpy(&lun->mode_pages.rigid_disk_page[
4190                                CTL_PAGE_SAVED], &rigid_disk_page_default,
4191                                sizeof(rigid_disk_page_default));
4192
4193                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4194                                 CTL_DEFAULT_HEADS;
4195
4196                         /*
4197                          * The divide method here will be more accurate,
4198                          * probably, but results in floating point being
4199                          * used in the kernel on i386 (__udivdi3()).  On the
4200                          * XScale, though, __udivdi3() is implemented in
4201                          * software.
4202                          *
4203                          * The shift method for cylinder calculation is
4204                          * accurate if sectors_per_cylinder is a power of
4205                          * 2.  Otherwise it might be slightly off -- you
4206                          * might have a bit of a truncation problem.
4207                          */
4208 #ifdef  __XSCALE__
4209                         cylinders = (lun->be_lun->maxlba + 1) /
4210                                 sectors_per_cylinder;
4211 #else
4212                         for (shift = 31; shift > 0; shift--) {
4213                                 if (sectors_per_cylinder & (1 << shift))
4214                                         break;
4215                         }
4216                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4217 #endif
4218
4219                         /*
4220                          * We've basically got 3 bytes, or 24 bits for the
4221                          * cylinder size in the mode page.  If we're over,
4222                          * just round down to 2^24.
4223                          */
4224                         if (cylinders > 0xffffff)
4225                                 cylinders = 0xffffff;
4226
4227                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4228                                 CTL_PAGE_CURRENT];
4229                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4230
4231                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4232                                 CTL_PAGE_DEFAULT];
4233                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4234
4235                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4236                                 CTL_PAGE_SAVED];
4237                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4238
4239                         page_index->page_data =
4240                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4241                         break;
4242                 }
4243                 case SMS_CACHING_PAGE: {
4244
4245                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4246                                 panic("invalid subpage value %d",
4247                                       page_index->subpage);
4248                         /*
4249                          * Defaults should be okay here, no calculations
4250                          * needed.
4251                          */
4252                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4253                                &caching_page_default,
4254                                sizeof(caching_page_default));
4255                         memcpy(&lun->mode_pages.caching_page[
4256                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4257                                sizeof(caching_page_changeable));
4258                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4259                                &caching_page_default,
4260                                sizeof(caching_page_default));
4261                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4262                                &caching_page_default,
4263                                sizeof(caching_page_default));
4264                         page_index->page_data =
4265                                 (uint8_t *)lun->mode_pages.caching_page;
4266                         break;
4267                 }
4268                 case SMS_CONTROL_MODE_PAGE: {
4269
4270                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4271                                 panic("invalid subpage value %d",
4272                                       page_index->subpage);
4273
4274                         /*
4275                          * Defaults should be okay here, no calculations
4276                          * needed.
4277                          */
4278                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4279                                &control_page_default,
4280                                sizeof(control_page_default));
4281                         memcpy(&lun->mode_pages.control_page[
4282                                CTL_PAGE_CHANGEABLE], &control_page_changeable,
4283                                sizeof(control_page_changeable));
4284                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4285                                &control_page_default,
4286                                sizeof(control_page_default));
4287                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4288                                &control_page_default,
4289                                sizeof(control_page_default));
4290                         page_index->page_data =
4291                                 (uint8_t *)lun->mode_pages.control_page;
4292                         break;
4293
4294                 }
4295                 case SMS_VENDOR_SPECIFIC_PAGE:{
4296                         switch (page_index->subpage) {
4297                         case PWR_SUBPAGE_CODE: {
4298                                 struct copan_power_subpage *current_page,
4299                                                            *saved_page;
4300
4301                                 memcpy(&lun->mode_pages.power_subpage[
4302                                        CTL_PAGE_CURRENT],
4303                                        &power_page_default,
4304                                        sizeof(power_page_default));
4305                                 memcpy(&lun->mode_pages.power_subpage[
4306                                        CTL_PAGE_CHANGEABLE],
4307                                        &power_page_changeable,
4308                                        sizeof(power_page_changeable));
4309                                 memcpy(&lun->mode_pages.power_subpage[
4310                                        CTL_PAGE_DEFAULT],
4311                                        &power_page_default,
4312                                        sizeof(power_page_default));
4313                                 memcpy(&lun->mode_pages.power_subpage[
4314                                        CTL_PAGE_SAVED],
4315                                        &power_page_default,
4316                                        sizeof(power_page_default));
4317                                 page_index->page_data =
4318                                     (uint8_t *)lun->mode_pages.power_subpage;
4319
4320                                 current_page = (struct copan_power_subpage *)
4321                                         (page_index->page_data +
4322                                          (page_index->page_len *
4323                                           CTL_PAGE_CURRENT));
4324                                 saved_page = (struct copan_power_subpage *)
4325                                         (page_index->page_data +
4326                                          (page_index->page_len *
4327                                           CTL_PAGE_SAVED));
4328                                 break;
4329                         }
4330                         case APS_SUBPAGE_CODE: {
4331                                 struct copan_aps_subpage *current_page,
4332                                                          *saved_page;
4333
4334                                 // This gets set multiple times but
4335                                 // it should always be the same. It's
4336                                 // only done during init so who cares.
4337                                 index_to_aps_page = i;
4338
4339                                 memcpy(&lun->mode_pages.aps_subpage[
4340                                        CTL_PAGE_CURRENT],
4341                                        &aps_page_default,
4342                                        sizeof(aps_page_default));
4343                                 memcpy(&lun->mode_pages.aps_subpage[
4344                                        CTL_PAGE_CHANGEABLE],
4345                                        &aps_page_changeable,
4346                                        sizeof(aps_page_changeable));
4347                                 memcpy(&lun->mode_pages.aps_subpage[
4348                                        CTL_PAGE_DEFAULT],
4349                                        &aps_page_default,
4350                                        sizeof(aps_page_default));
4351                                 memcpy(&lun->mode_pages.aps_subpage[
4352                                        CTL_PAGE_SAVED],
4353                                        &aps_page_default,
4354                                        sizeof(aps_page_default));
4355                                 page_index->page_data =
4356                                         (uint8_t *)lun->mode_pages.aps_subpage;
4357
4358                                 current_page = (struct copan_aps_subpage *)
4359                                         (page_index->page_data +
4360                                          (page_index->page_len *
4361                                           CTL_PAGE_CURRENT));
4362                                 saved_page = (struct copan_aps_subpage *)
4363                                         (page_index->page_data +
4364                                          (page_index->page_len *
4365                                           CTL_PAGE_SAVED));
4366                                 break;
4367                         }
4368                         case DBGCNF_SUBPAGE_CODE: {
4369                                 struct copan_debugconf_subpage *current_page,
4370                                                                *saved_page;
4371
4372                                 memcpy(&lun->mode_pages.debugconf_subpage[
4373                                        CTL_PAGE_CURRENT],
4374                                        &debugconf_page_default,
4375                                        sizeof(debugconf_page_default));
4376                                 memcpy(&lun->mode_pages.debugconf_subpage[
4377                                        CTL_PAGE_CHANGEABLE],
4378                                        &debugconf_page_changeable,
4379                                        sizeof(debugconf_page_changeable));
4380                                 memcpy(&lun->mode_pages.debugconf_subpage[
4381                                        CTL_PAGE_DEFAULT],
4382                                        &debugconf_page_default,
4383                                        sizeof(debugconf_page_default));
4384                                 memcpy(&lun->mode_pages.debugconf_subpage[
4385                                        CTL_PAGE_SAVED],
4386                                        &debugconf_page_default,
4387                                        sizeof(debugconf_page_default));
4388                                 page_index->page_data =
4389                                         (uint8_t *)lun->mode_pages.debugconf_subpage;
4390
4391                                 current_page = (struct copan_debugconf_subpage *)
4392                                         (page_index->page_data +
4393                                          (page_index->page_len *
4394                                           CTL_PAGE_CURRENT));
4395                                 saved_page = (struct copan_debugconf_subpage *)
4396                                         (page_index->page_data +
4397                                          (page_index->page_len *
4398                                           CTL_PAGE_SAVED));
4399                                 break;
4400                         }
4401                         default:
4402                                 panic("invalid subpage value %d",
4403                                       page_index->subpage);
4404                                 break;
4405                         }
4406                         break;
4407                 }
4408                 default:
4409                         panic("invalid page value %d",
4410                               page_index->page_code & SMPH_PC_MASK);
4411                         break;
4412         }
4413         }
4414
4415         return (CTL_RETVAL_COMPLETE);
4416 }
4417
4418 /*
4419  * LUN allocation.
4420  *
4421  * Requirements:
4422  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4423  *   wants us to allocate the LUN and he can block.
4424  * - ctl_softc is always set
4425  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4426  *
4427  * Returns 0 for success, non-zero (errno) for failure.
4428  */
4429 static int
4430 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4431               struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4432 {
4433         struct ctl_lun *nlun, *lun;
4434         struct ctl_port *port;
4435         struct scsi_vpd_id_descriptor *desc;
4436         struct scsi_vpd_id_t10 *t10id;
4437         const char *eui, *naa, *scsiname, *vendor;
4438         int lun_number, i, lun_malloced;
4439         int devidlen, idlen1, idlen2 = 0, len;
4440
4441         if (be_lun == NULL)
4442                 return (EINVAL);
4443
4444         /*
4445          * We currently only support Direct Access or Processor LUN types.
4446          */
4447         switch (be_lun->lun_type) {
4448         case T_DIRECT:
4449                 break;
4450         case T_PROCESSOR:
4451                 break;
4452         case T_SEQUENTIAL:
4453         case T_CHANGER:
4454         default:
4455                 be_lun->lun_config_status(be_lun->be_lun,
4456                                           CTL_LUN_CONFIG_FAILURE);
4457                 break;
4458         }
4459         if (ctl_lun == NULL) {
4460                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4461                 lun_malloced = 1;
4462         } else {
4463                 lun_malloced = 0;
4464                 lun = ctl_lun;
4465         }
4466
4467         memset(lun, 0, sizeof(*lun));
4468         if (lun_malloced)
4469                 lun->flags = CTL_LUN_MALLOCED;
4470
4471         /* Generate LUN ID. */
4472         devidlen = max(CTL_DEVID_MIN_LEN,
4473             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4474         idlen1 = sizeof(*t10id) + devidlen;
4475         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4476         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4477         if (scsiname != NULL) {
4478                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4479                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4480         }
4481         eui = ctl_get_opt(&be_lun->options, "eui");
4482         if (eui != NULL) {
4483                 len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4484         }
4485         naa = ctl_get_opt(&be_lun->options, "naa");
4486         if (naa != NULL) {
4487                 len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4488         }
4489         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4490             M_CTL, M_WAITOK | M_ZERO);
4491         lun->lun_devid->len = len;
4492         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4493         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4494         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4495         desc->length = idlen1;
4496         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4497         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4498         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4499                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4500         } else {
4501                 strncpy(t10id->vendor, vendor,
4502                     min(sizeof(t10id->vendor), strlen(vendor)));
4503         }
4504         strncpy((char *)t10id->vendor_spec_id,
4505             (char *)be_lun->device_id, devidlen);
4506         if (scsiname != NULL) {
4507                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4508                     desc->length);
4509                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4510                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4511                     SVPD_ID_TYPE_SCSI_NAME;
4512                 desc->length = idlen2;
4513                 strlcpy(desc->identifier, scsiname, idlen2);
4514         }
4515         if (eui != NULL) {
4516                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4517                     desc->length);
4518                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4519                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4520                     SVPD_ID_TYPE_EUI64;
4521                 desc->length = 8;
4522                 scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4523         }
4524         if (naa != NULL) {
4525                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4526                     desc->length);
4527                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4528                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4529                     SVPD_ID_TYPE_NAA;
4530                 desc->length = 8;
4531                 scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4532         }
4533
4534         mtx_lock(&ctl_softc->ctl_lock);
4535         /*
4536          * See if the caller requested a particular LUN number.  If so, see
4537          * if it is available.  Otherwise, allocate the first available LUN.
4538          */
4539         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4540                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4541                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4542                         mtx_unlock(&ctl_softc->ctl_lock);
4543                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4544                                 printf("ctl: requested LUN ID %d is higher "
4545                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4546                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4547                         } else {
4548                                 /*
4549                                  * XXX KDM return an error, or just assign
4550                                  * another LUN ID in this case??
4551                                  */
4552                                 printf("ctl: requested LUN ID %d is already "
4553                                        "in use\n", be_lun->req_lun_id);
4554                         }
4555                         if (lun->flags & CTL_LUN_MALLOCED)
4556                                 free(lun, M_CTL);
4557                         be_lun->lun_config_status(be_lun->be_lun,
4558                                                   CTL_LUN_CONFIG_FAILURE);
4559                         return (ENOSPC);
4560                 }
4561                 lun_number = be_lun->req_lun_id;
4562         } else {
4563                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4564                 if (lun_number == -1) {
4565                         mtx_unlock(&ctl_softc->ctl_lock);
4566                         printf("ctl: can't allocate LUN on target %ju, out of "
4567                                "LUNs\n", (uintmax_t)target_id.id);
4568                         if (lun->flags & CTL_LUN_MALLOCED)
4569                                 free(lun, M_CTL);
4570                         be_lun->lun_config_status(be_lun->be_lun,
4571                                                   CTL_LUN_CONFIG_FAILURE);
4572                         return (ENOSPC);
4573                 }
4574         }
4575         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4576
4577         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4578         lun->target = target_id;
4579         lun->lun = lun_number;
4580         lun->be_lun = be_lun;
4581         /*
4582          * The processor LUN is always enabled.  Disk LUNs come on line
4583          * disabled, and must be enabled by the backend.
4584          */
4585         lun->flags |= CTL_LUN_DISABLED;
4586         lun->backend = be_lun->be;
4587         be_lun->ctl_lun = lun;
4588         be_lun->lun_id = lun_number;
4589         atomic_add_int(&be_lun->be->num_luns, 1);
4590         if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4591                 lun->flags |= CTL_LUN_STOPPED;
4592
4593         if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4594                 lun->flags |= CTL_LUN_INOPERABLE;
4595
4596         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4597                 lun->flags |= CTL_LUN_PRIMARY_SC;
4598
4599         lun->ctl_softc = ctl_softc;
4600         TAILQ_INIT(&lun->ooa_queue);
4601         TAILQ_INIT(&lun->blocked_queue);
4602         STAILQ_INIT(&lun->error_list);
4603         ctl_tpc_lun_init(lun);
4604
4605         /*
4606          * Initialize the mode page index.
4607          */
4608         ctl_init_page_index(lun);
4609
4610         /*
4611          * Set the poweron UA for all initiators on this LUN only.
4612          */
4613         for (i = 0; i < CTL_MAX_INITIATORS; i++)
4614                 lun->pending_ua[i] = CTL_UA_POWERON;
4615
4616         /*
4617          * Now, before we insert this lun on the lun list, set the lun
4618          * inventory changed UA for all other luns.
4619          */
4620         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4621                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4622                         nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4623                 }
4624         }
4625
4626         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4627
4628         ctl_softc->ctl_luns[lun_number] = lun;
4629
4630         ctl_softc->num_luns++;
4631
4632         /* Setup statistics gathering */
4633         lun->stats.device_type = be_lun->lun_type;
4634         lun->stats.lun_number = lun_number;
4635         if (lun->stats.device_type == T_DIRECT)
4636                 lun->stats.blocksize = be_lun->blocksize;
4637         else
4638                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4639         for (i = 0;i < CTL_MAX_PORTS;i++)
4640                 lun->stats.ports[i].targ_port = i;
4641
4642         mtx_unlock(&ctl_softc->ctl_lock);
4643
4644         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4645
4646         /*
4647          * Run through each registered FETD and bring it online if it isn't
4648          * already.  Enable the target ID if it hasn't been enabled, and
4649          * enable this particular LUN.
4650          */
4651         STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4652                 int retval;
4653
4654                 retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4655                 if (retval != 0) {
4656                         printf("ctl_alloc_lun: FETD %s port %d returned error "
4657                                "%d for lun_enable on target %ju lun %d\n",
4658                                port->port_name, port->targ_port, retval,
4659                                (uintmax_t)target_id.id, lun_number);
4660                 } else
4661                         port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4662         }
4663         return (0);
4664 }
4665
4666 /*
4667  * Delete a LUN.
4668  * Assumptions:
4669  * - LUN has already been marked invalid and any pending I/O has been taken
4670  *   care of.
4671  */
4672 static int
4673 ctl_free_lun(struct ctl_lun *lun)
4674 {
4675         struct ctl_softc *softc;
4676 #if 0
4677         struct ctl_port *port;
4678 #endif
4679         struct ctl_lun *nlun;
4680         int i;
4681
4682         softc = lun->ctl_softc;
4683
4684         mtx_assert(&softc->ctl_lock, MA_OWNED);
4685
4686         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4687
4688         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4689
4690         softc->ctl_luns[lun->lun] = NULL;
4691
4692         if (!TAILQ_EMPTY(&lun->ooa_queue))
4693                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4694
4695         softc->num_luns--;
4696
4697         /*
4698          * XXX KDM this scheme only works for a single target/multiple LUN
4699          * setup.  It needs to be revamped for a multiple target scheme.
4700          *
4701          * XXX KDM this results in port->lun_disable() getting called twice,
4702          * once when ctl_disable_lun() is called, and a second time here.
4703          * We really need to re-think the LUN disable semantics.  There
4704          * should probably be several steps/levels to LUN removal:
4705          *  - disable
4706          *  - invalidate
4707          *  - free
4708          *
4709          * Right now we only have a disable method when communicating to
4710          * the front end ports, at least for individual LUNs.
4711          */
4712 #if 0
4713         STAILQ_FOREACH(port, &softc->port_list, links) {
4714                 int retval;
4715
4716                 retval = port->lun_disable(port->targ_lun_arg, lun->target,
4717                                          lun->lun);
4718                 if (retval != 0) {
4719                         printf("ctl_free_lun: FETD %s port %d returned error "
4720                                "%d for lun_disable on target %ju lun %jd\n",
4721                                port->port_name, port->targ_port, retval,
4722                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4723                 }
4724
4725                 if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4726                         port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4727
4728                         retval = port->targ_disable(port->targ_lun_arg,lun->target);
4729                         if (retval != 0) {
4730                                 printf("ctl_free_lun: FETD %s port %d "
4731                                        "returned error %d for targ_disable on "
4732                                        "target %ju\n", port->port_name,
4733                                        port->targ_port, retval,
4734                                        (uintmax_t)lun->target.id);
4735                         } else
4736                                 port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4737
4738                         if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4739                                 continue;
4740
4741 #if 0
4742                         port->port_offline(port->onoff_arg);
4743                         port->status &= ~CTL_PORT_STATUS_ONLINE;
4744 #endif
4745                 }
4746         }
4747 #endif
4748
4749         /*
4750          * Tell the backend to free resources, if this LUN has a backend.
4751          */
4752         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4753         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4754
4755         ctl_tpc_lun_shutdown(lun);
4756         mtx_destroy(&lun->lun_lock);
4757         free(lun->lun_devid, M_CTL);
4758         if (lun->flags & CTL_LUN_MALLOCED)
4759                 free(lun, M_CTL);
4760
4761         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4762                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4763                         nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4764                 }
4765         }
4766
4767         return (0);
4768 }
4769
4770 static void
4771 ctl_create_lun(struct ctl_be_lun *be_lun)
4772 {
4773         struct ctl_softc *ctl_softc;
4774
4775         ctl_softc = control_softc;
4776
4777         /*
4778          * ctl_alloc_lun() should handle all potential failure cases.
4779          */
4780         ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4781 }
4782
4783 int
4784 ctl_add_lun(struct ctl_be_lun *be_lun)
4785 {
4786         struct ctl_softc *ctl_softc = control_softc;
4787
4788         mtx_lock(&ctl_softc->ctl_lock);
4789         STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4790         mtx_unlock(&ctl_softc->ctl_lock);
4791         wakeup(&ctl_softc->pending_lun_queue);
4792
4793         return (0);
4794 }
4795
4796 int
4797 ctl_enable_lun(struct ctl_be_lun *be_lun)
4798 {
4799         struct ctl_softc *ctl_softc;
4800         struct ctl_port *port, *nport;
4801         struct ctl_lun *lun;
4802         int retval;
4803
4804         ctl_softc = control_softc;
4805
4806         lun = (struct ctl_lun *)be_lun->ctl_lun;
4807
4808         mtx_lock(&ctl_softc->ctl_lock);
4809         mtx_lock(&lun->lun_lock);
4810         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4811                 /*
4812                  * eh?  Why did we get called if the LUN is already
4813                  * enabled?
4814                  */
4815                 mtx_unlock(&lun->lun_lock);
4816                 mtx_unlock(&ctl_softc->ctl_lock);
4817                 return (0);
4818         }
4819         lun->flags &= ~CTL_LUN_DISABLED;
4820         mtx_unlock(&lun->lun_lock);
4821
4822         for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4823                 nport = STAILQ_NEXT(port, links);
4824
4825                 /*
4826                  * Drop the lock while we call the FETD's enable routine.
4827                  * This can lead to a callback into CTL (at least in the
4828                  * case of the internal initiator frontend.
4829                  */
4830                 mtx_unlock(&ctl_softc->ctl_lock);
4831                 retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4832                 mtx_lock(&ctl_softc->ctl_lock);
4833                 if (retval != 0) {
4834                         printf("%s: FETD %s port %d returned error "
4835                                "%d for lun_enable on target %ju lun %jd\n",
4836                                __func__, port->port_name, port->targ_port, retval,
4837                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4838                 }
4839 #if 0
4840                  else {
4841             /* NOTE:  TODO:  why does lun enable affect port status? */
4842                         port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4843                 }
4844 #endif
4845         }
4846
4847         mtx_unlock(&ctl_softc->ctl_lock);
4848
4849         return (0);
4850 }
4851
4852 int
4853 ctl_disable_lun(struct ctl_be_lun *be_lun)
4854 {
4855         struct ctl_softc *ctl_softc;
4856         struct ctl_port *port;
4857         struct ctl_lun *lun;
4858         int retval;
4859
4860         ctl_softc = control_softc;
4861
4862         lun = (struct ctl_lun *)be_lun->ctl_lun;
4863
4864         mtx_lock(&ctl_softc->ctl_lock);
4865         mtx_lock(&lun->lun_lock);
4866         if (lun->flags & CTL_LUN_DISABLED) {
4867                 mtx_unlock(&lun->lun_lock);
4868                 mtx_unlock(&ctl_softc->ctl_lock);
4869                 return (0);
4870         }
4871         lun->flags |= CTL_LUN_DISABLED;
4872         mtx_unlock(&lun->lun_lock);
4873
4874         STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4875                 mtx_unlock(&ctl_softc->ctl_lock);
4876                 /*
4877                  * Drop the lock before we call the frontend's disable
4878                  * routine, to avoid lock order reversals.
4879                  *
4880                  * XXX KDM what happens if the frontend list changes while
4881                  * we're traversing it?  It's unlikely, but should be handled.
4882                  */
4883                 retval = port->lun_disable(port->targ_lun_arg, lun->target,
4884                                          lun->lun);
4885                 mtx_lock(&ctl_softc->ctl_lock);
4886                 if (retval != 0) {
4887                         printf("ctl_alloc_lun: FETD %s port %d returned error "
4888                                "%d for lun_disable on target %ju lun %jd\n",
4889                                port->port_name, port->targ_port, retval,
4890                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4891                 }
4892         }
4893
4894         mtx_unlock(&ctl_softc->ctl_lock);
4895
4896         return (0);
4897 }
4898
4899 int
4900 ctl_start_lun(struct ctl_be_lun *be_lun)
4901 {
4902         struct ctl_softc *ctl_softc;
4903         struct ctl_lun *lun;
4904
4905         ctl_softc = control_softc;
4906
4907         lun = (struct ctl_lun *)be_lun->ctl_lun;
4908
4909         mtx_lock(&lun->lun_lock);
4910         lun->flags &= ~CTL_LUN_STOPPED;
4911         mtx_unlock(&lun->lun_lock);
4912
4913         return (0);
4914 }
4915
4916 int
4917 ctl_stop_lun(struct ctl_be_lun *be_lun)
4918 {
4919         struct ctl_softc *ctl_softc;
4920         struct ctl_lun *lun;
4921
4922         ctl_softc = control_softc;
4923
4924         lun = (struct ctl_lun *)be_lun->ctl_lun;
4925
4926         mtx_lock(&lun->lun_lock);
4927         lun->flags |= CTL_LUN_STOPPED;
4928         mtx_unlock(&lun->lun_lock);
4929
4930         return (0);
4931 }
4932
4933 int
4934 ctl_lun_offline(struct ctl_be_lun *be_lun)
4935 {
4936         struct ctl_softc *ctl_softc;
4937         struct ctl_lun *lun;
4938
4939         ctl_softc = control_softc;
4940
4941         lun = (struct ctl_lun *)be_lun->ctl_lun;
4942
4943         mtx_lock(&lun->lun_lock);
4944         lun->flags |= CTL_LUN_OFFLINE;
4945         mtx_unlock(&lun->lun_lock);
4946
4947         return (0);
4948 }
4949
4950 int
4951 ctl_lun_online(struct ctl_be_lun *be_lun)
4952 {
4953         struct ctl_softc *ctl_softc;
4954         struct ctl_lun *lun;
4955
4956         ctl_softc = control_softc;
4957
4958         lun = (struct ctl_lun *)be_lun->ctl_lun;
4959
4960         mtx_lock(&lun->lun_lock);
4961         lun->flags &= ~CTL_LUN_OFFLINE;
4962         mtx_unlock(&lun->lun_lock);
4963
4964         return (0);
4965 }
4966
4967 int
4968 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4969 {
4970         struct ctl_softc *ctl_softc;
4971         struct ctl_lun *lun;
4972
4973         ctl_softc = control_softc;
4974
4975         lun = (struct ctl_lun *)be_lun->ctl_lun;
4976
4977         mtx_lock(&lun->lun_lock);
4978
4979         /*
4980          * The LUN needs to be disabled before it can be marked invalid.
4981          */
4982         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4983                 mtx_unlock(&lun->lun_lock);
4984                 return (-1);
4985         }
4986         /*
4987          * Mark the LUN invalid.
4988          */
4989         lun->flags |= CTL_LUN_INVALID;
4990
4991         /*
4992          * If there is nothing in the OOA queue, go ahead and free the LUN.
4993          * If we have something in the OOA queue, we'll free it when the
4994          * last I/O completes.
4995          */
4996         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4997                 mtx_unlock(&lun->lun_lock);
4998                 mtx_lock(&ctl_softc->ctl_lock);
4999                 ctl_free_lun(lun);
5000                 mtx_unlock(&ctl_softc->ctl_lock);
5001         } else
5002                 mtx_unlock(&lun->lun_lock);
5003
5004         return (0);
5005 }
5006
5007 int
5008 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5009 {
5010         struct ctl_softc *ctl_softc;
5011         struct ctl_lun *lun;
5012
5013         ctl_softc = control_softc;
5014         lun = (struct ctl_lun *)be_lun->ctl_lun;
5015
5016         mtx_lock(&lun->lun_lock);
5017         lun->flags |= CTL_LUN_INOPERABLE;
5018         mtx_unlock(&lun->lun_lock);
5019
5020         return (0);
5021 }
5022
5023 int
5024 ctl_lun_operable(struct ctl_be_lun *be_lun)
5025 {
5026         struct ctl_softc *ctl_softc;
5027         struct ctl_lun *lun;
5028
5029         ctl_softc = control_softc;
5030         lun = (struct ctl_lun *)be_lun->ctl_lun;
5031
5032         mtx_lock(&lun->lun_lock);
5033         lun->flags &= ~CTL_LUN_INOPERABLE;
5034         mtx_unlock(&lun->lun_lock);
5035
5036         return (0);
5037 }
5038
5039 int
5040 ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5041                    int lock)
5042 {
5043         struct ctl_softc *softc;
5044         struct ctl_lun *lun;
5045         struct copan_aps_subpage *current_sp;
5046         struct ctl_page_index *page_index;
5047         int i;
5048
5049         softc = control_softc;
5050
5051         mtx_lock(&softc->ctl_lock);
5052
5053         lun = (struct ctl_lun *)be_lun->ctl_lun;
5054         mtx_lock(&lun->lun_lock);
5055
5056         page_index = NULL;
5057         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5058                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5059                      APS_PAGE_CODE)
5060                         continue;
5061
5062                 if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5063                         continue;
5064                 page_index = &lun->mode_pages.index[i];
5065         }
5066
5067         if (page_index == NULL) {
5068                 mtx_unlock(&lun->lun_lock);
5069                 mtx_unlock(&softc->ctl_lock);
5070                 printf("%s: APS subpage not found for lun %ju!\n", __func__,
5071                        (uintmax_t)lun->lun);
5072                 return (1);
5073         }
5074 #if 0
5075         if ((softc->aps_locked_lun != 0)
5076          && (softc->aps_locked_lun != lun->lun)) {
5077                 printf("%s: attempt to lock LUN %llu when %llu is already "
5078                        "locked\n");
5079                 mtx_unlock(&lun->lun_lock);
5080                 mtx_unlock(&softc->ctl_lock);
5081                 return (1);
5082         }
5083 #endif
5084
5085         current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5086                 (page_index->page_len * CTL_PAGE_CURRENT));
5087
5088         if (lock != 0) {
5089                 current_sp->lock_active = APS_LOCK_ACTIVE;
5090                 softc->aps_locked_lun = lun->lun;
5091         } else {
5092                 current_sp->lock_active = 0;
5093                 softc->aps_locked_lun = 0;
5094         }
5095
5096
5097         /*
5098          * If we're in HA mode, try to send the lock message to the other
5099          * side.
5100          */
5101         if (ctl_is_single == 0) {
5102                 int isc_retval;
5103                 union ctl_ha_msg lock_msg;
5104
5105                 lock_msg.hdr.nexus = *nexus;
5106                 lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5107                 if (lock != 0)
5108                         lock_msg.aps.lock_flag = 1;
5109                 else
5110                         lock_msg.aps.lock_flag = 0;
5111                 isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5112                                          sizeof(lock_msg), 0);
5113                 if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5114                         printf("%s: APS (lock=%d) error returned from "
5115                                "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5116                         mtx_unlock(&lun->lun_lock);
5117                         mtx_unlock(&softc->ctl_lock);
5118                         return (1);
5119                 }
5120         }
5121
5122         mtx_unlock(&lun->lun_lock);
5123         mtx_unlock(&softc->ctl_lock);
5124
5125         return (0);
5126 }
5127
5128 void
5129 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5130 {
5131         struct ctl_lun *lun;
5132         struct ctl_softc *softc;
5133         int i;
5134
5135         softc = control_softc;
5136
5137         lun = (struct ctl_lun *)be_lun->ctl_lun;
5138
5139         mtx_lock(&lun->lun_lock);
5140
5141         for (i = 0; i < CTL_MAX_INITIATORS; i++)
5142                 lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5143
5144         mtx_unlock(&lun->lun_lock);
5145 }
5146
5147 /*
5148  * Backend "memory move is complete" callback for requests that never
5149  * make it down to say RAIDCore's configuration code.
5150  */
5151 int
5152 ctl_config_move_done(union ctl_io *io)
5153 {
5154         int retval;
5155
5156         retval = CTL_RETVAL_COMPLETE;
5157
5158
5159         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5160         /*
5161          * XXX KDM this shouldn't happen, but what if it does?
5162          */
5163         if (io->io_hdr.io_type != CTL_IO_SCSI)
5164                 panic("I/O type isn't CTL_IO_SCSI!");
5165
5166         if ((io->io_hdr.port_status == 0)
5167          && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5168          && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5169                 io->io_hdr.status = CTL_SUCCESS;
5170         else if ((io->io_hdr.port_status != 0)
5171               && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5172               && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5173                 /*
5174                  * For hardware error sense keys, the sense key
5175                  * specific value is defined to be a retry count,
5176                  * but we use it to pass back an internal FETD
5177                  * error code.  XXX KDM  Hopefully the FETD is only
5178                  * using 16 bits for an error code, since that's
5179                  * all the space we have in the sks field.
5180                  */
5181                 ctl_set_internal_failure(&io->scsiio,
5182                                          /*sks_valid*/ 1,
5183                                          /*retry_count*/
5184                                          io->io_hdr.port_status);
5185                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5186                         free(io->scsiio.kern_data_ptr, M_CTL);
5187                 ctl_done(io);
5188                 goto bailout;
5189         }
5190
5191         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5192          || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5193          || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5194                 /*
5195                  * XXX KDM just assuming a single pointer here, and not a
5196                  * S/G list.  If we start using S/G lists for config data,
5197                  * we'll need to know how to clean them up here as well.
5198                  */
5199                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5200                         free(io->scsiio.kern_data_ptr, M_CTL);
5201                 /* Hopefully the user has already set the status... */
5202                 ctl_done(io);
5203         } else {
5204                 /*
5205                  * XXX KDM now we need to continue data movement.  Some
5206                  * options:
5207                  * - call ctl_scsiio() again?  We don't do this for data
5208                  *   writes, because for those at least we know ahead of
5209                  *   time where the write will go and how long it is.  For
5210                  *   config writes, though, that information is largely
5211                  *   contained within the write itself, thus we need to
5212                  *   parse out the data again.
5213                  *
5214                  * - Call some other function once the data is in?
5215                  */
5216
5217                 /*
5218                  * XXX KDM call ctl_scsiio() again for now, and check flag
5219                  * bits to see whether we're allocated or not.
5220                  */
5221                 retval = ctl_scsiio(&io->scsiio);
5222         }
5223 bailout:
5224         return (retval);
5225 }
5226
5227 /*
5228  * This gets called by a backend driver when it is done with a
5229  * data_submit method.
5230  */
5231 void
5232 ctl_data_submit_done(union ctl_io *io)
5233 {
5234         /*
5235          * If the IO_CONT flag is set, we need to call the supplied
5236          * function to continue processing the I/O, instead of completing
5237          * the I/O just yet.
5238          *
5239          * If there is an error, though, we don't want to keep processing.
5240          * Instead, just send status back to the initiator.
5241          */
5242         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5243             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5244             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5245              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5246                 io->scsiio.io_cont(io);
5247                 return;
5248         }
5249         ctl_done(io);
5250 }
5251
5252 /*
5253  * This gets called by a backend driver when it is done with a
5254  * configuration write.
5255  */
5256 void
5257 ctl_config_write_done(union ctl_io *io)
5258 {
5259         /*
5260          * If the IO_CONT flag is set, we need to call the supplied
5261          * function to continue processing the I/O, instead of completing
5262          * the I/O just yet.
5263          *
5264          * If there is an error, though, we don't want to keep processing.
5265          * Instead, just send status back to the initiator.
5266          */
5267         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5268          && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5269           || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5270                 io->scsiio.io_cont(io);
5271                 return;
5272         }
5273         /*
5274          * Since a configuration write can be done for commands that actually
5275          * have data allocated, like write buffer, and commands that have
5276          * no data, like start/stop unit, we need to check here.
5277          */
5278         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5279                 free(io->scsiio.kern_data_ptr, M_CTL);
5280         ctl_done(io);
5281 }
5282
5283 /*
5284  * SCSI release command.
5285  */
5286 int
5287 ctl_scsi_release(struct ctl_scsiio *ctsio)
5288 {
5289         int length, longid, thirdparty_id, resv_id;
5290         struct ctl_softc *ctl_softc;
5291         struct ctl_lun *lun;
5292
5293         length = 0;
5294         resv_id = 0;
5295
5296         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5297
5298         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5299         ctl_softc = control_softc;
5300
5301         switch (ctsio->cdb[0]) {
5302         case RELEASE_10: {
5303                 struct scsi_release_10 *cdb;
5304
5305                 cdb = (struct scsi_release_10 *)ctsio->cdb;
5306
5307                 if (cdb->byte2 & SR10_LONGID)
5308                         longid = 1;
5309                 else
5310                         thirdparty_id = cdb->thirdparty_id;
5311
5312                 resv_id = cdb->resv_id;
5313                 length = scsi_2btoul(cdb->length);
5314                 break;
5315         }
5316         }
5317
5318
5319         /*
5320          * XXX KDM right now, we only support LUN reservation.  We don't
5321          * support 3rd party reservations, or extent reservations, which
5322          * might actually need the parameter list.  If we've gotten this
5323          * far, we've got a LUN reservation.  Anything else got kicked out
5324          * above.  So, according to SPC, ignore the length.
5325          */
5326         length = 0;
5327
5328         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5329          && (length > 0)) {
5330                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5331                 ctsio->kern_data_len = length;
5332                 ctsio->kern_total_len = length;
5333                 ctsio->kern_data_resid = 0;
5334                 ctsio->kern_rel_offset = 0;
5335                 ctsio->kern_sg_entries = 0;
5336                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5337                 ctsio->be_move_done = ctl_config_move_done;
5338                 ctl_datamove((union ctl_io *)ctsio);
5339
5340                 return (CTL_RETVAL_COMPLETE);
5341         }
5342
5343         if (length > 0)
5344                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5345
5346         mtx_lock(&lun->lun_lock);
5347
5348         /*
5349          * According to SPC, it is not an error for an intiator to attempt
5350          * to release a reservation on a LUN that isn't reserved, or that
5351          * is reserved by another initiator.  The reservation can only be
5352          * released, though, by the initiator who made it or by one of
5353          * several reset type events.
5354          */
5355         if (lun->flags & CTL_LUN_RESERVED) {
5356                 if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5357                  && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5358                  && (ctsio->io_hdr.nexus.targ_target.id ==
5359                      lun->rsv_nexus.targ_target.id)) {
5360                         lun->flags &= ~CTL_LUN_RESERVED;
5361                 }
5362         }
5363
5364         mtx_unlock(&lun->lun_lock);
5365
5366         ctsio->scsi_status = SCSI_STATUS_OK;
5367         ctsio->io_hdr.status = CTL_SUCCESS;
5368
5369         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5370                 free(ctsio->kern_data_ptr, M_CTL);
5371                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5372         }
5373
5374         ctl_done((union ctl_io *)ctsio);
5375         return (CTL_RETVAL_COMPLETE);
5376 }
5377
5378 int
5379 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5380 {
5381         int extent, thirdparty, longid;
5382         int resv_id, length;
5383         uint64_t thirdparty_id;
5384         struct ctl_softc *ctl_softc;
5385         struct ctl_lun *lun;
5386
5387         extent = 0;
5388         thirdparty = 0;
5389         longid = 0;
5390         resv_id = 0;
5391         length = 0;
5392         thirdparty_id = 0;
5393
5394         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5395
5396         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5397         ctl_softc = control_softc;
5398
5399         switch (ctsio->cdb[0]) {
5400         case RESERVE_10: {
5401                 struct scsi_reserve_10 *cdb;
5402
5403                 cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5404
5405                 if (cdb->byte2 & SR10_LONGID)
5406                         longid = 1;
5407                 else
5408                         thirdparty_id = cdb->thirdparty_id;
5409
5410                 resv_id = cdb->resv_id;
5411                 length = scsi_2btoul(cdb->length);
5412                 break;
5413         }
5414         }
5415
5416         /*
5417          * XXX KDM right now, we only support LUN reservation.  We don't
5418          * support 3rd party reservations, or extent reservations, which
5419          * might actually need the parameter list.  If we've gotten this
5420          * far, we've got a LUN reservation.  Anything else got kicked out
5421          * above.  So, according to SPC, ignore the length.
5422          */
5423         length = 0;
5424
5425         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5426          && (length > 0)) {
5427                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5428                 ctsio->kern_data_len = length;
5429                 ctsio->kern_total_len = length;
5430                 ctsio->kern_data_resid = 0;
5431                 ctsio->kern_rel_offset = 0;
5432                 ctsio->kern_sg_entries = 0;
5433                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5434                 ctsio->be_move_done = ctl_config_move_done;
5435                 ctl_datamove((union ctl_io *)ctsio);
5436
5437                 return (CTL_RETVAL_COMPLETE);
5438         }
5439
5440         if (length > 0)
5441                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5442
5443         mtx_lock(&lun->lun_lock);
5444         if (lun->flags & CTL_LUN_RESERVED) {
5445                 if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5446                  || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5447                  || (ctsio->io_hdr.nexus.targ_target.id !=
5448                      lun->rsv_nexus.targ_target.id)) {
5449                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5450                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
5451                         goto bailout;
5452                 }
5453         }
5454
5455         lun->flags |= CTL_LUN_RESERVED;
5456         lun->rsv_nexus = ctsio->io_hdr.nexus;
5457
5458         ctsio->scsi_status = SCSI_STATUS_OK;
5459         ctsio->io_hdr.status = CTL_SUCCESS;
5460
5461 bailout:
5462         mtx_unlock(&lun->lun_lock);
5463
5464         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5465                 free(ctsio->kern_data_ptr, M_CTL);
5466                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5467         }
5468
5469         ctl_done((union ctl_io *)ctsio);
5470         return (CTL_RETVAL_COMPLETE);
5471 }
5472
5473 int
5474 ctl_start_stop(struct ctl_scsiio *ctsio)
5475 {
5476         struct scsi_start_stop_unit *cdb;
5477         struct ctl_lun *lun;
5478         struct ctl_softc *ctl_softc;
5479         int retval;
5480
5481         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5482
5483         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5484         ctl_softc = control_softc;
5485         retval = 0;
5486
5487         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5488
5489         /*
5490          * XXX KDM
5491          * We don't support the immediate bit on a stop unit.  In order to
5492          * do that, we would need to code up a way to know that a stop is
5493          * pending, and hold off any new commands until it completes, one
5494          * way or another.  Then we could accept or reject those commands
5495          * depending on its status.  We would almost need to do the reverse
5496          * of what we do below for an immediate start -- return the copy of
5497          * the ctl_io to the FETD with status to send to the host (and to
5498          * free the copy!) and then free the original I/O once the stop
5499          * actually completes.  That way, the OOA queue mechanism can work
5500          * to block commands that shouldn't proceed.  Another alternative
5501          * would be to put the copy in the queue in place of the original,
5502          * and return the original back to the caller.  That could be
5503          * slightly safer..
5504          */
5505         if ((cdb->byte2 & SSS_IMMED)
5506          && ((cdb->how & SSS_START) == 0)) {
5507                 ctl_set_invalid_field(ctsio,
5508                                       /*sks_valid*/ 1,
5509                                       /*command*/ 1,
5510                                       /*field*/ 1,
5511                                       /*bit_valid*/ 1,
5512                                       /*bit*/ 0);
5513                 ctl_done((union ctl_io *)ctsio);
5514                 return (CTL_RETVAL_COMPLETE);
5515         }
5516
5517         if ((lun->flags & CTL_LUN_PR_RESERVED)
5518          && ((cdb->how & SSS_START)==0)) {
5519                 uint32_t residx;
5520
5521                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5522                 if (!lun->per_res[residx].registered
5523                  || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5524
5525                         ctl_set_reservation_conflict(ctsio);
5526                         ctl_done((union ctl_io *)ctsio);
5527                         return (CTL_RETVAL_COMPLETE);
5528                 }
5529         }
5530
5531         /*
5532          * If there is no backend on this device, we can't start or stop
5533          * it.  In theory we shouldn't get any start/stop commands in the
5534          * first place at this level if the LUN doesn't have a backend.
5535          * That should get stopped by the command decode code.
5536          */
5537         if (lun->backend == NULL) {
5538                 ctl_set_invalid_opcode(ctsio);
5539                 ctl_done((union ctl_io *)ctsio);
5540                 return (CTL_RETVAL_COMPLETE);
5541         }
5542
5543         /*
5544          * XXX KDM Copan-specific offline behavior.
5545          * Figure out a reasonable way to port this?
5546          */
5547 #ifdef NEEDTOPORT
5548         mtx_lock(&lun->lun_lock);
5549
5550         if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5551          && (lun->flags & CTL_LUN_OFFLINE)) {
5552                 /*
5553                  * If the LUN is offline, and the on/offline bit isn't set,
5554                  * reject the start or stop.  Otherwise, let it through.
5555                  */
5556                 mtx_unlock(&lun->lun_lock);
5557                 ctl_set_lun_not_ready(ctsio);
5558                 ctl_done((union ctl_io *)ctsio);
5559         } else {
5560                 mtx_unlock(&lun->lun_lock);
5561 #endif /* NEEDTOPORT */
5562                 /*
5563                  * This could be a start or a stop when we're online,
5564                  * or a stop/offline or start/online.  A start or stop when
5565                  * we're offline is covered in the case above.
5566                  */
5567                 /*
5568                  * In the non-immediate case, we send the request to
5569                  * the backend and return status to the user when
5570                  * it is done.
5571                  *
5572                  * In the immediate case, we allocate a new ctl_io
5573                  * to hold a copy of the request, and send that to
5574                  * the backend.  We then set good status on the
5575                  * user's request and return it immediately.
5576                  */
5577                 if (cdb->byte2 & SSS_IMMED) {
5578                         union ctl_io *new_io;
5579
5580                         new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5581                         if (new_io == NULL) {
5582                                 ctl_set_busy(ctsio);
5583                                 ctl_done((union ctl_io *)ctsio);
5584                         } else {
5585                                 ctl_copy_io((union ctl_io *)ctsio,
5586                                             new_io);
5587                                 retval = lun->backend->config_write(new_io);
5588                                 ctl_set_success(ctsio);
5589                                 ctl_done((union ctl_io *)ctsio);
5590                         }
5591                 } else {
5592                         retval = lun->backend->config_write(
5593                                 (union ctl_io *)ctsio);
5594                 }
5595 #ifdef NEEDTOPORT
5596         }
5597 #endif
5598         return (retval);
5599 }
5600
5601 /*
5602  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5603  * we don't really do anything with the LBA and length fields if the user
5604  * passes them in.  Instead we'll just flush out the cache for the entire
5605  * LUN.
5606  */
5607 int
5608 ctl_sync_cache(struct ctl_scsiio *ctsio)
5609 {
5610         struct ctl_lun *lun;
5611         struct ctl_softc *ctl_softc;
5612         uint64_t starting_lba;
5613         uint32_t block_count;
5614         int retval;
5615
5616         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5617
5618         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5619         ctl_softc = control_softc;
5620         retval = 0;
5621
5622         switch (ctsio->cdb[0]) {
5623         case SYNCHRONIZE_CACHE: {
5624                 struct scsi_sync_cache *cdb;
5625                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5626
5627                 starting_lba = scsi_4btoul(cdb->begin_lba);
5628                 block_count = scsi_2btoul(cdb->lb_count);
5629                 break;
5630         }
5631         case SYNCHRONIZE_CACHE_16: {
5632                 struct scsi_sync_cache_16 *cdb;
5633                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5634
5635                 starting_lba = scsi_8btou64(cdb->begin_lba);
5636                 block_count = scsi_4btoul(cdb->lb_count);
5637                 break;
5638         }
5639         default:
5640                 ctl_set_invalid_opcode(ctsio);
5641                 ctl_done((union ctl_io *)ctsio);
5642                 goto bailout;
5643                 break; /* NOTREACHED */
5644         }
5645
5646         /*
5647          * We check the LBA and length, but don't do anything with them.
5648          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5649          * get flushed.  This check will just help satisfy anyone who wants
5650          * to see an error for an out of range LBA.
5651          */
5652         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5653                 ctl_set_lba_out_of_range(ctsio);
5654                 ctl_done((union ctl_io *)ctsio);
5655                 goto bailout;
5656         }
5657
5658         /*
5659          * If this LUN has no backend, we can't flush the cache anyway.
5660          */
5661         if (lun->backend == NULL) {
5662                 ctl_set_invalid_opcode(ctsio);
5663                 ctl_done((union ctl_io *)ctsio);
5664                 goto bailout;
5665         }
5666
5667         /*
5668          * Check to see whether we're configured to send the SYNCHRONIZE
5669          * CACHE command directly to the back end.
5670          */
5671         mtx_lock(&lun->lun_lock);
5672         if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5673          && (++(lun->sync_count) >= lun->sync_interval)) {
5674                 lun->sync_count = 0;
5675                 mtx_unlock(&lun->lun_lock);
5676                 retval = lun->backend->config_write((union ctl_io *)ctsio);
5677         } else {
5678                 mtx_unlock(&lun->lun_lock);
5679                 ctl_set_success(ctsio);
5680                 ctl_done((union ctl_io *)ctsio);
5681         }
5682
5683 bailout:
5684
5685         return (retval);
5686 }
5687
5688 int
5689 ctl_format(struct ctl_scsiio *ctsio)
5690 {
5691         struct scsi_format *cdb;
5692         struct ctl_lun *lun;
5693         struct ctl_softc *ctl_softc;
5694         int length, defect_list_len;
5695
5696         CTL_DEBUG_PRINT(("ctl_format\n"));
5697
5698         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5699         ctl_softc = control_softc;
5700
5701         cdb = (struct scsi_format *)ctsio->cdb;
5702
5703         length = 0;
5704         if (cdb->byte2 & SF_FMTDATA) {
5705                 if (cdb->byte2 & SF_LONGLIST)
5706                         length = sizeof(struct scsi_format_header_long);
5707                 else
5708                         length = sizeof(struct scsi_format_header_short);
5709         }
5710
5711         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5712          && (length > 0)) {
5713                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5714                 ctsio->kern_data_len = length;
5715                 ctsio->kern_total_len = length;
5716                 ctsio->kern_data_resid = 0;
5717                 ctsio->kern_rel_offset = 0;
5718                 ctsio->kern_sg_entries = 0;
5719                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5720                 ctsio->be_move_done = ctl_config_move_done;
5721                 ctl_datamove((union ctl_io *)ctsio);
5722
5723                 return (CTL_RETVAL_COMPLETE);
5724         }
5725
5726         defect_list_len = 0;
5727
5728         if (cdb->byte2 & SF_FMTDATA) {
5729                 if (cdb->byte2 & SF_LONGLIST) {
5730                         struct scsi_format_header_long *header;
5731
5732                         header = (struct scsi_format_header_long *)
5733                                 ctsio->kern_data_ptr;
5734
5735                         defect_list_len = scsi_4btoul(header->defect_list_len);
5736                         if (defect_list_len != 0) {
5737                                 ctl_set_invalid_field(ctsio,
5738                                                       /*sks_valid*/ 1,
5739                                                       /*command*/ 0,
5740                                                       /*field*/ 2,
5741                                                       /*bit_valid*/ 0,
5742                                                       /*bit*/ 0);
5743                                 goto bailout;
5744                         }
5745                 } else {
5746                         struct scsi_format_header_short *header;
5747
5748                         header = (struct scsi_format_header_short *)
5749                                 ctsio->kern_data_ptr;
5750
5751                         defect_list_len = scsi_2btoul(header->defect_list_len);
5752                         if (defect_list_len != 0) {
5753                                 ctl_set_invalid_field(ctsio,
5754                                                       /*sks_valid*/ 1,
5755                                                       /*command*/ 0,
5756                                                       /*field*/ 2,
5757                                                       /*bit_valid*/ 0,
5758                                                       /*bit*/ 0);
5759                                 goto bailout;
5760                         }
5761                 }
5762         }
5763
5764         /*
5765          * The format command will clear out the "Medium format corrupted"
5766          * status if set by the configuration code.  That status is really
5767          * just a way to notify the host that we have lost the media, and
5768          * get them to issue a command that will basically make them think
5769          * they're blowing away the media.
5770          */
5771         mtx_lock(&lun->lun_lock);
5772         lun->flags &= ~CTL_LUN_INOPERABLE;
5773         mtx_unlock(&lun->lun_lock);
5774
5775         ctsio->scsi_status = SCSI_STATUS_OK;
5776         ctsio->io_hdr.status = CTL_SUCCESS;
5777 bailout:
5778
5779         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5780                 free(ctsio->kern_data_ptr, M_CTL);
5781                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5782         }
5783
5784         ctl_done((union ctl_io *)ctsio);
5785         return (CTL_RETVAL_COMPLETE);
5786 }
5787
5788 int
5789 ctl_read_buffer(struct ctl_scsiio *ctsio)
5790 {
5791         struct scsi_read_buffer *cdb;
5792         struct ctl_lun *lun;
5793         int buffer_offset, len;
5794         static uint8_t descr[4];
5795         static uint8_t echo_descr[4] = { 0 };
5796
5797         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5798
5799         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5800         cdb = (struct scsi_read_buffer *)ctsio->cdb;
5801
5802         if (lun->flags & CTL_LUN_PR_RESERVED) {
5803                 uint32_t residx;
5804
5805                 /*
5806                  * XXX KDM need a lock here.
5807                  */
5808                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5809                 if ((lun->res_type == SPR_TYPE_EX_AC
5810                   && residx != lun->pr_res_idx)
5811                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
5812                    || lun->res_type == SPR_TYPE_EX_AC_AR)
5813                   && !lun->per_res[residx].registered)) {
5814                         ctl_set_reservation_conflict(ctsio);
5815                         ctl_done((union ctl_io *)ctsio);
5816                         return (CTL_RETVAL_COMPLETE);
5817                 }
5818         }
5819
5820         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5821             (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5822             (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5823                 ctl_set_invalid_field(ctsio,
5824                                       /*sks_valid*/ 1,
5825                                       /*command*/ 1,
5826                                       /*field*/ 1,
5827                                       /*bit_valid*/ 1,
5828                                       /*bit*/ 4);
5829                 ctl_done((union ctl_io *)ctsio);
5830                 return (CTL_RETVAL_COMPLETE);
5831         }
5832
5833         len = scsi_3btoul(cdb->length);
5834         buffer_offset = scsi_3btoul(cdb->offset);
5835
5836         if (buffer_offset + len > sizeof(lun->write_buffer)) {
5837                 ctl_set_invalid_field(ctsio,
5838                                       /*sks_valid*/ 1,
5839                                       /*command*/ 1,
5840                                       /*field*/ 6,
5841                                       /*bit_valid*/ 0,
5842                                       /*bit*/ 0);
5843                 ctl_done((union ctl_io *)ctsio);
5844                 return (CTL_RETVAL_COMPLETE);
5845         }
5846
5847         if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5848                 descr[0] = 0;
5849                 scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5850                 ctsio->kern_data_ptr = descr;
5851                 len = min(len, sizeof(descr));
5852         } else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5853                 ctsio->kern_data_ptr = echo_descr;
5854                 len = min(len, sizeof(echo_descr));
5855         } else
5856                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5857         ctsio->kern_data_len = len;
5858         ctsio->kern_total_len = len;
5859         ctsio->kern_data_resid = 0;
5860         ctsio->kern_rel_offset = 0;
5861         ctsio->kern_sg_entries = 0;
5862         ctsio->be_move_done = ctl_config_move_done;
5863         ctl_datamove((union ctl_io *)ctsio);
5864
5865         return (CTL_RETVAL_COMPLETE);
5866 }
5867
5868 int
5869 ctl_write_buffer(struct ctl_scsiio *ctsio)
5870 {
5871         struct scsi_write_buffer *cdb;
5872         struct ctl_lun *lun;
5873         int buffer_offset, len;
5874
5875         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5876
5877         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5878         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5879
5880         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5881                 ctl_set_invalid_field(ctsio,
5882                                       /*sks_valid*/ 1,
5883                                       /*command*/ 1,
5884                                       /*field*/ 1,
5885                                       /*bit_valid*/ 1,
5886                                       /*bit*/ 4);
5887                 ctl_done((union ctl_io *)ctsio);
5888                 return (CTL_RETVAL_COMPLETE);
5889         }
5890
5891         len = scsi_3btoul(cdb->length);
5892         buffer_offset = scsi_3btoul(cdb->offset);
5893
5894         if (buffer_offset + len > sizeof(lun->write_buffer)) {
5895                 ctl_set_invalid_field(ctsio,
5896                                       /*sks_valid*/ 1,
5897                                       /*command*/ 1,
5898                                       /*field*/ 6,
5899                                       /*bit_valid*/ 0,
5900                                       /*bit*/ 0);
5901                 ctl_done((union ctl_io *)ctsio);
5902                 return (CTL_RETVAL_COMPLETE);
5903         }
5904
5905         /*
5906          * If we've got a kernel request that hasn't been malloced yet,
5907          * malloc it and tell the caller the data buffer is here.
5908          */
5909         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5910                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5911                 ctsio->kern_data_len = len;
5912                 ctsio->kern_total_len = len;
5913                 ctsio->kern_data_resid = 0;
5914                 ctsio->kern_rel_offset = 0;
5915                 ctsio->kern_sg_entries = 0;
5916                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5917                 ctsio->be_move_done = ctl_config_move_done;
5918                 ctl_datamove((union ctl_io *)ctsio);
5919
5920                 return (CTL_RETVAL_COMPLETE);
5921         }
5922
5923         ctl_done((union ctl_io *)ctsio);
5924
5925         return (CTL_RETVAL_COMPLETE);
5926 }
5927
5928 int
5929 ctl_write_same(struct ctl_scsiio *ctsio)
5930 {
5931         struct ctl_lun *lun;
5932         struct ctl_lba_len_flags *lbalen;
5933         uint64_t lba;
5934         uint32_t num_blocks;
5935         int len, retval;
5936         uint8_t byte2;
5937
5938         retval = CTL_RETVAL_COMPLETE;
5939
5940         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5941
5942         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5943
5944         switch (ctsio->cdb[0]) {
5945         case WRITE_SAME_10: {
5946                 struct scsi_write_same_10 *cdb;
5947
5948                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5949
5950                 lba = scsi_4btoul(cdb->addr);
5951                 num_blocks = scsi_2btoul(cdb->length);
5952                 byte2 = cdb->byte2;
5953                 break;
5954         }
5955         case WRITE_SAME_16: {
5956                 struct scsi_write_same_16 *cdb;
5957
5958                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5959
5960                 lba = scsi_8btou64(cdb->addr);
5961                 num_blocks = scsi_4btoul(cdb->length);
5962                 byte2 = cdb->byte2;
5963                 break;
5964         }
5965         default:
5966                 /*
5967                  * We got a command we don't support.  This shouldn't
5968                  * happen, commands should be filtered out above us.
5969                  */
5970                 ctl_set_invalid_opcode(ctsio);
5971                 ctl_done((union ctl_io *)ctsio);
5972
5973                 return (CTL_RETVAL_COMPLETE);
5974                 break; /* NOTREACHED */
5975         }
5976
5977         /*
5978          * The first check is to make sure we're in bounds, the second
5979          * check is to catch wrap-around problems.  If the lba + num blocks
5980          * is less than the lba, then we've wrapped around and the block
5981          * range is invalid anyway.
5982          */
5983         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5984          || ((lba + num_blocks) < lba)) {
5985                 ctl_set_lba_out_of_range(ctsio);
5986                 ctl_done((union ctl_io *)ctsio);
5987                 return (CTL_RETVAL_COMPLETE);
5988         }
5989
5990         /* Zero number of blocks means "to the last logical block" */
5991         if (num_blocks == 0) {
5992                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5993                         ctl_set_invalid_field(ctsio,
5994                                               /*sks_valid*/ 0,
5995                                               /*command*/ 1,
5996                                               /*field*/ 0,
5997                                               /*bit_valid*/ 0,
5998                                               /*bit*/ 0);
5999                         ctl_done((union ctl_io *)ctsio);
6000                         return (CTL_RETVAL_COMPLETE);
6001                 }
6002                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
6003         }
6004
6005         len = lun->be_lun->blocksize;
6006
6007         /*
6008          * If we've got a kernel request that hasn't been malloced yet,
6009          * malloc it and tell the caller the data buffer is here.
6010          */
6011         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6012                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6013                 ctsio->kern_data_len = len;
6014                 ctsio->kern_total_len = len;
6015                 ctsio->kern_data_resid = 0;
6016                 ctsio->kern_rel_offset = 0;
6017                 ctsio->kern_sg_entries = 0;
6018                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6019                 ctsio->be_move_done = ctl_config_move_done;
6020                 ctl_datamove((union ctl_io *)ctsio);
6021
6022                 return (CTL_RETVAL_COMPLETE);
6023         }
6024
6025         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6026         lbalen->lba = lba;
6027         lbalen->len = num_blocks;
6028         lbalen->flags = byte2;
6029         retval = lun->backend->config_write((union ctl_io *)ctsio);
6030
6031         return (retval);
6032 }
6033
6034 int
6035 ctl_unmap(struct ctl_scsiio *ctsio)
6036 {
6037         struct ctl_lun *lun;
6038         struct scsi_unmap *cdb;
6039         struct ctl_ptr_len_flags *ptrlen;
6040         struct scsi_unmap_header *hdr;
6041         struct scsi_unmap_desc *buf, *end;
6042         uint64_t lba;
6043         uint32_t num_blocks;
6044         int len, retval;
6045         uint8_t byte2;
6046
6047         retval = CTL_RETVAL_COMPLETE;
6048
6049         CTL_DEBUG_PRINT(("ctl_unmap\n"));
6050
6051         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6052         cdb = (struct scsi_unmap *)ctsio->cdb;
6053
6054         len = scsi_2btoul(cdb->length);
6055         byte2 = cdb->byte2;
6056
6057         /*
6058          * If we've got a kernel request that hasn't been malloced yet,
6059          * malloc it and tell the caller the data buffer is here.
6060          */
6061         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6062                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6063                 ctsio->kern_data_len = len;
6064                 ctsio->kern_total_len = len;
6065                 ctsio->kern_data_resid = 0;
6066                 ctsio->kern_rel_offset = 0;
6067                 ctsio->kern_sg_entries = 0;
6068                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6069                 ctsio->be_move_done = ctl_config_move_done;
6070                 ctl_datamove((union ctl_io *)ctsio);
6071
6072                 return (CTL_RETVAL_COMPLETE);
6073         }
6074
6075         len = ctsio->kern_total_len - ctsio->kern_data_resid;
6076         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6077         if (len < sizeof (*hdr) ||
6078             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6079             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6080             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6081                 ctl_set_invalid_field(ctsio,
6082                                       /*sks_valid*/ 0,
6083                                       /*command*/ 0,
6084                                       /*field*/ 0,
6085                                       /*bit_valid*/ 0,
6086                                       /*bit*/ 0);
6087                 ctl_done((union ctl_io *)ctsio);
6088                 return (CTL_RETVAL_COMPLETE);
6089         }
6090         len = scsi_2btoul(hdr->desc_length);
6091         buf = (struct scsi_unmap_desc *)(hdr + 1);
6092         end = buf + len / sizeof(*buf);
6093
6094         ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6095         ptrlen->ptr = (void *)buf;
6096         ptrlen->len = len;
6097         ptrlen->flags = byte2;
6098
6099         for (; buf < end; buf++) {
6100                 lba = scsi_8btou64(buf->lba);
6101                 num_blocks = scsi_4btoul(buf->length);
6102                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6103                  || ((lba + num_blocks) < lba)) {
6104                         ctl_set_lba_out_of_range(ctsio);
6105                         ctl_done((union ctl_io *)ctsio);
6106                         return (CTL_RETVAL_COMPLETE);
6107                 }
6108         }
6109
6110         retval = lun->backend->config_write((union ctl_io *)ctsio);
6111
6112         return (retval);
6113 }
6114
6115 /*
6116  * Note that this function currently doesn't actually do anything inside
6117  * CTL to enforce things if the DQue bit is turned on.
6118  *
6119  * Also note that this function can't be used in the default case, because
6120  * the DQue bit isn't set in the changeable mask for the control mode page
6121  * anyway.  This is just here as an example for how to implement a page
6122  * handler, and a placeholder in case we want to allow the user to turn
6123  * tagged queueing on and off.
6124  *
6125  * The D_SENSE bit handling is functional, however, and will turn
6126  * descriptor sense on and off for a given LUN.
6127  */
6128 int
6129 ctl_control_page_handler(struct ctl_scsiio *ctsio,
6130                          struct ctl_page_index *page_index, uint8_t *page_ptr)
6131 {
6132         struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6133         struct ctl_lun *lun;
6134         struct ctl_softc *softc;
6135         int set_ua;
6136         uint32_t initidx;
6137
6138         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6139         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6140         set_ua = 0;
6141
6142         user_cp = (struct scsi_control_page *)page_ptr;
6143         current_cp = (struct scsi_control_page *)
6144                 (page_index->page_data + (page_index->page_len *
6145                 CTL_PAGE_CURRENT));
6146         saved_cp = (struct scsi_control_page *)
6147                 (page_index->page_data + (page_index->page_len *
6148                 CTL_PAGE_SAVED));
6149
6150         softc = control_softc;
6151
6152         mtx_lock(&lun->lun_lock);
6153         if (((current_cp->rlec & SCP_DSENSE) == 0)
6154          && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6155                 /*
6156                  * Descriptor sense is currently turned off and the user
6157                  * wants to turn it on.
6158                  */
6159                 current_cp->rlec |= SCP_DSENSE;
6160                 saved_cp->rlec |= SCP_DSENSE;
6161                 lun->flags |= CTL_LUN_SENSE_DESC;
6162                 set_ua = 1;
6163         } else if (((current_cp->rlec & SCP_DSENSE) != 0)
6164                 && ((user_cp->rlec & SCP_DSENSE) == 0)) {
6165                 /*
6166                  * Descriptor sense is currently turned on, and the user
6167                  * wants to turn it off.
6168                  */
6169                 current_cp->rlec &= ~SCP_DSENSE;
6170                 saved_cp->rlec &= ~SCP_DSENSE;
6171                 lun->flags &= ~CTL_LUN_SENSE_DESC;
6172                 set_ua = 1;
6173         }
6174         if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
6175                 if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6176 #ifdef NEEDTOPORT
6177                         csevent_log(CSC_CTL | CSC_SHELF_SW |
6178                                     CTL_UNTAG_TO_UNTAG,
6179                                     csevent_LogType_Trace,
6180                                     csevent_Severity_Information,
6181                                     csevent_AlertLevel_Green,
6182                                     csevent_FRU_Firmware,
6183                                     csevent_FRU_Unknown,
6184                                     "Received untagged to untagged transition");
6185 #endif /* NEEDTOPORT */
6186                 } else {
6187 #ifdef NEEDTOPORT
6188                         csevent_log(CSC_CTL | CSC_SHELF_SW |
6189                                     CTL_UNTAG_TO_TAG,
6190                                     csevent_LogType_ConfigChange,
6191                                     csevent_Severity_Information,
6192                                     csevent_AlertLevel_Green,
6193                                     csevent_FRU_Firmware,
6194                                     csevent_FRU_Unknown,
6195                                     "Received untagged to tagged "
6196                                     "queueing transition");
6197 #endif /* NEEDTOPORT */
6198
6199                         current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6200                         saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6201                         set_ua = 1;
6202                 }
6203         } else {
6204                 if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6205 #ifdef NEEDTOPORT
6206                         csevent_log(CSC_CTL | CSC_SHELF_SW |
6207                                     CTL_TAG_TO_UNTAG,
6208                                     csevent_LogType_ConfigChange,
6209                                     csevent_Severity_Warning,
6210                                     csevent_AlertLevel_Yellow,
6211                                     csevent_FRU_Firmware,
6212                                     csevent_FRU_Unknown,
6213                                     "Received tagged queueing to untagged "
6214                                     "transition");
6215 #endif /* NEEDTOPORT */
6216
6217                         current_cp->queue_flags |= SCP_QUEUE_DQUE;
6218                         saved_cp->queue_flags |= SCP_QUEUE_DQUE;
6219                         set_ua = 1;
6220                 } else {
6221 #ifdef NEEDTOPORT
6222                         csevent_log(CSC_CTL | CSC_SHELF_SW |
6223                                     CTL_TAG_TO_TAG,
6224                                     csevent_LogType_Trace,
6225                                     csevent_Severity_Information,
6226                                     csevent_AlertLevel_Green,
6227                                     csevent_FRU_Firmware,
6228                                     csevent_FRU_Unknown,
6229                                     "Received tagged queueing to tagged "
6230                                     "queueing transition");
6231 #endif /* NEEDTOPORT */
6232                 }
6233         }
6234         if (set_ua != 0) {
6235                 int i;
6236                 /*
6237                  * Let other initiators know that the mode
6238                  * parameters for this LUN have changed.
6239                  */
6240                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6241                         if (i == initidx)
6242                                 continue;
6243
6244                         lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6245                 }
6246         }
6247         mtx_unlock(&lun->lun_lock);
6248
6249         return (0);
6250 }
6251
6252 int
6253 ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6254                      struct ctl_page_index *page_index, uint8_t *page_ptr)
6255 {
6256         return (0);
6257 }
6258
6259 int
6260 ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6261                            struct ctl_page_index *page_index, int pc)
6262 {
6263         struct copan_power_subpage *page;
6264
6265         page = (struct copan_power_subpage *)page_index->page_data +
6266                 (page_index->page_len * pc);
6267
6268         switch (pc) {
6269         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6270                 /*
6271                  * We don't update the changable bits for this page.
6272                  */
6273                 break;
6274         case SMS_PAGE_CTRL_CURRENT >> 6:
6275         case SMS_PAGE_CTRL_DEFAULT >> 6:
6276         case SMS_PAGE_CTRL_SAVED >> 6:
6277 #ifdef NEEDTOPORT
6278                 ctl_update_power_subpage(page);
6279 #endif
6280                 break;
6281         default:
6282 #ifdef NEEDTOPORT
6283                 EPRINT(0, "Invalid PC %d!!", pc);
6284 #endif
6285                 break;
6286         }
6287         return (0);
6288 }
6289
6290
6291 int
6292 ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6293                    struct ctl_page_index *page_index, uint8_t *page_ptr)
6294 {
6295         struct copan_aps_subpage *user_sp;
6296         struct copan_aps_subpage *current_sp;
6297         union ctl_modepage_info *modepage_info;
6298         struct ctl_softc *softc;
6299         struct ctl_lun *lun;
6300         int retval;
6301
6302         retval = CTL_RETVAL_COMPLETE;
6303         current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6304                      (page_index->page_len * CTL_PAGE_CURRENT));
6305         softc = control_softc;
6306         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6307
6308         user_sp = (struct copan_aps_subpage *)page_ptr;
6309
6310         modepage_info = (union ctl_modepage_info *)
6311                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6312
6313         modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6314         modepage_info->header.subpage = page_index->subpage;
6315         modepage_info->aps.lock_active = user_sp->lock_active;
6316
6317         mtx_lock(&softc->ctl_lock);
6318
6319         /*
6320          * If there is a request to lock the LUN and another LUN is locked
6321          * this is an error. If the requested LUN is already locked ignore
6322          * the request. If no LUN is locked attempt to lock it.
6323          * if there is a request to unlock the LUN and the LUN is currently
6324          * locked attempt to unlock it. Otherwise ignore the request. i.e.
6325          * if another LUN is locked or no LUN is locked.
6326          */
6327         if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6328                 if (softc->aps_locked_lun == lun->lun) {
6329                         /*
6330                          * This LUN is already locked, so we're done.
6331                          */
6332                         retval = CTL_RETVAL_COMPLETE;
6333                 } else if (softc->aps_locked_lun == 0) {
6334                         /*
6335                          * No one has the lock, pass the request to the
6336                          * backend.
6337                          */
6338                         retval = lun->backend->config_write(
6339                                 (union ctl_io *)ctsio);
6340                 } else {
6341                         /*
6342                          * Someone else has the lock, throw out the request.
6343                          */
6344                         ctl_set_already_locked(ctsio);
6345                         free(ctsio->kern_data_ptr, M_CTL);
6346                         ctl_done((union ctl_io *)ctsio);
6347
6348                         /*
6349                          * Set the return value so that ctl_do_mode_select()
6350                          * won't try to complete the command.  We already
6351                          * completed it here.
6352                          */
6353                         retval = CTL_RETVAL_ERROR;
6354                 }
6355         } else if (softc->aps_locked_lun == lun->lun) {
6356                 /*
6357                  * This LUN is locked, so pass the unlock request to the
6358                  * backend.
6359                  */
6360                 retval = lun->backend->config_write((union ctl_io *)ctsio);
6361         }
6362         mtx_unlock(&softc->ctl_lock);
6363
6364         return (retval);
6365 }
6366
6367 int
6368 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6369                                 struct ctl_page_index *page_index,
6370                                 uint8_t *page_ptr)
6371 {
6372         uint8_t *c;
6373         int i;
6374
6375         c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6376         ctl_time_io_secs =
6377                 (c[0] << 8) |
6378                 (c[1] << 0) |
6379                 0;
6380         CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6381         printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6382         printf("page data:");
6383         for (i=0; i<8; i++)
6384                 printf(" %.2x",page_ptr[i]);
6385         printf("\n");
6386         return (0);
6387 }
6388
6389 int
6390 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6391                                struct ctl_page_index *page_index,
6392                                int pc)
6393 {
6394         struct copan_debugconf_subpage *page;
6395
6396         page = (struct copan_debugconf_subpage *)page_index->page_data +
6397                 (page_index->page_len * pc);
6398
6399         switch (pc) {
6400         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6401         case SMS_PAGE_CTRL_DEFAULT >> 6:
6402         case SMS_PAGE_CTRL_SAVED >> 6:
6403                 /*
6404                  * We don't update the changable or default bits for this page.
6405                  */
6406                 break;
6407         case SMS_PAGE_CTRL_CURRENT >> 6:
6408                 page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6409                 page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6410                 break;
6411         default:
6412 #ifdef NEEDTOPORT
6413                 EPRINT(0, "Invalid PC %d!!", pc);
6414 #endif /* NEEDTOPORT */
6415                 break;
6416         }
6417         return (0);
6418 }
6419
6420
6421 static int
6422 ctl_do_mode_select(union ctl_io *io)
6423 {
6424         struct scsi_mode_page_header *page_header;
6425         struct ctl_page_index *page_index;
6426         struct ctl_scsiio *ctsio;
6427         int control_dev, page_len;
6428         int page_len_offset, page_len_size;
6429         union ctl_modepage_info *modepage_info;
6430         struct ctl_lun *lun;
6431         int *len_left, *len_used;
6432         int retval, i;
6433
6434         ctsio = &io->scsiio;
6435         page_index = NULL;
6436         page_len = 0;
6437         retval = CTL_RETVAL_COMPLETE;
6438
6439         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6440
6441         if (lun->be_lun->lun_type != T_DIRECT)
6442                 control_dev = 1;
6443         else
6444                 control_dev = 0;
6445
6446         modepage_info = (union ctl_modepage_info *)
6447                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6448         len_left = &modepage_info->header.len_left;
6449         len_used = &modepage_info->header.len_used;
6450
6451 do_next_page:
6452
6453         page_header = (struct scsi_mode_page_header *)
6454                 (ctsio->kern_data_ptr + *len_used);
6455
6456         if (*len_left == 0) {
6457                 free(ctsio->kern_data_ptr, M_CTL);
6458                 ctl_set_success(ctsio);
6459                 ctl_done((union ctl_io *)ctsio);
6460                 return (CTL_RETVAL_COMPLETE);
6461         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6462
6463                 free(ctsio->kern_data_ptr, M_CTL);
6464                 ctl_set_param_len_error(ctsio);
6465                 ctl_done((union ctl_io *)ctsio);
6466                 return (CTL_RETVAL_COMPLETE);
6467
6468         } else if ((page_header->page_code & SMPH_SPF)
6469                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6470
6471                 free(ctsio->kern_data_ptr, M_CTL);
6472                 ctl_set_param_len_error(ctsio);
6473                 ctl_done((union ctl_io *)ctsio);
6474                 return (CTL_RETVAL_COMPLETE);
6475         }
6476
6477
6478         /*
6479          * XXX KDM should we do something with the block descriptor?
6480          */
6481         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6482
6483                 if ((control_dev != 0)
6484                  && (lun->mode_pages.index[i].page_flags &
6485                      CTL_PAGE_FLAG_DISK_ONLY))
6486                         continue;
6487
6488                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6489                     (page_header->page_code & SMPH_PC_MASK))
6490                         continue;
6491
6492                 /*
6493                  * If neither page has a subpage code, then we've got a
6494                  * match.
6495                  */
6496                 if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6497                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6498                         page_index = &lun->mode_pages.index[i];
6499                         page_len = page_header->page_length;
6500                         break;
6501                 }
6502
6503                 /*
6504                  * If both pages have subpages, then the subpage numbers
6505                  * have to match.
6506                  */
6507                 if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6508                   && (page_header->page_code & SMPH_SPF)) {
6509                         struct scsi_mode_page_header_sp *sph;
6510
6511                         sph = (struct scsi_mode_page_header_sp *)page_header;
6512
6513                         if (lun->mode_pages.index[i].subpage ==
6514                             sph->subpage) {
6515                                 page_index = &lun->mode_pages.index[i];
6516                                 page_len = scsi_2btoul(sph->page_length);
6517                                 break;
6518                         }
6519                 }
6520         }
6521
6522         /*
6523          * If we couldn't find the page, or if we don't have a mode select
6524          * handler for it, send back an error to the user.
6525          */
6526         if ((page_index == NULL)
6527          || (page_index->select_handler == NULL)) {
6528                 ctl_set_invalid_field(ctsio,
6529                                       /*sks_valid*/ 1,
6530                                       /*command*/ 0,
6531                                       /*field*/ *len_used,
6532                                       /*bit_valid*/ 0,
6533                                       /*bit*/ 0);
6534                 free(ctsio->kern_data_ptr, M_CTL);
6535                 ctl_done((union ctl_io *)ctsio);
6536                 return (CTL_RETVAL_COMPLETE);
6537         }
6538
6539         if (page_index->page_code & SMPH_SPF) {
6540                 page_len_offset = 2;
6541                 page_len_size = 2;
6542         } else {
6543                 page_len_size = 1;
6544                 page_len_offset = 1;
6545         }
6546
6547         /*
6548          * If the length the initiator gives us isn't the one we specify in
6549          * the mode page header, or if they didn't specify enough data in
6550          * the CDB to avoid truncating this page, kick out the request.
6551          */
6552         if ((page_len != (page_index->page_len - page_len_offset -
6553                           page_len_size))
6554          || (*len_left < page_index->page_len)) {
6555
6556
6557                 ctl_set_invalid_field(ctsio,
6558                                       /*sks_valid*/ 1,
6559                                       /*command*/ 0,
6560                                       /*field*/ *len_used + page_len_offset,
6561                                       /*bit_valid*/ 0,
6562                                       /*bit*/ 0);
6563                 free(ctsio->kern_data_ptr, M_CTL);
6564                 ctl_done((union ctl_io *)ctsio);
6565                 return (CTL_RETVAL_COMPLETE);
6566         }
6567
6568         /*
6569          * Run through the mode page, checking to make sure that the bits
6570          * the user changed are actually legal for him to change.
6571          */
6572         for (i = 0; i < page_index->page_len; i++) {
6573                 uint8_t *user_byte, *change_mask, *current_byte;
6574                 int bad_bit;
6575                 int j;
6576
6577                 user_byte = (uint8_t *)page_header + i;
6578                 change_mask = page_index->page_data +
6579                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6580                 current_byte = page_index->page_data +
6581                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6582
6583                 /*
6584                  * Check to see whether the user set any bits in this byte
6585                  * that he is not allowed to set.
6586                  */
6587                 if ((*user_byte & ~(*change_mask)) ==
6588                     (*current_byte & ~(*change_mask)))
6589                         continue;
6590
6591                 /*
6592                  * Go through bit by bit to determine which one is illegal.
6593                  */
6594                 bad_bit = 0;
6595                 for (j = 7; j >= 0; j--) {
6596                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6597                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6598                                 bad_bit = i;
6599                                 break;
6600                         }
6601                 }
6602                 ctl_set_invalid_field(ctsio,
6603                                       /*sks_valid*/ 1,
6604                                       /*command*/ 0,
6605                                       /*field*/ *len_used + i,
6606                                       /*bit_valid*/ 1,
6607                                       /*bit*/ bad_bit);
6608                 free(ctsio->kern_data_ptr, M_CTL);
6609                 ctl_done((union ctl_io *)ctsio);
6610                 return (CTL_RETVAL_COMPLETE);
6611         }
6612
6613         /*
6614          * Decrement these before we call the page handler, since we may
6615          * end up getting called back one way or another before the handler
6616          * returns to this context.
6617          */
6618         *len_left -= page_index->page_len;
6619         *len_used += page_index->page_len;
6620
6621         retval = page_index->select_handler(ctsio, page_index,
6622                                             (uint8_t *)page_header);
6623
6624         /*
6625          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6626          * wait until this queued command completes to finish processing
6627          * the mode page.  If it returns anything other than
6628          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6629          * already set the sense information, freed the data pointer, and
6630          * completed the io for us.
6631          */
6632         if (retval != CTL_RETVAL_COMPLETE)
6633                 goto bailout_no_done;
6634
6635         /*
6636          * If the initiator sent us more than one page, parse the next one.
6637          */
6638         if (*len_left > 0)
6639                 goto do_next_page;
6640
6641         ctl_set_success(ctsio);
6642         free(ctsio->kern_data_ptr, M_CTL);
6643         ctl_done((union ctl_io *)ctsio);
6644
6645 bailout_no_done:
6646
6647         return (CTL_RETVAL_COMPLETE);
6648
6649 }
6650
6651 int
6652 ctl_mode_select(struct ctl_scsiio *ctsio)
6653 {
6654         int param_len, pf, sp;
6655         int header_size, bd_len;
6656         int len_left, len_used;
6657         struct ctl_page_index *page_index;
6658         struct ctl_lun *lun;
6659         int control_dev, page_len;
6660         union ctl_modepage_info *modepage_info;
6661         int retval;
6662
6663         pf = 0;
6664         sp = 0;
6665         page_len = 0;
6666         len_used = 0;
6667         len_left = 0;
6668         retval = 0;
6669         bd_len = 0;
6670         page_index = NULL;
6671
6672         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6673
6674         if (lun->be_lun->lun_type != T_DIRECT)
6675                 control_dev = 1;
6676         else
6677                 control_dev = 0;
6678
6679         switch (ctsio->cdb[0]) {
6680         case MODE_SELECT_6: {
6681                 struct scsi_mode_select_6 *cdb;
6682
6683                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6684
6685                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6686                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6687
6688                 param_len = cdb->length;
6689                 header_size = sizeof(struct scsi_mode_header_6);
6690                 break;
6691         }
6692         case MODE_SELECT_10: {
6693                 struct scsi_mode_select_10 *cdb;
6694
6695                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6696
6697                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6698                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6699
6700                 param_len = scsi_2btoul(cdb->length);
6701                 header_size = sizeof(struct scsi_mode_header_10);
6702                 break;
6703         }
6704         default:
6705                 ctl_set_invalid_opcode(ctsio);
6706                 ctl_done((union ctl_io *)ctsio);
6707                 return (CTL_RETVAL_COMPLETE);
6708                 break; /* NOTREACHED */
6709         }
6710
6711         /*
6712          * From SPC-3:
6713          * "A parameter list length of zero indicates that the Data-Out Buffer
6714          * shall be empty. This condition shall not be considered as an error."
6715          */
6716         if (param_len == 0) {
6717                 ctl_set_success(ctsio);
6718                 ctl_done((union ctl_io *)ctsio);
6719                 return (CTL_RETVAL_COMPLETE);
6720         }
6721
6722         /*
6723          * Since we'll hit this the first time through, prior to
6724          * allocation, we don't need to free a data buffer here.
6725          */
6726         if (param_len < header_size) {
6727                 ctl_set_param_len_error(ctsio);
6728                 ctl_done((union ctl_io *)ctsio);
6729                 return (CTL_RETVAL_COMPLETE);
6730         }
6731
6732         /*
6733          * Allocate the data buffer and grab the user's data.  In theory,
6734          * we shouldn't have to sanity check the parameter list length here
6735          * because the maximum size is 64K.  We should be able to malloc
6736          * that much without too many problems.
6737          */
6738         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6739                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6740                 ctsio->kern_data_len = param_len;
6741                 ctsio->kern_total_len = param_len;
6742                 ctsio->kern_data_resid = 0;
6743                 ctsio->kern_rel_offset = 0;
6744                 ctsio->kern_sg_entries = 0;
6745                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6746                 ctsio->be_move_done = ctl_config_move_done;
6747                 ctl_datamove((union ctl_io *)ctsio);
6748
6749                 return (CTL_RETVAL_COMPLETE);
6750         }
6751
6752         switch (ctsio->cdb[0]) {
6753         case MODE_SELECT_6: {
6754                 struct scsi_mode_header_6 *mh6;
6755
6756                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6757                 bd_len = mh6->blk_desc_len;
6758                 break;
6759         }
6760         case MODE_SELECT_10: {
6761                 struct scsi_mode_header_10 *mh10;
6762
6763                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6764                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6765                 break;
6766         }
6767         default:
6768                 panic("Invalid CDB type %#x", ctsio->cdb[0]);
6769                 break;
6770         }
6771
6772         if (param_len < (header_size + bd_len)) {
6773                 free(ctsio->kern_data_ptr, M_CTL);
6774                 ctl_set_param_len_error(ctsio);
6775                 ctl_done((union ctl_io *)ctsio);
6776                 return (CTL_RETVAL_COMPLETE);
6777         }
6778
6779         /*
6780          * Set the IO_CONT flag, so that if this I/O gets passed to
6781          * ctl_config_write_done(), it'll get passed back to
6782          * ctl_do_mode_select() for further processing, or completion if
6783          * we're all done.
6784          */
6785         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6786         ctsio->io_cont = ctl_do_mode_select;
6787
6788         modepage_info = (union ctl_modepage_info *)
6789                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6790
6791         memset(modepage_info, 0, sizeof(*modepage_info));
6792
6793         len_left = param_len - header_size - bd_len;
6794         len_used = header_size + bd_len;
6795
6796         modepage_info->header.len_left = len_left;
6797         modepage_info->header.len_used = len_used;
6798
6799         return (ctl_do_mode_select((union ctl_io *)ctsio));
6800 }
6801
6802 int
6803 ctl_mode_sense(struct ctl_scsiio *ctsio)
6804 {
6805         struct ctl_lun *lun;
6806         int pc, page_code, dbd, llba, subpage;
6807         int alloc_len, page_len, header_len, total_len;
6808         struct scsi_mode_block_descr *block_desc;
6809         struct ctl_page_index *page_index;
6810         int control_dev;
6811
6812         dbd = 0;
6813         llba = 0;
6814         block_desc = NULL;
6815         page_index = NULL;
6816
6817         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6818
6819         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6820
6821         if (lun->be_lun->lun_type != T_DIRECT)
6822                 control_dev = 1;
6823         else
6824                 control_dev = 0;
6825
6826         if (lun->flags & CTL_LUN_PR_RESERVED) {
6827                 uint32_t residx;
6828
6829                 /*
6830                  * XXX KDM need a lock here.
6831                  */
6832                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6833                 if ((lun->res_type == SPR_TYPE_EX_AC
6834                   && residx != lun->pr_res_idx)
6835                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
6836                    || lun->res_type == SPR_TYPE_EX_AC_AR)
6837                   && !lun->per_res[residx].registered)) {
6838                         ctl_set_reservation_conflict(ctsio);
6839                         ctl_done((union ctl_io *)ctsio);
6840                         return (CTL_RETVAL_COMPLETE);
6841                 }
6842         }
6843
6844         switch (ctsio->cdb[0]) {
6845         case MODE_SENSE_6: {
6846                 struct scsi_mode_sense_6 *cdb;
6847
6848                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6849
6850                 header_len = sizeof(struct scsi_mode_hdr_6);
6851                 if (cdb->byte2 & SMS_DBD)
6852                         dbd = 1;
6853                 else
6854                         header_len += sizeof(struct scsi_mode_block_descr);
6855
6856                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6857                 page_code = cdb->page & SMS_PAGE_CODE;
6858                 subpage = cdb->subpage;
6859                 alloc_len = cdb->length;
6860                 break;
6861         }
6862         case MODE_SENSE_10: {
6863                 struct scsi_mode_sense_10 *cdb;
6864
6865                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6866
6867                 header_len = sizeof(struct scsi_mode_hdr_10);
6868
6869                 if (cdb->byte2 & SMS_DBD)
6870                         dbd = 1;
6871                 else
6872                         header_len += sizeof(struct scsi_mode_block_descr);
6873                 if (cdb->byte2 & SMS10_LLBAA)
6874                         llba = 1;
6875                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6876                 page_code = cdb->page & SMS_PAGE_CODE;
6877                 subpage = cdb->subpage;
6878                 alloc_len = scsi_2btoul(cdb->length);
6879                 break;
6880         }
6881         default:
6882                 ctl_set_invalid_opcode(ctsio);
6883                 ctl_done((union ctl_io *)ctsio);
6884                 return (CTL_RETVAL_COMPLETE);
6885                 break; /* NOTREACHED */
6886         }
6887
6888         /*
6889          * We have to make a first pass through to calculate the size of
6890          * the pages that match the user's query.  Then we allocate enough
6891          * memory to hold it, and actually copy the data into the buffer.
6892          */
6893         switch (page_code) {
6894         case SMS_ALL_PAGES_PAGE: {
6895                 int i;
6896
6897                 page_len = 0;
6898
6899                 /*
6900                  * At the moment, values other than 0 and 0xff here are
6901                  * reserved according to SPC-3.
6902                  */
6903                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6904                  && (subpage != SMS_SUBPAGE_ALL)) {
6905                         ctl_set_invalid_field(ctsio,
6906                                               /*sks_valid*/ 1,
6907                                               /*command*/ 1,
6908                                               /*field*/ 3,
6909                                               /*bit_valid*/ 0,
6910                                               /*bit*/ 0);
6911                         ctl_done((union ctl_io *)ctsio);
6912                         return (CTL_RETVAL_COMPLETE);
6913                 }
6914
6915                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6916                         if ((control_dev != 0)
6917                          && (lun->mode_pages.index[i].page_flags &
6918                              CTL_PAGE_FLAG_DISK_ONLY))
6919                                 continue;
6920
6921                         /*
6922                          * We don't use this subpage if the user didn't
6923                          * request all subpages.
6924                          */
6925                         if ((lun->mode_pages.index[i].subpage != 0)
6926                          && (subpage == SMS_SUBPAGE_PAGE_0))
6927                                 continue;
6928
6929 #if 0
6930                         printf("found page %#x len %d\n",
6931                                lun->mode_pages.index[i].page_code &
6932                                SMPH_PC_MASK,
6933                                lun->mode_pages.index[i].page_len);
6934 #endif
6935                         page_len += lun->mode_pages.index[i].page_len;
6936                 }
6937                 break;
6938         }
6939         default: {
6940                 int i;
6941
6942                 page_len = 0;
6943
6944                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6945                         /* Look for the right page code */
6946                         if ((lun->mode_pages.index[i].page_code &
6947                              SMPH_PC_MASK) != page_code)
6948                                 continue;
6949
6950                         /* Look for the right subpage or the subpage wildcard*/
6951                         if ((lun->mode_pages.index[i].subpage != subpage)
6952                          && (subpage != SMS_SUBPAGE_ALL))
6953                                 continue;
6954
6955                         /* Make sure the page is supported for this dev type */
6956                         if ((control_dev != 0)
6957                          && (lun->mode_pages.index[i].page_flags &
6958                              CTL_PAGE_FLAG_DISK_ONLY))
6959                                 continue;
6960
6961 #if 0
6962                         printf("found page %#x len %d\n",
6963                                lun->mode_pages.index[i].page_code &
6964                                SMPH_PC_MASK,
6965                                lun->mode_pages.index[i].page_len);
6966 #endif
6967
6968                         page_len += lun->mode_pages.index[i].page_len;
6969                 }
6970
6971                 if (page_len == 0) {
6972                         ctl_set_invalid_field(ctsio,
6973                                               /*sks_valid*/ 1,
6974                                               /*command*/ 1,
6975                                               /*field*/ 2,
6976                                               /*bit_valid*/ 1,
6977                                               /*bit*/ 5);
6978                         ctl_done((union ctl_io *)ctsio);
6979                         return (CTL_RETVAL_COMPLETE);
6980                 }
6981                 break;
6982         }
6983         }
6984
6985         total_len = header_len + page_len;
6986 #if 0
6987         printf("header_len = %d, page_len = %d, total_len = %d\n",
6988                header_len, page_len, total_len);
6989 #endif
6990
6991         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6992         ctsio->kern_sg_entries = 0;
6993         ctsio->kern_data_resid = 0;
6994         ctsio->kern_rel_offset = 0;
6995         if (total_len < alloc_len) {
6996                 ctsio->residual = alloc_len - total_len;
6997                 ctsio->kern_data_len = total_len;
6998                 ctsio->kern_total_len = total_len;
6999         } else {
7000                 ctsio->residual = 0;
7001                 ctsio->kern_data_len = alloc_len;
7002                 ctsio->kern_total_len = alloc_len;
7003         }
7004
7005         switch (ctsio->cdb[0]) {
7006         case MODE_SENSE_6: {
7007                 struct scsi_mode_hdr_6 *header;
7008
7009                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7010
7011                 header->datalen = ctl_min(total_len - 1, 254);
7012
7013                 if (dbd)
7014                         header->block_descr_len = 0;
7015                 else
7016                         header->block_descr_len =
7017                                 sizeof(struct scsi_mode_block_descr);
7018                 block_desc = (struct scsi_mode_block_descr *)&header[1];
7019                 break;
7020         }
7021         case MODE_SENSE_10: {
7022                 struct scsi_mode_hdr_10 *header;
7023                 int datalen;
7024
7025                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7026
7027                 datalen = ctl_min(total_len - 2, 65533);
7028                 scsi_ulto2b(datalen, header->datalen);
7029                 if (dbd)
7030                         scsi_ulto2b(0, header->block_descr_len);
7031                 else
7032                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7033                                     header->block_descr_len);
7034                 block_desc = (struct scsi_mode_block_descr *)&header[1];
7035                 break;
7036         }
7037         default:
7038                 panic("invalid CDB type %#x", ctsio->cdb[0]);
7039                 break; /* NOTREACHED */
7040         }
7041
7042         /*
7043          * If we've got a disk, use its blocksize in the block
7044          * descriptor.  Otherwise, just set it to 0.
7045          */
7046         if (dbd == 0) {
7047                 if (control_dev != 0)
7048                         scsi_ulto3b(lun->be_lun->blocksize,
7049                                     block_desc->block_len);
7050                 else
7051                         scsi_ulto3b(0, block_desc->block_len);
7052         }
7053
7054         switch (page_code) {
7055         case SMS_ALL_PAGES_PAGE: {
7056                 int i, data_used;
7057
7058                 data_used = header_len;
7059                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7060                         struct ctl_page_index *page_index;
7061
7062                         page_index = &lun->mode_pages.index[i];
7063
7064                         if ((control_dev != 0)
7065                          && (page_index->page_flags &
7066                             CTL_PAGE_FLAG_DISK_ONLY))
7067                                 continue;
7068
7069                         /*
7070                          * We don't use this subpage if the user didn't
7071                          * request all subpages.  We already checked (above)
7072                          * to make sure the user only specified a subpage
7073                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7074                          */
7075                         if ((page_index->subpage != 0)
7076                          && (subpage == SMS_SUBPAGE_PAGE_0))
7077                                 continue;
7078
7079                         /*
7080                          * Call the handler, if it exists, to update the
7081                          * page to the latest values.
7082                          */
7083                         if (page_index->sense_handler != NULL)
7084                                 page_index->sense_handler(ctsio, page_index,pc);
7085
7086                         memcpy(ctsio->kern_data_ptr + data_used,
7087                                page_index->page_data +
7088                                (page_index->page_len * pc),
7089                                page_index->page_len);
7090                         data_used += page_index->page_len;
7091                 }
7092                 break;
7093         }
7094         default: {
7095                 int i, data_used;
7096
7097                 data_used = header_len;
7098
7099                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7100                         struct ctl_page_index *page_index;
7101
7102                         page_index = &lun->mode_pages.index[i];
7103
7104                         /* Look for the right page code */
7105                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7106                                 continue;
7107
7108                         /* Look for the right subpage or the subpage wildcard*/
7109                         if ((page_index->subpage != subpage)
7110                          && (subpage != SMS_SUBPAGE_ALL))
7111                                 continue;
7112
7113                         /* Make sure the page is supported for this dev type */
7114                         if ((control_dev != 0)
7115                          && (page_index->page_flags &
7116                              CTL_PAGE_FLAG_DISK_ONLY))
7117                                 continue;
7118
7119                         /*
7120                          * Call the handler, if it exists, to update the
7121                          * page to the latest values.
7122                          */
7123                         if (page_index->sense_handler != NULL)
7124                                 page_index->sense_handler(ctsio, page_index,pc);
7125
7126                         memcpy(ctsio->kern_data_ptr + data_used,
7127                                page_index->page_data +
7128                                (page_index->page_len * pc),
7129                                page_index->page_len);
7130                         data_used += page_index->page_len;
7131                 }
7132                 break;
7133         }
7134         }
7135
7136         ctsio->scsi_status = SCSI_STATUS_OK;
7137
7138         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7139         ctsio->be_move_done = ctl_config_move_done;
7140         ctl_datamove((union ctl_io *)ctsio);
7141
7142         return (CTL_RETVAL_COMPLETE);
7143 }
7144
7145 int
7146 ctl_read_capacity(struct ctl_scsiio *ctsio)
7147 {
7148         struct scsi_read_capacity *cdb;
7149         struct scsi_read_capacity_data *data;
7150         struct ctl_lun *lun;
7151         uint32_t lba;
7152
7153         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7154
7155         cdb = (struct scsi_read_capacity *)ctsio->cdb;
7156
7157         lba = scsi_4btoul(cdb->addr);
7158         if (((cdb->pmi & SRC_PMI) == 0)
7159          && (lba != 0)) {
7160                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7161                                       /*sks_valid*/ 1,
7162                                       /*command*/ 1,
7163                                       /*field*/ 2,
7164                                       /*bit_valid*/ 0,
7165                                       /*bit*/ 0);
7166                 ctl_done((union ctl_io *)ctsio);
7167                 return (CTL_RETVAL_COMPLETE);
7168         }
7169
7170         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7171
7172         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7173         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7174         ctsio->residual = 0;
7175         ctsio->kern_data_len = sizeof(*data);
7176         ctsio->kern_total_len = sizeof(*data);
7177         ctsio->kern_data_resid = 0;
7178         ctsio->kern_rel_offset = 0;
7179         ctsio->kern_sg_entries = 0;
7180
7181         /*
7182          * If the maximum LBA is greater than 0xfffffffe, the user must
7183          * issue a SERVICE ACTION IN (16) command, with the read capacity
7184          * serivce action set.
7185          */
7186         if (lun->be_lun->maxlba > 0xfffffffe)
7187                 scsi_ulto4b(0xffffffff, data->addr);
7188         else
7189                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7190
7191         /*
7192          * XXX KDM this may not be 512 bytes...
7193          */
7194         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7195
7196         ctsio->scsi_status = SCSI_STATUS_OK;
7197
7198         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7199         ctsio->be_move_done = ctl_config_move_done;
7200         ctl_datamove((union ctl_io *)ctsio);
7201
7202         return (CTL_RETVAL_COMPLETE);
7203 }
7204
7205 int
7206 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7207 {
7208         struct scsi_read_capacity_16 *cdb;
7209         struct scsi_read_capacity_data_long *data;
7210         struct ctl_lun *lun;
7211         uint64_t lba;
7212         uint32_t alloc_len;
7213
7214         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7215
7216         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7217
7218         alloc_len = scsi_4btoul(cdb->alloc_len);
7219         lba = scsi_8btou64(cdb->addr);
7220
7221         if ((cdb->reladr & SRC16_PMI)
7222          && (lba != 0)) {
7223                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7224                                       /*sks_valid*/ 1,
7225                                       /*command*/ 1,
7226                                       /*field*/ 2,
7227                                       /*bit_valid*/ 0,
7228                                       /*bit*/ 0);
7229                 ctl_done((union ctl_io *)ctsio);
7230                 return (CTL_RETVAL_COMPLETE);
7231         }
7232
7233         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7234
7235         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7236         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7237
7238         if (sizeof(*data) < alloc_len) {
7239                 ctsio->residual = alloc_len - sizeof(*data);
7240                 ctsio->kern_data_len = sizeof(*data);
7241                 ctsio->kern_total_len = sizeof(*data);
7242         } else {
7243                 ctsio->residual = 0;
7244                 ctsio->kern_data_len = alloc_len;
7245                 ctsio->kern_total_len = alloc_len;
7246         }
7247         ctsio->kern_data_resid = 0;
7248         ctsio->kern_rel_offset = 0;
7249         ctsio->kern_sg_entries = 0;
7250
7251         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7252         /* XXX KDM this may not be 512 bytes... */
7253         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7254         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7255         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7256         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7257                 data->lalba_lbp[0] |= SRC16_LBPME;
7258
7259         ctsio->scsi_status = SCSI_STATUS_OK;
7260
7261         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7262         ctsio->be_move_done = ctl_config_move_done;
7263         ctl_datamove((union ctl_io *)ctsio);
7264
7265         return (CTL_RETVAL_COMPLETE);
7266 }
7267
7268 int
7269 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7270 {
7271         struct scsi_maintenance_in *cdb;
7272         int retval;
7273         int alloc_len, ext, total_len = 0, g, p, pc, pg;
7274         int num_target_port_groups, num_target_ports, single;
7275         struct ctl_lun *lun;
7276         struct ctl_softc *softc;
7277         struct ctl_port *port;
7278         struct scsi_target_group_data *rtg_ptr;
7279         struct scsi_target_group_data_extended *rtg_ext_ptr;
7280         struct scsi_target_port_group_descriptor *tpg_desc;
7281
7282         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7283
7284         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7285         softc = control_softc;
7286         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7287
7288         retval = CTL_RETVAL_COMPLETE;
7289
7290         switch (cdb->byte2 & STG_PDF_MASK) {
7291         case STG_PDF_LENGTH:
7292                 ext = 0;
7293                 break;
7294         case STG_PDF_EXTENDED:
7295                 ext = 1;
7296                 break;
7297         default:
7298                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7299                                       /*sks_valid*/ 1,
7300                                       /*command*/ 1,
7301                                       /*field*/ 2,
7302                                       /*bit_valid*/ 1,
7303                                       /*bit*/ 5);
7304                 ctl_done((union ctl_io *)ctsio);
7305                 return(retval);
7306         }
7307
7308         single = ctl_is_single;
7309         if (single)
7310                 num_target_port_groups = 1;
7311         else
7312                 num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7313         num_target_ports = 0;
7314         mtx_lock(&softc->ctl_lock);
7315         STAILQ_FOREACH(port, &softc->port_list, links) {
7316                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7317                         continue;
7318                 if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7319                         continue;
7320                 num_target_ports++;
7321         }
7322         mtx_unlock(&softc->ctl_lock);
7323
7324         if (ext)
7325                 total_len = sizeof(struct scsi_target_group_data_extended);
7326         else
7327                 total_len = sizeof(struct scsi_target_group_data);
7328         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7329                 num_target_port_groups +
7330             sizeof(struct scsi_target_port_descriptor) *
7331                 num_target_ports * num_target_port_groups;
7332
7333         alloc_len = scsi_4btoul(cdb->length);
7334
7335         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7336
7337         ctsio->kern_sg_entries = 0;
7338
7339         if (total_len < alloc_len) {
7340                 ctsio->residual = alloc_len - total_len;
7341                 ctsio->kern_data_len = total_len;
7342                 ctsio->kern_total_len = total_len;
7343         } else {
7344                 ctsio->residual = 0;
7345                 ctsio->kern_data_len = alloc_len;
7346                 ctsio->kern_total_len = alloc_len;
7347         }
7348         ctsio->kern_data_resid = 0;
7349         ctsio->kern_rel_offset = 0;
7350
7351         if (ext) {
7352                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7353                     ctsio->kern_data_ptr;
7354                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7355                 rtg_ext_ptr->format_type = 0x10;
7356                 rtg_ext_ptr->implicit_transition_time = 0;
7357                 tpg_desc = &rtg_ext_ptr->groups[0];
7358         } else {
7359                 rtg_ptr = (struct scsi_target_group_data *)
7360                     ctsio->kern_data_ptr;
7361                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7362                 tpg_desc = &rtg_ptr->groups[0];
7363         }
7364
7365         pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7366         mtx_lock(&softc->ctl_lock);
7367         for (g = 0; g < num_target_port_groups; g++) {
7368                 if (g == pg)
7369                         tpg_desc->pref_state = TPG_PRIMARY |
7370                             TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7371                 else
7372                         tpg_desc->pref_state =
7373                             TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7374                 tpg_desc->support = TPG_AO_SUP;
7375                 if (!single)
7376                         tpg_desc->support |= TPG_AN_SUP;
7377                 scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7378                 tpg_desc->status = TPG_IMPLICIT;
7379                 pc = 0;
7380                 STAILQ_FOREACH(port, &softc->port_list, links) {
7381                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7382                                 continue;
7383                         if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7384                             CTL_MAX_LUNS)
7385                                 continue;
7386                         p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7387                         scsi_ulto2b(p, tpg_desc->descriptors[pc].
7388                             relative_target_port_identifier);
7389                         pc++;
7390                 }
7391                 tpg_desc->target_port_count = pc;
7392                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7393                     &tpg_desc->descriptors[pc];
7394         }
7395         mtx_unlock(&softc->ctl_lock);
7396
7397         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7398         ctsio->be_move_done = ctl_config_move_done;
7399
7400         CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7401                          ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7402                          ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7403                          ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7404                          ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7405
7406         ctl_datamove((union ctl_io *)ctsio);
7407         return(retval);
7408 }
7409
7410 int
7411 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7412 {
7413         struct ctl_lun *lun;
7414         struct scsi_report_supported_opcodes *cdb;
7415         const struct ctl_cmd_entry *entry, *sentry;
7416         struct scsi_report_supported_opcodes_all *all;
7417         struct scsi_report_supported_opcodes_descr *descr;
7418         struct scsi_report_supported_opcodes_one *one;
7419         int retval;
7420         int alloc_len, total_len;
7421         int opcode, service_action, i, j, num;
7422
7423         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7424
7425         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7426         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7427
7428         retval = CTL_RETVAL_COMPLETE;
7429
7430         opcode = cdb->requested_opcode;
7431         service_action = scsi_2btoul(cdb->requested_service_action);
7432         switch (cdb->options & RSO_OPTIONS_MASK) {
7433         case RSO_OPTIONS_ALL:
7434                 num = 0;
7435                 for (i = 0; i < 256; i++) {
7436                         entry = &ctl_cmd_table[i];
7437                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7438                                 for (j = 0; j < 32; j++) {
7439                                         sentry = &((const struct ctl_cmd_entry *)
7440                                             entry->execute)[j];
7441                                         if (ctl_cmd_applicable(
7442                                             lun->be_lun->lun_type, sentry))
7443                                                 num++;
7444                                 }
7445                         } else {
7446                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7447                                     entry))
7448                                         num++;
7449                         }
7450                 }
7451                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7452                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7453                 break;
7454         case RSO_OPTIONS_OC:
7455                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7456                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7457                                               /*sks_valid*/ 1,
7458                                               /*command*/ 1,
7459                                               /*field*/ 2,
7460                                               /*bit_valid*/ 1,
7461                                               /*bit*/ 2);
7462                         ctl_done((union ctl_io *)ctsio);
7463                         return (CTL_RETVAL_COMPLETE);
7464                 }
7465                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7466                 break;
7467         case RSO_OPTIONS_OC_SA:
7468                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7469                     service_action >= 32) {
7470                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7471                                               /*sks_valid*/ 1,
7472                                               /*command*/ 1,
7473                                               /*field*/ 2,
7474                                               /*bit_valid*/ 1,
7475                                               /*bit*/ 2);
7476                         ctl_done((union ctl_io *)ctsio);
7477                         return (CTL_RETVAL_COMPLETE);
7478                 }
7479                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7480                 break;
7481         default:
7482                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7483                                       /*sks_valid*/ 1,
7484                                       /*command*/ 1,
7485                                       /*field*/ 2,
7486                                       /*bit_valid*/ 1,
7487                                       /*bit*/ 2);
7488                 ctl_done((union ctl_io *)ctsio);
7489                 return (CTL_RETVAL_COMPLETE);
7490         }
7491
7492         alloc_len = scsi_4btoul(cdb->length);
7493
7494         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7495
7496         ctsio->kern_sg_entries = 0;
7497
7498         if (total_len < alloc_len) {
7499                 ctsio->residual = alloc_len - total_len;
7500                 ctsio->kern_data_len = total_len;
7501                 ctsio->kern_total_len = total_len;
7502         } else {
7503                 ctsio->residual = 0;
7504                 ctsio->kern_data_len = alloc_len;
7505                 ctsio->kern_total_len = alloc_len;
7506         }
7507         ctsio->kern_data_resid = 0;
7508         ctsio->kern_rel_offset = 0;
7509
7510         switch (cdb->options & RSO_OPTIONS_MASK) {
7511         case RSO_OPTIONS_ALL:
7512                 all = (struct scsi_report_supported_opcodes_all *)
7513                     ctsio->kern_data_ptr;
7514                 num = 0;
7515                 for (i = 0; i < 256; i++) {
7516                         entry = &ctl_cmd_table[i];
7517                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7518                                 for (j = 0; j < 32; j++) {
7519                                         sentry = &((const struct ctl_cmd_entry *)
7520                                             entry->execute)[j];
7521                                         if (!ctl_cmd_applicable(
7522                                             lun->be_lun->lun_type, sentry))
7523                                                 continue;
7524                                         descr = &all->descr[num++];
7525                                         descr->opcode = i;
7526                                         scsi_ulto2b(j, descr->service_action);
7527                                         descr->flags = RSO_SERVACTV;
7528                                         scsi_ulto2b(sentry->length,
7529                                             descr->cdb_length);
7530                                 }
7531                         } else {
7532                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7533                                     entry))
7534                                         continue;
7535                                 descr = &all->descr[num++];
7536                                 descr->opcode = i;
7537                                 scsi_ulto2b(0, descr->service_action);
7538                                 descr->flags = 0;
7539                                 scsi_ulto2b(entry->length, descr->cdb_length);
7540                         }
7541                 }
7542                 scsi_ulto4b(
7543                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7544                     all->length);
7545                 break;
7546         case RSO_OPTIONS_OC:
7547                 one = (struct scsi_report_supported_opcodes_one *)
7548                     ctsio->kern_data_ptr;
7549                 entry = &ctl_cmd_table[opcode];
7550                 goto fill_one;
7551         case RSO_OPTIONS_OC_SA:
7552                 one = (struct scsi_report_supported_opcodes_one *)
7553                     ctsio->kern_data_ptr;
7554                 entry = &ctl_cmd_table[opcode];
7555                 entry = &((const struct ctl_cmd_entry *)
7556                     entry->execute)[service_action];
7557 fill_one:
7558                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7559                         one->support = 3;
7560                         scsi_ulto2b(entry->length, one->cdb_length);
7561                         one->cdb_usage[0] = opcode;
7562                         memcpy(&one->cdb_usage[1], entry->usage,
7563                             entry->length - 1);
7564                 } else
7565                         one->support = 1;
7566                 break;
7567         }
7568
7569         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7570         ctsio->be_move_done = ctl_config_move_done;
7571
7572         ctl_datamove((union ctl_io *)ctsio);
7573         return(retval);
7574 }
7575
7576 int
7577 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7578 {
7579         struct ctl_lun *lun;
7580         struct scsi_report_supported_tmf *cdb;
7581         struct scsi_report_supported_tmf_data *data;
7582         int retval;
7583         int alloc_len, total_len;
7584
7585         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7586
7587         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7588         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7589
7590         retval = CTL_RETVAL_COMPLETE;
7591
7592         total_len = sizeof(struct scsi_report_supported_tmf_data);
7593         alloc_len = scsi_4btoul(cdb->length);
7594
7595         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7596
7597         ctsio->kern_sg_entries = 0;
7598
7599         if (total_len < alloc_len) {
7600                 ctsio->residual = alloc_len - total_len;
7601                 ctsio->kern_data_len = total_len;
7602                 ctsio->kern_total_len = total_len;
7603         } else {
7604                 ctsio->residual = 0;
7605                 ctsio->kern_data_len = alloc_len;
7606                 ctsio->kern_total_len = alloc_len;
7607         }
7608         ctsio->kern_data_resid = 0;
7609         ctsio->kern_rel_offset = 0;
7610
7611         data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7612         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7613         data->byte2 |= RST_ITNRS;
7614
7615         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7616         ctsio->be_move_done = ctl_config_move_done;
7617
7618         ctl_datamove((union ctl_io *)ctsio);
7619         return (retval);
7620 }
7621
7622 int
7623 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7624 {
7625         struct ctl_lun *lun;
7626         struct scsi_report_timestamp *cdb;
7627         struct scsi_report_timestamp_data *data;
7628         struct timeval tv;
7629         int64_t timestamp;
7630         int retval;
7631         int alloc_len, total_len;
7632
7633         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7634
7635         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7636         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7637
7638         retval = CTL_RETVAL_COMPLETE;
7639
7640         total_len = sizeof(struct scsi_report_timestamp_data);
7641         alloc_len = scsi_4btoul(cdb->length);
7642
7643         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7644
7645         ctsio->kern_sg_entries = 0;
7646
7647         if (total_len < alloc_len) {
7648                 ctsio->residual = alloc_len - total_len;
7649                 ctsio->kern_data_len = total_len;
7650                 ctsio->kern_total_len = total_len;
7651         } else {
7652                 ctsio->residual = 0;
7653                 ctsio->kern_data_len = alloc_len;
7654                 ctsio->kern_total_len = alloc_len;
7655         }
7656         ctsio->kern_data_resid = 0;
7657         ctsio->kern_rel_offset = 0;
7658
7659         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7660         scsi_ulto2b(sizeof(*data) - 2, data->length);
7661         data->origin = RTS_ORIG_OUTSIDE;
7662         getmicrotime(&tv);
7663         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7664         scsi_ulto4b(timestamp >> 16, data->timestamp);
7665         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7666
7667         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7668         ctsio->be_move_done = ctl_config_move_done;
7669
7670         ctl_datamove((union ctl_io *)ctsio);
7671         return (retval);
7672 }
7673
7674 int
7675 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7676 {
7677         struct scsi_per_res_in *cdb;
7678         int alloc_len, total_len = 0;
7679         /* struct scsi_per_res_in_rsrv in_data; */
7680         struct ctl_lun *lun;
7681         struct ctl_softc *softc;
7682
7683         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7684
7685         softc = control_softc;
7686
7687         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7688
7689         alloc_len = scsi_2btoul(cdb->length);
7690
7691         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7692
7693 retry:
7694         mtx_lock(&lun->lun_lock);
7695         switch (cdb->action) {
7696         case SPRI_RK: /* read keys */
7697                 total_len = sizeof(struct scsi_per_res_in_keys) +
7698                         lun->pr_key_count *
7699                         sizeof(struct scsi_per_res_key);
7700                 break;
7701         case SPRI_RR: /* read reservation */
7702                 if (lun->flags & CTL_LUN_PR_RESERVED)
7703                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7704                 else
7705                         total_len = sizeof(struct scsi_per_res_in_header);
7706                 break;
7707         case SPRI_RC: /* report capabilities */
7708                 total_len = sizeof(struct scsi_per_res_cap);
7709                 break;
7710         case SPRI_RS: /* read full status */
7711                 total_len = sizeof(struct scsi_per_res_in_header) +
7712                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7713                     lun->pr_key_count;
7714                 break;
7715         default:
7716                 panic("Invalid PR type %x", cdb->action);
7717         }
7718         mtx_unlock(&lun->lun_lock);
7719
7720         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7721
7722         if (total_len < alloc_len) {
7723                 ctsio->residual = alloc_len - total_len;
7724                 ctsio->kern_data_len = total_len;
7725                 ctsio->kern_total_len = total_len;
7726         } else {
7727                 ctsio->residual = 0;
7728                 ctsio->kern_data_len = alloc_len;
7729                 ctsio->kern_total_len = alloc_len;
7730         }
7731
7732         ctsio->kern_data_resid = 0;
7733         ctsio->kern_rel_offset = 0;
7734         ctsio->kern_sg_entries = 0;
7735
7736         mtx_lock(&lun->lun_lock);
7737         switch (cdb->action) {
7738         case SPRI_RK: { // read keys
7739         struct scsi_per_res_in_keys *res_keys;
7740                 int i, key_count;
7741
7742                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7743
7744                 /*
7745                  * We had to drop the lock to allocate our buffer, which
7746                  * leaves time for someone to come in with another
7747                  * persistent reservation.  (That is unlikely, though,
7748                  * since this should be the only persistent reservation
7749                  * command active right now.)
7750                  */
7751                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7752                     (lun->pr_key_count *
7753                      sizeof(struct scsi_per_res_key)))){
7754                         mtx_unlock(&lun->lun_lock);
7755                         free(ctsio->kern_data_ptr, M_CTL);
7756                         printf("%s: reservation length changed, retrying\n",
7757                                __func__);
7758                         goto retry;
7759                 }
7760
7761                 scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7762
7763                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7764                              lun->pr_key_count, res_keys->header.length);
7765
7766                 for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7767                         if (!lun->per_res[i].registered)
7768                                 continue;
7769
7770                         /*
7771                          * We used lun->pr_key_count to calculate the
7772                          * size to allocate.  If it turns out the number of
7773                          * initiators with the registered flag set is
7774                          * larger than that (i.e. they haven't been kept in
7775                          * sync), we've got a problem.
7776                          */
7777                         if (key_count >= lun->pr_key_count) {
7778 #ifdef NEEDTOPORT
7779                                 csevent_log(CSC_CTL | CSC_SHELF_SW |
7780                                             CTL_PR_ERROR,
7781                                             csevent_LogType_Fault,
7782                                             csevent_AlertLevel_Yellow,
7783                                             csevent_FRU_ShelfController,
7784                                             csevent_FRU_Firmware,
7785                                         csevent_FRU_Unknown,
7786                                             "registered keys %d >= key "
7787                                             "count %d", key_count,
7788                                             lun->pr_key_count);
7789 #endif
7790                                 key_count++;
7791                                 continue;
7792                         }
7793                         memcpy(res_keys->keys[key_count].key,
7794                                lun->per_res[i].res_key.key,
7795                                ctl_min(sizeof(res_keys->keys[key_count].key),
7796                                sizeof(lun->per_res[i].res_key)));
7797                         key_count++;
7798                 }
7799                 break;
7800         }
7801         case SPRI_RR: { // read reservation
7802                 struct scsi_per_res_in_rsrv *res;
7803                 int tmp_len, header_only;
7804
7805                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7806
7807                 scsi_ulto4b(lun->PRGeneration, res->header.generation);
7808
7809                 if (lun->flags & CTL_LUN_PR_RESERVED)
7810                 {
7811                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7812                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7813                                     res->header.length);
7814                         header_only = 0;
7815                 } else {
7816                         tmp_len = sizeof(struct scsi_per_res_in_header);
7817                         scsi_ulto4b(0, res->header.length);
7818                         header_only = 1;
7819                 }
7820
7821                 /*
7822                  * We had to drop the lock to allocate our buffer, which
7823                  * leaves time for someone to come in with another
7824                  * persistent reservation.  (That is unlikely, though,
7825                  * since this should be the only persistent reservation
7826                  * command active right now.)
7827                  */
7828                 if (tmp_len != total_len) {
7829                         mtx_unlock(&lun->lun_lock);
7830                         free(ctsio->kern_data_ptr, M_CTL);
7831                         printf("%s: reservation status changed, retrying\n",
7832                                __func__);
7833                         goto retry;
7834                 }
7835
7836                 /*
7837                  * No reservation held, so we're done.
7838                  */
7839                 if (header_only != 0)
7840                         break;
7841
7842                 /*
7843                  * If the registration is an All Registrants type, the key
7844                  * is 0, since it doesn't really matter.
7845                  */
7846                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7847                         memcpy(res->data.reservation,
7848                                &lun->per_res[lun->pr_res_idx].res_key,
7849                                sizeof(struct scsi_per_res_key));
7850                 }
7851                 res->data.scopetype = lun->res_type;
7852                 break;
7853         }
7854         case SPRI_RC:     //report capabilities
7855         {
7856                 struct scsi_per_res_cap *res_cap;
7857                 uint16_t type_mask;
7858
7859                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7860                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7861                 res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_3;
7862                 type_mask = SPRI_TM_WR_EX_AR |
7863                             SPRI_TM_EX_AC_RO |
7864                             SPRI_TM_WR_EX_RO |
7865                             SPRI_TM_EX_AC |
7866                             SPRI_TM_WR_EX |
7867                             SPRI_TM_EX_AC_AR;
7868                 scsi_ulto2b(type_mask, res_cap->type_mask);
7869                 break;
7870         }
7871         case SPRI_RS: { // read full status
7872                 struct scsi_per_res_in_full *res_status;
7873                 struct scsi_per_res_in_full_desc *res_desc;
7874                 struct ctl_port *port;
7875                 int i, len;
7876
7877                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7878
7879                 /*
7880                  * We had to drop the lock to allocate our buffer, which
7881                  * leaves time for someone to come in with another
7882                  * persistent reservation.  (That is unlikely, though,
7883                  * since this should be the only persistent reservation
7884                  * command active right now.)
7885                  */
7886                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7887                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7888                      lun->pr_key_count)){
7889                         mtx_unlock(&lun->lun_lock);
7890                         free(ctsio->kern_data_ptr, M_CTL);
7891                         printf("%s: reservation length changed, retrying\n",
7892                                __func__);
7893                         goto retry;
7894                 }
7895
7896                 scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7897
7898                 res_desc = &res_status->desc[0];
7899                 for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7900                         if (!lun->per_res[i].registered)
7901                                 continue;
7902
7903                         memcpy(&res_desc->res_key, &lun->per_res[i].res_key.key,
7904                             sizeof(res_desc->res_key));
7905                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7906                             (lun->pr_res_idx == i ||
7907                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7908                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7909                                 res_desc->scopetype = lun->res_type;
7910                         }
7911                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7912                             res_desc->rel_trgt_port_id);
7913                         len = 0;
7914                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7915                         if (port != NULL)
7916                                 len = ctl_create_iid(port,
7917                                     i % CTL_MAX_INIT_PER_PORT,
7918                                     res_desc->transport_id);
7919                         scsi_ulto4b(len, res_desc->additional_length);
7920                         res_desc = (struct scsi_per_res_in_full_desc *)
7921                             &res_desc->transport_id[len];
7922                 }
7923                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7924                     res_status->header.length);
7925                 break;
7926         }
7927         default:
7928                 /*
7929                  * This is a bug, because we just checked for this above,
7930                  * and should have returned an error.
7931                  */
7932                 panic("Invalid PR type %x", cdb->action);
7933                 break; /* NOTREACHED */
7934         }
7935         mtx_unlock(&lun->lun_lock);
7936
7937         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7938         ctsio->be_move_done = ctl_config_move_done;
7939
7940         CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7941                          ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7942                          ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7943                          ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7944                          ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7945
7946         ctl_datamove((union ctl_io *)ctsio);
7947
7948         return (CTL_RETVAL_COMPLETE);
7949 }
7950
7951 /*
7952  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7953  * it should return.
7954  */
7955 static int
7956 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7957                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7958                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7959                 struct scsi_per_res_out_parms* param)
7960 {
7961         union ctl_ha_msg persis_io;
7962         int retval, i;
7963         int isc_retval;
7964
7965         retval = 0;
7966
7967         mtx_lock(&lun->lun_lock);
7968         if (sa_res_key == 0) {
7969                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7970                         /* validate scope and type */
7971                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7972                              SPR_LU_SCOPE) {
7973                                 mtx_unlock(&lun->lun_lock);
7974                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7975                                                       /*sks_valid*/ 1,
7976                                                       /*command*/ 1,
7977                                                       /*field*/ 2,
7978                                                       /*bit_valid*/ 1,
7979                                                       /*bit*/ 4);
7980                                 ctl_done((union ctl_io *)ctsio);
7981                                 return (1);
7982                         }
7983
7984                         if (type>8 || type==2 || type==4 || type==0) {
7985                                 mtx_unlock(&lun->lun_lock);
7986                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7987                                                       /*sks_valid*/ 1,
7988                                                       /*command*/ 1,
7989                                                       /*field*/ 2,
7990                                                       /*bit_valid*/ 1,
7991                                                       /*bit*/ 0);
7992                                 ctl_done((union ctl_io *)ctsio);
7993                                 return (1);
7994                         }
7995
7996                         /* temporarily unregister this nexus */
7997                         lun->per_res[residx].registered = 0;
7998
7999                         /*
8000                          * Unregister everybody else and build UA for
8001                          * them
8002                          */
8003                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8004                                 if (lun->per_res[i].registered == 0)
8005                                         continue;
8006
8007                                 if (!persis_offset
8008                                  && i <CTL_MAX_INITIATORS)
8009                                         lun->pending_ua[i] |=
8010                                                 CTL_UA_REG_PREEMPT;
8011                                 else if (persis_offset
8012                                       && i >= persis_offset)
8013                                         lun->pending_ua[i-persis_offset] |=
8014                                                 CTL_UA_REG_PREEMPT;
8015                                 lun->per_res[i].registered = 0;
8016                                 memset(&lun->per_res[i].res_key, 0,
8017                                        sizeof(struct scsi_per_res_key));
8018                         }
8019                         lun->per_res[residx].registered = 1;
8020                         lun->pr_key_count = 1;
8021                         lun->res_type = type;
8022                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8023                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8024                                 lun->pr_res_idx = residx;
8025
8026                         /* send msg to other side */
8027                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8028                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8029                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8030                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8031                         persis_io.pr.pr_info.res_type = type;
8032                         memcpy(persis_io.pr.pr_info.sa_res_key,
8033                                param->serv_act_res_key,
8034                                sizeof(param->serv_act_res_key));
8035                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8036                              &persis_io, sizeof(persis_io), 0)) >
8037                              CTL_HA_STATUS_SUCCESS) {
8038                                 printf("CTL:Persis Out error returned "
8039                                        "from ctl_ha_msg_send %d\n",
8040                                        isc_retval);
8041                         }
8042                 } else {
8043                         /* not all registrants */
8044                         mtx_unlock(&lun->lun_lock);
8045                         free(ctsio->kern_data_ptr, M_CTL);
8046                         ctl_set_invalid_field(ctsio,
8047                                               /*sks_valid*/ 1,
8048                                               /*command*/ 0,
8049                                               /*field*/ 8,
8050                                               /*bit_valid*/ 0,
8051                                               /*bit*/ 0);
8052                         ctl_done((union ctl_io *)ctsio);
8053                         return (1);
8054                 }
8055         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8056                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
8057                 int found = 0;
8058
8059                 if (res_key == sa_res_key) {
8060                         /* special case */
8061                         /*
8062                          * The spec implies this is not good but doesn't
8063                          * say what to do. There are two choices either
8064                          * generate a res conflict or check condition
8065                          * with illegal field in parameter data. Since
8066                          * that is what is done when the sa_res_key is
8067                          * zero I'll take that approach since this has
8068                          * to do with the sa_res_key.
8069                          */
8070                         mtx_unlock(&lun->lun_lock);
8071                         free(ctsio->kern_data_ptr, M_CTL);
8072                         ctl_set_invalid_field(ctsio,
8073                                               /*sks_valid*/ 1,
8074                                               /*command*/ 0,
8075                                               /*field*/ 8,
8076                                               /*bit_valid*/ 0,
8077                                               /*bit*/ 0);
8078                         ctl_done((union ctl_io *)ctsio);
8079                         return (1);
8080                 }
8081
8082                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8083                         if (lun->per_res[i].registered
8084                          && memcmp(param->serv_act_res_key,
8085                             lun->per_res[i].res_key.key,
8086                             sizeof(struct scsi_per_res_key)) != 0)
8087                                 continue;
8088
8089                         found = 1;
8090                         lun->per_res[i].registered = 0;
8091                         memset(&lun->per_res[i].res_key, 0,
8092                                sizeof(struct scsi_per_res_key));
8093                         lun->pr_key_count--;
8094
8095                         if (!persis_offset && i < CTL_MAX_INITIATORS)
8096                                 lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8097                         else if (persis_offset && i >= persis_offset)
8098                                 lun->pending_ua[i-persis_offset] |=
8099                                         CTL_UA_REG_PREEMPT;
8100                 }
8101                 if (!found) {
8102                         mtx_unlock(&lun->lun_lock);
8103                         free(ctsio->kern_data_ptr, M_CTL);
8104                         ctl_set_reservation_conflict(ctsio);
8105                         ctl_done((union ctl_io *)ctsio);
8106                         return (CTL_RETVAL_COMPLETE);
8107                 }
8108                 /* send msg to other side */
8109                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8110                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8111                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8112                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
8113                 persis_io.pr.pr_info.res_type = type;
8114                 memcpy(persis_io.pr.pr_info.sa_res_key,
8115                        param->serv_act_res_key,
8116                        sizeof(param->serv_act_res_key));
8117                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8118                      &persis_io, sizeof(persis_io), 0)) >
8119                      CTL_HA_STATUS_SUCCESS) {
8120                         printf("CTL:Persis Out error returned from "
8121                                "ctl_ha_msg_send %d\n", isc_retval);
8122                 }
8123         } else {
8124                 /* Reserved but not all registrants */
8125                 /* sa_res_key is res holder */
8126                 if (memcmp(param->serv_act_res_key,
8127                    lun->per_res[lun->pr_res_idx].res_key.key,
8128                    sizeof(struct scsi_per_res_key)) == 0) {
8129                         /* validate scope and type */
8130                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8131                              SPR_LU_SCOPE) {
8132                                 mtx_unlock(&lun->lun_lock);
8133                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8134                                                       /*sks_valid*/ 1,
8135                                                       /*command*/ 1,
8136                                                       /*field*/ 2,
8137                                                       /*bit_valid*/ 1,
8138                                                       /*bit*/ 4);
8139                                 ctl_done((union ctl_io *)ctsio);
8140                                 return (1);
8141                         }
8142
8143                         if (type>8 || type==2 || type==4 || type==0) {
8144                                 mtx_unlock(&lun->lun_lock);
8145                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8146                                                       /*sks_valid*/ 1,
8147                                                       /*command*/ 1,
8148                                                       /*field*/ 2,
8149                                                       /*bit_valid*/ 1,
8150                                                       /*bit*/ 0);
8151                                 ctl_done((union ctl_io *)ctsio);
8152                                 return (1);
8153                         }
8154
8155                         /*
8156                          * Do the following:
8157                          * if sa_res_key != res_key remove all
8158                          * registrants w/sa_res_key and generate UA
8159                          * for these registrants(Registrations
8160                          * Preempted) if it wasn't an exclusive
8161                          * reservation generate UA(Reservations
8162                          * Preempted) for all other registered nexuses
8163                          * if the type has changed. Establish the new
8164                          * reservation and holder. If res_key and
8165                          * sa_res_key are the same do the above
8166                          * except don't unregister the res holder.
8167                          */
8168
8169                         /*
8170                          * Temporarily unregister so it won't get
8171                          * removed or UA generated
8172                          */
8173                         lun->per_res[residx].registered = 0;
8174                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8175                                 if (lun->per_res[i].registered == 0)
8176                                         continue;
8177
8178                                 if (memcmp(param->serv_act_res_key,
8179                                     lun->per_res[i].res_key.key,
8180                                     sizeof(struct scsi_per_res_key)) == 0) {
8181                                         lun->per_res[i].registered = 0;
8182                                         memset(&lun->per_res[i].res_key,
8183                                                0,
8184                                                sizeof(struct scsi_per_res_key));
8185                                         lun->pr_key_count--;
8186
8187                                         if (!persis_offset
8188                                          && i < CTL_MAX_INITIATORS)
8189                                                 lun->pending_ua[i] |=
8190                                                         CTL_UA_REG_PREEMPT;
8191                                         else if (persis_offset
8192                                               && i >= persis_offset)
8193                                                 lun->pending_ua[i-persis_offset] |=
8194                                                   CTL_UA_REG_PREEMPT;
8195                                 } else if (type != lun->res_type
8196                                         && (lun->res_type == SPR_TYPE_WR_EX_RO
8197                                          || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8198                                                 if (!persis_offset
8199                                                  && i < CTL_MAX_INITIATORS)
8200                                                         lun->pending_ua[i] |=
8201                                                         CTL_UA_RES_RELEASE;
8202                                                 else if (persis_offset
8203                                                       && i >= persis_offset)
8204                                                         lun->pending_ua[
8205                                                         i-persis_offset] |=
8206                                                         CTL_UA_RES_RELEASE;
8207                                 }
8208                         }
8209                         lun->per_res[residx].registered = 1;
8210                         lun->res_type = type;
8211                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8212                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8213                                 lun->pr_res_idx = residx;
8214                         else
8215                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8216
8217                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8218                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8219                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8220                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8221                         persis_io.pr.pr_info.res_type = type;
8222                         memcpy(persis_io.pr.pr_info.sa_res_key,
8223                                param->serv_act_res_key,
8224                                sizeof(param->serv_act_res_key));
8225                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8226                              &persis_io, sizeof(persis_io), 0)) >
8227                              CTL_HA_STATUS_SUCCESS) {
8228                                 printf("CTL:Persis Out error returned "
8229                                        "from ctl_ha_msg_send %d\n",
8230                                        isc_retval);
8231                         }
8232                 } else {
8233                         /*
8234                          * sa_res_key is not the res holder just
8235                          * remove registrants
8236                          */
8237                         int found=0;
8238
8239                         for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8240                                 if (memcmp(param->serv_act_res_key,
8241                                     lun->per_res[i].res_key.key,
8242                                     sizeof(struct scsi_per_res_key)) != 0)
8243                                         continue;
8244
8245                                 found = 1;
8246                                 lun->per_res[i].registered = 0;
8247                                 memset(&lun->per_res[i].res_key, 0,
8248                                        sizeof(struct scsi_per_res_key));
8249                                 lun->pr_key_count--;
8250
8251                                 if (!persis_offset
8252                                  && i < CTL_MAX_INITIATORS)
8253                                         lun->pending_ua[i] |=
8254                                                 CTL_UA_REG_PREEMPT;
8255                                 else if (persis_offset
8256                                       && i >= persis_offset)
8257                                         lun->pending_ua[i-persis_offset] |=
8258                                                 CTL_UA_REG_PREEMPT;
8259                         }
8260
8261                         if (!found) {
8262                                 mtx_unlock(&lun->lun_lock);
8263                                 free(ctsio->kern_data_ptr, M_CTL);
8264                                 ctl_set_reservation_conflict(ctsio);
8265                                 ctl_done((union ctl_io *)ctsio);
8266                                 return (1);
8267                         }
8268                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8269                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8270                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8271                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8272                         persis_io.pr.pr_info.res_type = type;
8273                         memcpy(persis_io.pr.pr_info.sa_res_key,
8274                                param->serv_act_res_key,
8275                                sizeof(param->serv_act_res_key));
8276                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8277                              &persis_io, sizeof(persis_io), 0)) >
8278                              CTL_HA_STATUS_SUCCESS) {
8279                                 printf("CTL:Persis Out error returned "
8280                                        "from ctl_ha_msg_send %d\n",
8281                                 isc_retval);
8282                         }
8283                 }
8284         }
8285
8286         lun->PRGeneration++;
8287         mtx_unlock(&lun->lun_lock);
8288
8289         return (retval);
8290 }
8291
8292 static void
8293 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8294 {
8295         int i;
8296
8297         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8298          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8299          || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8300                    msg->pr.pr_info.sa_res_key,
8301                    sizeof(struct scsi_per_res_key)) != 0) {
8302                 uint64_t sa_res_key;
8303                 sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8304
8305                 if (sa_res_key == 0) {
8306                         /* temporarily unregister this nexus */
8307                         lun->per_res[msg->pr.pr_info.residx].registered = 0;
8308
8309                         /*
8310                          * Unregister everybody else and build UA for
8311                          * them
8312                          */
8313                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8314                                 if (lun->per_res[i].registered == 0)
8315                                         continue;
8316
8317                                 if (!persis_offset
8318                                  && i < CTL_MAX_INITIATORS)
8319                                         lun->pending_ua[i] |=
8320                                                 CTL_UA_REG_PREEMPT;
8321                                 else if (persis_offset && i >= persis_offset)
8322                                         lun->pending_ua[i - persis_offset] |=
8323                                                 CTL_UA_REG_PREEMPT;
8324                                 lun->per_res[i].registered = 0;
8325                                 memset(&lun->per_res[i].res_key, 0,
8326                                        sizeof(struct scsi_per_res_key));
8327                         }
8328
8329                         lun->per_res[msg->pr.pr_info.residx].registered = 1;
8330                         lun->pr_key_count = 1;
8331                         lun->res_type = msg->pr.pr_info.res_type;
8332                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8333                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8334                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8335                 } else {
8336                         for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8337                                 if (memcmp(msg->pr.pr_info.sa_res_key,
8338                                    lun->per_res[i].res_key.key,
8339                                    sizeof(struct scsi_per_res_key)) != 0)
8340                                         continue;
8341
8342                                 lun->per_res[i].registered = 0;
8343                                 memset(&lun->per_res[i].res_key, 0,
8344                                        sizeof(struct scsi_per_res_key));
8345                                 lun->pr_key_count--;
8346
8347                                 if (!persis_offset
8348                                  && i < persis_offset)
8349                                         lun->pending_ua[i] |=
8350                                                 CTL_UA_REG_PREEMPT;
8351                                 else if (persis_offset
8352                                       && i >= persis_offset)
8353                                         lun->pending_ua[i - persis_offset] |=
8354                                                 CTL_UA_REG_PREEMPT;
8355                         }
8356                 }
8357         } else {
8358                 /*
8359                  * Temporarily unregister so it won't get removed
8360                  * or UA generated
8361                  */
8362                 lun->per_res[msg->pr.pr_info.residx].registered = 0;
8363                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8364                         if (lun->per_res[i].registered == 0)
8365                                 continue;
8366
8367                         if (memcmp(msg->pr.pr_info.sa_res_key,
8368                            lun->per_res[i].res_key.key,
8369                            sizeof(struct scsi_per_res_key)) == 0) {
8370                                 lun->per_res[i].registered = 0;
8371                                 memset(&lun->per_res[i].res_key, 0,
8372                                        sizeof(struct scsi_per_res_key));
8373                                 lun->pr_key_count--;
8374                                 if (!persis_offset
8375                                  && i < CTL_MAX_INITIATORS)
8376                                         lun->pending_ua[i] |=
8377                                                 CTL_UA_REG_PREEMPT;
8378                                 else if (persis_offset
8379                                       && i >= persis_offset)
8380                                         lun->pending_ua[i - persis_offset] |=
8381                                                 CTL_UA_REG_PREEMPT;
8382                         } else if (msg->pr.pr_info.res_type != lun->res_type
8383                                 && (lun->res_type == SPR_TYPE_WR_EX_RO
8384                                  || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8385                                         if (!persis_offset
8386                                          && i < persis_offset)
8387                                                 lun->pending_ua[i] |=
8388                                                         CTL_UA_RES_RELEASE;
8389                                         else if (persis_offset
8390                                               && i >= persis_offset)
8391                                         lun->pending_ua[i - persis_offset] |=
8392                                                 CTL_UA_RES_RELEASE;
8393                         }
8394                 }
8395                 lun->per_res[msg->pr.pr_info.residx].registered = 1;
8396                 lun->res_type = msg->pr.pr_info.res_type;
8397                 if (lun->res_type != SPR_TYPE_WR_EX_AR
8398                  && lun->res_type != SPR_TYPE_EX_AC_AR)
8399                         lun->pr_res_idx = msg->pr.pr_info.residx;
8400                 else
8401                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8402         }
8403         lun->PRGeneration++;
8404
8405 }
8406
8407
8408 int
8409 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8410 {
8411         int retval;
8412         int isc_retval;
8413         u_int32_t param_len;
8414         struct scsi_per_res_out *cdb;
8415         struct ctl_lun *lun;
8416         struct scsi_per_res_out_parms* param;
8417         struct ctl_softc *softc;
8418         uint32_t residx;
8419         uint64_t res_key, sa_res_key;
8420         uint8_t type;
8421         union ctl_ha_msg persis_io;
8422         int    i;
8423
8424         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8425
8426         retval = CTL_RETVAL_COMPLETE;
8427
8428         softc = control_softc;
8429
8430         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8431         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8432
8433         /*
8434          * We only support whole-LUN scope.  The scope & type are ignored for
8435          * register, register and ignore existing key and clear.
8436          * We sometimes ignore scope and type on preempts too!!
8437          * Verify reservation type here as well.
8438          */
8439         type = cdb->scope_type & SPR_TYPE_MASK;
8440         if ((cdb->action == SPRO_RESERVE)
8441          || (cdb->action == SPRO_RELEASE)) {
8442                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8443                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8444                                               /*sks_valid*/ 1,
8445                                               /*command*/ 1,
8446                                               /*field*/ 2,
8447                                               /*bit_valid*/ 1,
8448                                               /*bit*/ 4);
8449                         ctl_done((union ctl_io *)ctsio);
8450                         return (CTL_RETVAL_COMPLETE);
8451                 }
8452
8453                 if (type>8 || type==2 || type==4 || type==0) {
8454                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8455                                               /*sks_valid*/ 1,
8456                                               /*command*/ 1,
8457                                               /*field*/ 2,
8458                                               /*bit_valid*/ 1,
8459                                               /*bit*/ 0);
8460                         ctl_done((union ctl_io *)ctsio);
8461                         return (CTL_RETVAL_COMPLETE);
8462                 }
8463         }
8464
8465         param_len = scsi_4btoul(cdb->length);
8466
8467         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8468                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8469                 ctsio->kern_data_len = param_len;
8470                 ctsio->kern_total_len = param_len;
8471                 ctsio->kern_data_resid = 0;
8472                 ctsio->kern_rel_offset = 0;
8473                 ctsio->kern_sg_entries = 0;
8474                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8475                 ctsio->be_move_done = ctl_config_move_done;
8476                 ctl_datamove((union ctl_io *)ctsio);
8477
8478                 return (CTL_RETVAL_COMPLETE);
8479         }
8480
8481         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8482
8483         residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8484         res_key = scsi_8btou64(param->res_key.key);
8485         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8486
8487         /*
8488          * Validate the reservation key here except for SPRO_REG_IGNO
8489          * This must be done for all other service actions
8490          */
8491         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8492                 mtx_lock(&lun->lun_lock);
8493                 if (lun->per_res[residx].registered) {
8494                     if (memcmp(param->res_key.key,
8495                                lun->per_res[residx].res_key.key,
8496                                ctl_min(sizeof(param->res_key),
8497                                sizeof(lun->per_res[residx].res_key))) != 0) {
8498                                 /*
8499                                  * The current key passed in doesn't match
8500                                  * the one the initiator previously
8501                                  * registered.
8502                                  */
8503                                 mtx_unlock(&lun->lun_lock);
8504                                 free(ctsio->kern_data_ptr, M_CTL);
8505                                 ctl_set_reservation_conflict(ctsio);
8506                                 ctl_done((union ctl_io *)ctsio);
8507                                 return (CTL_RETVAL_COMPLETE);
8508                         }
8509                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8510                         /*
8511                          * We are not registered
8512                          */
8513                         mtx_unlock(&lun->lun_lock);
8514                         free(ctsio->kern_data_ptr, M_CTL);
8515                         ctl_set_reservation_conflict(ctsio);
8516                         ctl_done((union ctl_io *)ctsio);
8517                         return (CTL_RETVAL_COMPLETE);
8518                 } else if (res_key != 0) {
8519                         /*
8520                          * We are not registered and trying to register but
8521                          * the register key isn't zero.
8522                          */
8523                         mtx_unlock(&lun->lun_lock);
8524                         free(ctsio->kern_data_ptr, M_CTL);
8525                         ctl_set_reservation_conflict(ctsio);
8526                         ctl_done((union ctl_io *)ctsio);
8527                         return (CTL_RETVAL_COMPLETE);
8528                 }
8529                 mtx_unlock(&lun->lun_lock);
8530         }
8531
8532         switch (cdb->action & SPRO_ACTION_MASK) {
8533         case SPRO_REGISTER:
8534         case SPRO_REG_IGNO: {
8535
8536 #if 0
8537                 printf("Registration received\n");
8538 #endif
8539
8540                 /*
8541                  * We don't support any of these options, as we report in
8542                  * the read capabilities request (see
8543                  * ctl_persistent_reserve_in(), above).
8544                  */
8545                 if ((param->flags & SPR_SPEC_I_PT)
8546                  || (param->flags & SPR_ALL_TG_PT)
8547                  || (param->flags & SPR_APTPL)) {
8548                         int bit_ptr;
8549
8550                         if (param->flags & SPR_APTPL)
8551                                 bit_ptr = 0;
8552                         else if (param->flags & SPR_ALL_TG_PT)
8553                                 bit_ptr = 2;
8554                         else /* SPR_SPEC_I_PT */
8555                                 bit_ptr = 3;
8556
8557                         free(ctsio->kern_data_ptr, M_CTL);
8558                         ctl_set_invalid_field(ctsio,
8559                                               /*sks_valid*/ 1,
8560                                               /*command*/ 0,
8561                                               /*field*/ 20,
8562                                               /*bit_valid*/ 1,
8563                                               /*bit*/ bit_ptr);
8564                         ctl_done((union ctl_io *)ctsio);
8565                         return (CTL_RETVAL_COMPLETE);
8566                 }
8567
8568                 mtx_lock(&lun->lun_lock);
8569
8570                 /*
8571                  * The initiator wants to clear the
8572                  * key/unregister.
8573                  */
8574                 if (sa_res_key == 0) {
8575                         if ((res_key == 0
8576                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8577                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8578                           && !lun->per_res[residx].registered)) {
8579                                 mtx_unlock(&lun->lun_lock);
8580                                 goto done;
8581                         }
8582
8583                         lun->per_res[residx].registered = 0;
8584                         memset(&lun->per_res[residx].res_key,
8585                                0, sizeof(lun->per_res[residx].res_key));
8586                         lun->pr_key_count--;
8587
8588                         if (residx == lun->pr_res_idx) {
8589                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8590                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8591
8592                                 if ((lun->res_type == SPR_TYPE_WR_EX_RO
8593                                   || lun->res_type == SPR_TYPE_EX_AC_RO)
8594                                  && lun->pr_key_count) {
8595                                         /*
8596                                          * If the reservation is a registrants
8597                                          * only type we need to generate a UA
8598                                          * for other registered inits.  The
8599                                          * sense code should be RESERVATIONS
8600                                          * RELEASED
8601                                          */
8602
8603                                         for (i = 0; i < CTL_MAX_INITIATORS;i++){
8604                                                 if (lun->per_res[
8605                                                     i+persis_offset].registered
8606                                                     == 0)
8607                                                         continue;
8608                                                 lun->pending_ua[i] |=
8609                                                         CTL_UA_RES_RELEASE;
8610                                         }
8611                                 }
8612                                 lun->res_type = 0;
8613                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8614                                 if (lun->pr_key_count==0) {
8615                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8616                                         lun->res_type = 0;
8617                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8618                                 }
8619                         }
8620                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8621                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8622                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8623                         persis_io.pr.pr_info.residx = residx;
8624                         if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8625                              &persis_io, sizeof(persis_io), 0 )) >
8626                              CTL_HA_STATUS_SUCCESS) {
8627                                 printf("CTL:Persis Out error returned from "
8628                                        "ctl_ha_msg_send %d\n", isc_retval);
8629                         }
8630                 } else /* sa_res_key != 0 */ {
8631
8632                         /*
8633                          * If we aren't registered currently then increment
8634                          * the key count and set the registered flag.
8635                          */
8636                         if (!lun->per_res[residx].registered) {
8637                                 lun->pr_key_count++;
8638                                 lun->per_res[residx].registered = 1;
8639                         }
8640
8641                         memcpy(&lun->per_res[residx].res_key,
8642                                param->serv_act_res_key,
8643                                ctl_min(sizeof(param->serv_act_res_key),
8644                                sizeof(lun->per_res[residx].res_key)));
8645
8646                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8647                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8648                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8649                         persis_io.pr.pr_info.residx = residx;
8650                         memcpy(persis_io.pr.pr_info.sa_res_key,
8651                                param->serv_act_res_key,
8652                                sizeof(param->serv_act_res_key));
8653                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8654                              &persis_io, sizeof(persis_io), 0)) >
8655                              CTL_HA_STATUS_SUCCESS) {
8656                                 printf("CTL:Persis Out error returned from "
8657                                        "ctl_ha_msg_send %d\n", isc_retval);
8658                         }
8659                 }
8660                 lun->PRGeneration++;
8661                 mtx_unlock(&lun->lun_lock);
8662
8663                 break;
8664         }
8665         case SPRO_RESERVE:
8666 #if 0
8667                 printf("Reserve executed type %d\n", type);
8668 #endif
8669                 mtx_lock(&lun->lun_lock);
8670                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8671                         /*
8672                          * if this isn't the reservation holder and it's
8673                          * not a "all registrants" type or if the type is
8674                          * different then we have a conflict
8675                          */
8676                         if ((lun->pr_res_idx != residx
8677                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8678                          || lun->res_type != type) {
8679                                 mtx_unlock(&lun->lun_lock);
8680                                 free(ctsio->kern_data_ptr, M_CTL);
8681                                 ctl_set_reservation_conflict(ctsio);
8682                                 ctl_done((union ctl_io *)ctsio);
8683                                 return (CTL_RETVAL_COMPLETE);
8684                         }
8685                         mtx_unlock(&lun->lun_lock);
8686                 } else /* create a reservation */ {
8687                         /*
8688                          * If it's not an "all registrants" type record
8689                          * reservation holder
8690                          */
8691                         if (type != SPR_TYPE_WR_EX_AR
8692                          && type != SPR_TYPE_EX_AC_AR)
8693                                 lun->pr_res_idx = residx; /* Res holder */
8694                         else
8695                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8696
8697                         lun->flags |= CTL_LUN_PR_RESERVED;
8698                         lun->res_type = type;
8699
8700                         mtx_unlock(&lun->lun_lock);
8701
8702                         /* send msg to other side */
8703                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8704                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8705                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8706                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8707                         persis_io.pr.pr_info.res_type = type;
8708                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8709                              &persis_io, sizeof(persis_io), 0)) >
8710                              CTL_HA_STATUS_SUCCESS) {
8711                                 printf("CTL:Persis Out error returned from "
8712                                        "ctl_ha_msg_send %d\n", isc_retval);
8713                         }
8714                 }
8715                 break;
8716
8717         case SPRO_RELEASE:
8718                 mtx_lock(&lun->lun_lock);
8719                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8720                         /* No reservation exists return good status */
8721                         mtx_unlock(&lun->lun_lock);
8722                         goto done;
8723                 }
8724                 /*
8725                  * Is this nexus a reservation holder?
8726                  */
8727                 if (lun->pr_res_idx != residx
8728                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8729                         /*
8730                          * not a res holder return good status but
8731                          * do nothing
8732                          */
8733                         mtx_unlock(&lun->lun_lock);
8734                         goto done;
8735                 }
8736
8737                 if (lun->res_type != type) {
8738                         mtx_unlock(&lun->lun_lock);
8739                         free(ctsio->kern_data_ptr, M_CTL);
8740                         ctl_set_illegal_pr_release(ctsio);
8741                         ctl_done((union ctl_io *)ctsio);
8742                         return (CTL_RETVAL_COMPLETE);
8743                 }
8744
8745                 /* okay to release */
8746                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8747                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8748                 lun->res_type = 0;
8749
8750                 /*
8751                  * if this isn't an exclusive access
8752                  * res generate UA for all other
8753                  * registrants.
8754                  */
8755                 if (type != SPR_TYPE_EX_AC
8756                  && type != SPR_TYPE_WR_EX) {
8757                         /*
8758                          * temporarily unregister so we don't generate UA
8759                          */
8760                         lun->per_res[residx].registered = 0;
8761
8762                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8763                                 if (lun->per_res[i+persis_offset].registered
8764                                     == 0)
8765                                         continue;
8766                                 lun->pending_ua[i] |=
8767                                         CTL_UA_RES_RELEASE;
8768                         }
8769
8770                         lun->per_res[residx].registered = 1;
8771                 }
8772                 mtx_unlock(&lun->lun_lock);
8773                 /* Send msg to other side */
8774                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8775                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8776                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8777                 if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8778                      sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8779                         printf("CTL:Persis Out error returned from "
8780                                "ctl_ha_msg_send %d\n", isc_retval);
8781                 }
8782                 break;
8783
8784         case SPRO_CLEAR:
8785                 /* send msg to other side */
8786
8787                 mtx_lock(&lun->lun_lock);
8788                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8789                 lun->res_type = 0;
8790                 lun->pr_key_count = 0;
8791                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8792
8793
8794                 memset(&lun->per_res[residx].res_key,
8795                        0, sizeof(lun->per_res[residx].res_key));
8796                 lun->per_res[residx].registered = 0;
8797
8798                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8799                         if (lun->per_res[i].registered) {
8800                                 if (!persis_offset && i < CTL_MAX_INITIATORS)
8801                                         lun->pending_ua[i] |=
8802                                                 CTL_UA_RES_PREEMPT;
8803                                 else if (persis_offset && i >= persis_offset)
8804                                         lun->pending_ua[i-persis_offset] |=
8805                                             CTL_UA_RES_PREEMPT;
8806
8807                                 memset(&lun->per_res[i].res_key,
8808                                        0, sizeof(struct scsi_per_res_key));
8809                                 lun->per_res[i].registered = 0;
8810                         }
8811                 lun->PRGeneration++;
8812                 mtx_unlock(&lun->lun_lock);
8813                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8814                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8815                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8816                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8817                      sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8818                         printf("CTL:Persis Out error returned from "
8819                                "ctl_ha_msg_send %d\n", isc_retval);
8820                 }
8821                 break;
8822
8823         case SPRO_PREEMPT: {
8824                 int nretval;
8825
8826                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8827                                           residx, ctsio, cdb, param);
8828                 if (nretval != 0)
8829                         return (CTL_RETVAL_COMPLETE);
8830                 break;
8831         }
8832         default:
8833                 panic("Invalid PR type %x", cdb->action);
8834         }
8835
8836 done:
8837         free(ctsio->kern_data_ptr, M_CTL);
8838         ctl_set_success(ctsio);
8839         ctl_done((union ctl_io *)ctsio);
8840
8841         return (retval);
8842 }
8843
8844 /*
8845  * This routine is for handling a message from the other SC pertaining to
8846  * persistent reserve out. All the error checking will have been done
8847  * so only perorming the action need be done here to keep the two
8848  * in sync.
8849  */
8850 static void
8851 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8852 {
8853         struct ctl_lun *lun;
8854         struct ctl_softc *softc;
8855         int i;
8856         uint32_t targ_lun;
8857
8858         softc = control_softc;
8859
8860         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8861         lun = softc->ctl_luns[targ_lun];
8862         mtx_lock(&lun->lun_lock);
8863         switch(msg->pr.pr_info.action) {
8864         case CTL_PR_REG_KEY:
8865                 if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8866                         lun->per_res[msg->pr.pr_info.residx].registered = 1;
8867                         lun->pr_key_count++;
8868                 }
8869                 lun->PRGeneration++;
8870                 memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8871                        msg->pr.pr_info.sa_res_key,
8872                        sizeof(struct scsi_per_res_key));
8873                 break;
8874
8875         case CTL_PR_UNREG_KEY:
8876                 lun->per_res[msg->pr.pr_info.residx].registered = 0;
8877                 memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8878                        0, sizeof(struct scsi_per_res_key));
8879                 lun->pr_key_count--;
8880
8881                 /* XXX Need to see if the reservation has been released */
8882                 /* if so do we need to generate UA? */
8883                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8884                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8885                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8886
8887                         if ((lun->res_type == SPR_TYPE_WR_EX_RO
8888                           || lun->res_type == SPR_TYPE_EX_AC_RO)
8889                          && lun->pr_key_count) {
8890                                 /*
8891                                  * If the reservation is a registrants
8892                                  * only type we need to generate a UA
8893                                  * for other registered inits.  The
8894                                  * sense code should be RESERVATIONS
8895                                  * RELEASED
8896                                  */
8897
8898                                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8899                                         if (lun->per_res[i+
8900                                             persis_offset].registered == 0)
8901                                                 continue;
8902
8903                                         lun->pending_ua[i] |=
8904                                                 CTL_UA_RES_RELEASE;
8905                                 }
8906                         }
8907                         lun->res_type = 0;
8908                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8909                         if (lun->pr_key_count==0) {
8910                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8911                                 lun->res_type = 0;
8912                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8913                         }
8914                 }
8915                 lun->PRGeneration++;
8916                 break;
8917
8918         case CTL_PR_RESERVE:
8919                 lun->flags |= CTL_LUN_PR_RESERVED;
8920                 lun->res_type = msg->pr.pr_info.res_type;
8921                 lun->pr_res_idx = msg->pr.pr_info.residx;
8922
8923                 break;
8924
8925         case CTL_PR_RELEASE:
8926                 /*
8927                  * if this isn't an exclusive access res generate UA for all
8928                  * other registrants.
8929                  */
8930                 if (lun->res_type != SPR_TYPE_EX_AC
8931                  && lun->res_type != SPR_TYPE_WR_EX) {
8932                         for (i = 0; i < CTL_MAX_INITIATORS; i++)
8933                                 if (lun->per_res[i+persis_offset].registered)
8934                                         lun->pending_ua[i] |=
8935                                                 CTL_UA_RES_RELEASE;
8936                 }
8937
8938                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8939                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8940                 lun->res_type = 0;
8941                 break;
8942
8943         case CTL_PR_PREEMPT:
8944                 ctl_pro_preempt_other(lun, msg);
8945                 break;
8946         case CTL_PR_CLEAR:
8947                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8948                 lun->res_type = 0;
8949                 lun->pr_key_count = 0;
8950                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8951
8952                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8953                         if (lun->per_res[i].registered == 0)
8954                                 continue;
8955                         if (!persis_offset
8956                          && i < CTL_MAX_INITIATORS)
8957                                 lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
8958                         else if (persis_offset
8959                               && i >= persis_offset)
8960                                 lun->pending_ua[i-persis_offset] |=
8961                                         CTL_UA_RES_PREEMPT;
8962                         memset(&lun->per_res[i].res_key, 0,
8963                                sizeof(struct scsi_per_res_key));
8964                         lun->per_res[i].registered = 0;
8965                 }
8966                 lun->PRGeneration++;
8967                 break;
8968         }
8969
8970         mtx_unlock(&lun->lun_lock);
8971 }
8972
8973 int
8974 ctl_read_write(struct ctl_scsiio *ctsio)
8975 {
8976         struct ctl_lun *lun;
8977         struct ctl_lba_len_flags *lbalen;
8978         uint64_t lba;
8979         uint32_t num_blocks;
8980         int fua, dpo;
8981         int retval;
8982         int isread;
8983
8984         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8985
8986         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8987
8988         fua = 0;
8989         dpo = 0;
8990
8991         retval = CTL_RETVAL_COMPLETE;
8992
8993         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8994               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8995         if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8996                 uint32_t residx;
8997
8998                 /*
8999                  * XXX KDM need a lock here.
9000                  */
9001                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9002                 if ((lun->res_type == SPR_TYPE_EX_AC
9003                   && residx != lun->pr_res_idx)
9004                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
9005                    || lun->res_type == SPR_TYPE_EX_AC_AR)
9006                   && !lun->per_res[residx].registered)) {
9007                         ctl_set_reservation_conflict(ctsio);
9008                         ctl_done((union ctl_io *)ctsio);
9009                         return (CTL_RETVAL_COMPLETE);
9010                 }
9011         }
9012
9013         switch (ctsio->cdb[0]) {
9014         case READ_6:
9015         case WRITE_6: {
9016                 struct scsi_rw_6 *cdb;
9017
9018                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
9019
9020                 lba = scsi_3btoul(cdb->addr);
9021                 /* only 5 bits are valid in the most significant address byte */
9022                 lba &= 0x1fffff;
9023                 num_blocks = cdb->length;
9024                 /*
9025                  * This is correct according to SBC-2.
9026                  */
9027                 if (num_blocks == 0)
9028                         num_blocks = 256;
9029                 break;
9030         }
9031         case READ_10:
9032         case WRITE_10: {
9033                 struct scsi_rw_10 *cdb;
9034
9035                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
9036
9037                 if (cdb->byte2 & SRW10_FUA)
9038                         fua = 1;
9039                 if (cdb->byte2 & SRW10_DPO)
9040                         dpo = 1;
9041
9042                 lba = scsi_4btoul(cdb->addr);
9043                 num_blocks = scsi_2btoul(cdb->length);
9044                 break;
9045         }
9046         case WRITE_VERIFY_10: {
9047                 struct scsi_write_verify_10 *cdb;
9048
9049                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9050
9051                 /*
9052                  * XXX KDM we should do actual write verify support at some
9053                  * point.  This is obviously fake, we're just translating
9054                  * things to a write.  So we don't even bother checking the
9055                  * BYTCHK field, since we don't do any verification.  If
9056                  * the user asks for it, we'll just pretend we did it.
9057                  */
9058                 if (cdb->byte2 & SWV_DPO)
9059                         dpo = 1;
9060
9061                 lba = scsi_4btoul(cdb->addr);
9062                 num_blocks = scsi_2btoul(cdb->length);
9063                 break;
9064         }
9065         case READ_12:
9066         case WRITE_12: {
9067                 struct scsi_rw_12 *cdb;
9068
9069                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
9070
9071                 if (cdb->byte2 & SRW12_FUA)
9072                         fua = 1;
9073                 if (cdb->byte2 & SRW12_DPO)
9074                         dpo = 1;
9075                 lba = scsi_4btoul(cdb->addr);
9076                 num_blocks = scsi_4btoul(cdb->length);
9077                 break;
9078         }
9079         case WRITE_VERIFY_12: {
9080                 struct scsi_write_verify_12 *cdb;
9081
9082                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9083
9084                 if (cdb->byte2 & SWV_DPO)
9085                         dpo = 1;
9086
9087                 lba = scsi_4btoul(cdb->addr);
9088                 num_blocks = scsi_4btoul(cdb->length);
9089
9090                 break;
9091         }
9092         case READ_16:
9093         case WRITE_16: {
9094                 struct scsi_rw_16 *cdb;
9095
9096                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9097
9098                 if (cdb->byte2 & SRW12_FUA)
9099                         fua = 1;
9100                 if (cdb->byte2 & SRW12_DPO)
9101                         dpo = 1;
9102
9103                 lba = scsi_8btou64(cdb->addr);
9104                 num_blocks = scsi_4btoul(cdb->length);
9105                 break;
9106         }
9107         case WRITE_VERIFY_16: {
9108                 struct scsi_write_verify_16 *cdb;
9109
9110                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9111
9112                 if (cdb->byte2 & SWV_DPO)
9113                         dpo = 1;
9114
9115                 lba = scsi_8btou64(cdb->addr);
9116                 num_blocks = scsi_4btoul(cdb->length);
9117                 break;
9118         }
9119         default:
9120                 /*
9121                  * We got a command we don't support.  This shouldn't
9122                  * happen, commands should be filtered out above us.
9123                  */
9124                 ctl_set_invalid_opcode(ctsio);
9125                 ctl_done((union ctl_io *)ctsio);
9126
9127                 return (CTL_RETVAL_COMPLETE);
9128                 break; /* NOTREACHED */
9129         }
9130
9131         /*
9132          * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
9133          * interesting for us, but if RAIDCore is in write-back mode,
9134          * getting it to do write-through for a particular transaction may
9135          * not be possible.
9136          */
9137
9138         /*
9139          * The first check is to make sure we're in bounds, the second
9140          * check is to catch wrap-around problems.  If the lba + num blocks
9141          * is less than the lba, then we've wrapped around and the block
9142          * range is invalid anyway.
9143          */
9144         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9145          || ((lba + num_blocks) < lba)) {
9146                 ctl_set_lba_out_of_range(ctsio);
9147                 ctl_done((union ctl_io *)ctsio);
9148                 return (CTL_RETVAL_COMPLETE);
9149         }
9150
9151         /*
9152          * According to SBC-3, a transfer length of 0 is not an error.
9153          * Note that this cannot happen with WRITE(6) or READ(6), since 0
9154          * translates to 256 blocks for those commands.
9155          */
9156         if (num_blocks == 0) {
9157                 ctl_set_success(ctsio);
9158                 ctl_done((union ctl_io *)ctsio);
9159                 return (CTL_RETVAL_COMPLETE);
9160         }
9161
9162         lbalen = (struct ctl_lba_len_flags *)
9163             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9164         lbalen->lba = lba;
9165         lbalen->len = num_blocks;
9166         lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
9167
9168         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9169         ctsio->kern_rel_offset = 0;
9170
9171         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9172
9173         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9174
9175         return (retval);
9176 }
9177
9178 static int
9179 ctl_cnw_cont(union ctl_io *io)
9180 {
9181         struct ctl_scsiio *ctsio;
9182         struct ctl_lun *lun;
9183         struct ctl_lba_len_flags *lbalen;
9184         int retval;
9185
9186         ctsio = &io->scsiio;
9187         ctsio->io_hdr.status = CTL_STATUS_NONE;
9188         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9189         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9190         lbalen = (struct ctl_lba_len_flags *)
9191             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9192         lbalen->flags = CTL_LLF_WRITE;
9193
9194         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9195         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9196         return (retval);
9197 }
9198
9199 int
9200 ctl_cnw(struct ctl_scsiio *ctsio)
9201 {
9202         struct ctl_lun *lun;
9203         struct ctl_lba_len_flags *lbalen;
9204         uint64_t lba;
9205         uint32_t num_blocks;
9206         int fua, dpo;
9207         int retval;
9208
9209         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9210
9211         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9212
9213         fua = 0;
9214         dpo = 0;
9215
9216         retval = CTL_RETVAL_COMPLETE;
9217
9218         switch (ctsio->cdb[0]) {
9219         case COMPARE_AND_WRITE: {
9220                 struct scsi_compare_and_write *cdb;
9221
9222                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9223
9224                 if (cdb->byte2 & SRW10_FUA)
9225                         fua = 1;
9226                 if (cdb->byte2 & SRW10_DPO)
9227                         dpo = 1;
9228                 lba = scsi_8btou64(cdb->addr);
9229                 num_blocks = cdb->length;
9230                 break;
9231         }
9232         default:
9233                 /*
9234                  * We got a command we don't support.  This shouldn't
9235                  * happen, commands should be filtered out above us.
9236                  */
9237                 ctl_set_invalid_opcode(ctsio);
9238                 ctl_done((union ctl_io *)ctsio);
9239
9240                 return (CTL_RETVAL_COMPLETE);
9241                 break; /* NOTREACHED */
9242         }
9243
9244         /*
9245          * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
9246          * interesting for us, but if RAIDCore is in write-back mode,
9247          * getting it to do write-through for a particular transaction may
9248          * not be possible.
9249          */
9250
9251         /*
9252          * The first check is to make sure we're in bounds, the second
9253          * check is to catch wrap-around problems.  If the lba + num blocks
9254          * is less than the lba, then we've wrapped around and the block
9255          * range is invalid anyway.
9256          */
9257         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9258          || ((lba + num_blocks) < lba)) {
9259                 ctl_set_lba_out_of_range(ctsio);
9260                 ctl_done((union ctl_io *)ctsio);
9261                 return (CTL_RETVAL_COMPLETE);
9262         }
9263
9264         /*
9265          * According to SBC-3, a transfer length of 0 is not an error.
9266          */
9267         if (num_blocks == 0) {
9268                 ctl_set_success(ctsio);
9269                 ctl_done((union ctl_io *)ctsio);
9270                 return (CTL_RETVAL_COMPLETE);
9271         }
9272
9273         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9274         ctsio->kern_rel_offset = 0;
9275
9276         /*
9277          * Set the IO_CONT flag, so that if this I/O gets passed to
9278          * ctl_data_submit_done(), it'll get passed back to
9279          * ctl_ctl_cnw_cont() for further processing.
9280          */
9281         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9282         ctsio->io_cont = ctl_cnw_cont;
9283
9284         lbalen = (struct ctl_lba_len_flags *)
9285             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9286         lbalen->lba = lba;
9287         lbalen->len = num_blocks;
9288         lbalen->flags = CTL_LLF_COMPARE;
9289
9290         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9291         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9292         return (retval);
9293 }
9294
9295 int
9296 ctl_verify(struct ctl_scsiio *ctsio)
9297 {
9298         struct ctl_lun *lun;
9299         struct ctl_lba_len_flags *lbalen;
9300         uint64_t lba;
9301         uint32_t num_blocks;
9302         int bytchk, dpo;
9303         int retval;
9304
9305         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9306
9307         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9308
9309         bytchk = 0;
9310         dpo = 0;
9311         retval = CTL_RETVAL_COMPLETE;
9312
9313         switch (ctsio->cdb[0]) {
9314         case VERIFY_10: {
9315                 struct scsi_verify_10 *cdb;
9316
9317                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
9318                 if (cdb->byte2 & SVFY_BYTCHK)
9319                         bytchk = 1;
9320                 if (cdb->byte2 & SVFY_DPO)
9321                         dpo = 1;
9322                 lba = scsi_4btoul(cdb->addr);
9323                 num_blocks = scsi_2btoul(cdb->length);
9324                 break;
9325         }
9326         case VERIFY_12: {
9327                 struct scsi_verify_12 *cdb;
9328
9329                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
9330                 if (cdb->byte2 & SVFY_BYTCHK)
9331                         bytchk = 1;
9332                 if (cdb->byte2 & SVFY_DPO)
9333                         dpo = 1;
9334                 lba = scsi_4btoul(cdb->addr);
9335                 num_blocks = scsi_4btoul(cdb->length);
9336                 break;
9337         }
9338         case VERIFY_16: {
9339                 struct scsi_rw_16 *cdb;
9340
9341                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9342                 if (cdb->byte2 & SVFY_BYTCHK)
9343                         bytchk = 1;
9344                 if (cdb->byte2 & SVFY_DPO)
9345                         dpo = 1;
9346                 lba = scsi_8btou64(cdb->addr);
9347                 num_blocks = scsi_4btoul(cdb->length);
9348                 break;
9349         }
9350         default:
9351                 /*
9352                  * We got a command we don't support.  This shouldn't
9353                  * happen, commands should be filtered out above us.
9354                  */
9355                 ctl_set_invalid_opcode(ctsio);
9356                 ctl_done((union ctl_io *)ctsio);
9357                 return (CTL_RETVAL_COMPLETE);
9358         }
9359
9360         /*
9361          * The first check is to make sure we're in bounds, the second
9362          * check is to catch wrap-around problems.  If the lba + num blocks
9363          * is less than the lba, then we've wrapped around and the block
9364          * range is invalid anyway.
9365          */
9366         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9367          || ((lba + num_blocks) < lba)) {
9368                 ctl_set_lba_out_of_range(ctsio);
9369                 ctl_done((union ctl_io *)ctsio);
9370                 return (CTL_RETVAL_COMPLETE);
9371         }
9372
9373         /*
9374          * According to SBC-3, a transfer length of 0 is not an error.
9375          */
9376         if (num_blocks == 0) {
9377                 ctl_set_success(ctsio);
9378                 ctl_done((union ctl_io *)ctsio);
9379                 return (CTL_RETVAL_COMPLETE);
9380         }
9381
9382         lbalen = (struct ctl_lba_len_flags *)
9383             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9384         lbalen->lba = lba;
9385         lbalen->len = num_blocks;
9386         if (bytchk) {
9387                 lbalen->flags = CTL_LLF_COMPARE;
9388                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9389         } else {
9390                 lbalen->flags = CTL_LLF_VERIFY;
9391                 ctsio->kern_total_len = 0;
9392         }
9393         ctsio->kern_rel_offset = 0;
9394
9395         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9396         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9397         return (retval);
9398 }
9399
9400 int
9401 ctl_report_luns(struct ctl_scsiio *ctsio)
9402 {
9403         struct scsi_report_luns *cdb;
9404         struct scsi_report_luns_data *lun_data;
9405         struct ctl_lun *lun, *request_lun;
9406         int num_luns, retval;
9407         uint32_t alloc_len, lun_datalen;
9408         int num_filled, well_known;
9409         uint32_t initidx, targ_lun_id, lun_id;
9410
9411         retval = CTL_RETVAL_COMPLETE;
9412         well_known = 0;
9413
9414         cdb = (struct scsi_report_luns *)ctsio->cdb;
9415
9416         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9417
9418         mtx_lock(&control_softc->ctl_lock);
9419         num_luns = control_softc->num_luns;
9420         mtx_unlock(&control_softc->ctl_lock);
9421
9422         switch (cdb->select_report) {
9423         case RPL_REPORT_DEFAULT:
9424         case RPL_REPORT_ALL:
9425                 break;
9426         case RPL_REPORT_WELLKNOWN:
9427                 well_known = 1;
9428                 num_luns = 0;
9429                 break;
9430         default:
9431                 ctl_set_invalid_field(ctsio,
9432                                       /*sks_valid*/ 1,
9433                                       /*command*/ 1,
9434                                       /*field*/ 2,
9435                                       /*bit_valid*/ 0,
9436                                       /*bit*/ 0);
9437                 ctl_done((union ctl_io *)ctsio);
9438                 return (retval);
9439                 break; /* NOTREACHED */
9440         }
9441
9442         alloc_len = scsi_4btoul(cdb->length);
9443         /*
9444          * The initiator has to allocate at least 16 bytes for this request,
9445          * so he can at least get the header and the first LUN.  Otherwise
9446          * we reject the request (per SPC-3 rev 14, section 6.21).
9447          */
9448         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9449             sizeof(struct scsi_report_luns_lundata))) {
9450                 ctl_set_invalid_field(ctsio,
9451                                       /*sks_valid*/ 1,
9452                                       /*command*/ 1,
9453                                       /*field*/ 6,
9454                                       /*bit_valid*/ 0,
9455                                       /*bit*/ 0);
9456                 ctl_done((union ctl_io *)ctsio);
9457                 return (retval);
9458         }
9459
9460         request_lun = (struct ctl_lun *)
9461                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9462
9463         lun_datalen = sizeof(*lun_data) +
9464                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9465
9466         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9467         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9468         ctsio->kern_sg_entries = 0;
9469
9470         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9471
9472         mtx_lock(&control_softc->ctl_lock);
9473         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9474                 lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9475                 if (lun_id >= CTL_MAX_LUNS)
9476                         continue;
9477                 lun = control_softc->ctl_luns[lun_id];
9478                 if (lun == NULL)
9479                         continue;
9480
9481                 if (targ_lun_id <= 0xff) {
9482                         /*
9483                          * Peripheral addressing method, bus number 0.
9484                          */
9485                         lun_data->luns[num_filled].lundata[0] =
9486                                 RPL_LUNDATA_ATYP_PERIPH;
9487                         lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9488                         num_filled++;
9489                 } else if (targ_lun_id <= 0x3fff) {
9490                         /*
9491                          * Flat addressing method.
9492                          */
9493                         lun_data->luns[num_filled].lundata[0] =
9494                                 RPL_LUNDATA_ATYP_FLAT |
9495                                 (targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9496 #ifdef OLDCTLHEADERS
9497                                 (SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9498                                 (targ_lun_id & SRLD_BUS_LUN_MASK);
9499 #endif
9500                         lun_data->luns[num_filled].lundata[1] =
9501 #ifdef OLDCTLHEADERS
9502                                 targ_lun_id >> SRLD_BUS_LUN_BITS;
9503 #endif
9504                                 targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9505                         num_filled++;
9506                 } else {
9507                         printf("ctl_report_luns: bogus LUN number %jd, "
9508                                "skipping\n", (intmax_t)targ_lun_id);
9509                 }
9510                 /*
9511                  * According to SPC-3, rev 14 section 6.21:
9512                  *
9513                  * "The execution of a REPORT LUNS command to any valid and
9514                  * installed logical unit shall clear the REPORTED LUNS DATA
9515                  * HAS CHANGED unit attention condition for all logical
9516                  * units of that target with respect to the requesting
9517                  * initiator. A valid and installed logical unit is one
9518                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9519                  * INQUIRY data (see 6.4.2)."
9520                  *
9521                  * If request_lun is NULL, the LUN this report luns command
9522                  * was issued to is either disabled or doesn't exist. In that
9523                  * case, we shouldn't clear any pending lun change unit
9524                  * attention.
9525                  */
9526                 if (request_lun != NULL) {
9527                         mtx_lock(&lun->lun_lock);
9528                         lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9529                         mtx_unlock(&lun->lun_lock);
9530                 }
9531         }
9532         mtx_unlock(&control_softc->ctl_lock);
9533
9534         /*
9535          * It's quite possible that we've returned fewer LUNs than we allocated
9536          * space for.  Trim it.
9537          */
9538         lun_datalen = sizeof(*lun_data) +
9539                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9540
9541         if (lun_datalen < alloc_len) {
9542                 ctsio->residual = alloc_len - lun_datalen;
9543                 ctsio->kern_data_len = lun_datalen;
9544                 ctsio->kern_total_len = lun_datalen;
9545         } else {
9546                 ctsio->residual = 0;
9547                 ctsio->kern_data_len = alloc_len;
9548                 ctsio->kern_total_len = alloc_len;
9549         }
9550         ctsio->kern_data_resid = 0;
9551         ctsio->kern_rel_offset = 0;
9552         ctsio->kern_sg_entries = 0;
9553
9554         /*
9555          * We set this to the actual data length, regardless of how much
9556          * space we actually have to return results.  If the user looks at
9557          * this value, he'll know whether or not he allocated enough space
9558          * and reissue the command if necessary.  We don't support well
9559          * known logical units, so if the user asks for that, return none.
9560          */
9561         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9562
9563         /*
9564          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9565          * this request.
9566          */
9567         ctsio->scsi_status = SCSI_STATUS_OK;
9568
9569         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9570         ctsio->be_move_done = ctl_config_move_done;
9571         ctl_datamove((union ctl_io *)ctsio);
9572
9573         return (retval);
9574 }
9575
9576 int
9577 ctl_request_sense(struct ctl_scsiio *ctsio)
9578 {
9579         struct scsi_request_sense *cdb;
9580         struct scsi_sense_data *sense_ptr;
9581         struct ctl_lun *lun;
9582         uint32_t initidx;
9583         int have_error;
9584         scsi_sense_data_type sense_format;
9585
9586         cdb = (struct scsi_request_sense *)ctsio->cdb;
9587
9588         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9589
9590         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9591
9592         /*
9593          * Determine which sense format the user wants.
9594          */
9595         if (cdb->byte2 & SRS_DESC)
9596                 sense_format = SSD_TYPE_DESC;
9597         else
9598                 sense_format = SSD_TYPE_FIXED;
9599
9600         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9601         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9602         ctsio->kern_sg_entries = 0;
9603
9604         /*
9605          * struct scsi_sense_data, which is currently set to 256 bytes, is
9606          * larger than the largest allowed value for the length field in the
9607          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9608          */
9609         ctsio->residual = 0;
9610         ctsio->kern_data_len = cdb->length;
9611         ctsio->kern_total_len = cdb->length;
9612
9613         ctsio->kern_data_resid = 0;
9614         ctsio->kern_rel_offset = 0;
9615         ctsio->kern_sg_entries = 0;
9616
9617         /*
9618          * If we don't have a LUN, we don't have any pending sense.
9619          */
9620         if (lun == NULL)
9621                 goto no_sense;
9622
9623         have_error = 0;
9624         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9625         /*
9626          * Check for pending sense, and then for pending unit attentions.
9627          * Pending sense gets returned first, then pending unit attentions.
9628          */
9629         mtx_lock(&lun->lun_lock);
9630 #ifdef CTL_WITH_CA
9631         if (ctl_is_set(lun->have_ca, initidx)) {
9632                 scsi_sense_data_type stored_format;
9633
9634                 /*
9635                  * Check to see which sense format was used for the stored
9636                  * sense data.
9637                  */
9638                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9639
9640                 /*
9641                  * If the user requested a different sense format than the
9642                  * one we stored, then we need to convert it to the other
9643                  * format.  If we're going from descriptor to fixed format
9644                  * sense data, we may lose things in translation, depending
9645                  * on what options were used.
9646                  *
9647                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9648                  * for some reason we'll just copy it out as-is.
9649                  */
9650                 if ((stored_format == SSD_TYPE_FIXED)
9651                  && (sense_format == SSD_TYPE_DESC))
9652                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9653                             &lun->pending_sense[initidx],
9654                             (struct scsi_sense_data_desc *)sense_ptr);
9655                 else if ((stored_format == SSD_TYPE_DESC)
9656                       && (sense_format == SSD_TYPE_FIXED))
9657                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9658                             &lun->pending_sense[initidx],
9659                             (struct scsi_sense_data_fixed *)sense_ptr);
9660                 else
9661                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9662                                ctl_min(sizeof(*sense_ptr),
9663                                sizeof(lun->pending_sense[initidx])));
9664
9665                 ctl_clear_mask(lun->have_ca, initidx);
9666                 have_error = 1;
9667         } else
9668 #endif
9669         if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9670                 ctl_ua_type ua_type;
9671
9672                 ua_type = ctl_build_ua(lun->pending_ua[initidx],
9673                                        sense_ptr, sense_format);
9674                 if (ua_type != CTL_UA_NONE) {
9675                         have_error = 1;
9676                         /* We're reporting this UA, so clear it */
9677                         lun->pending_ua[initidx] &= ~ua_type;
9678                 }
9679         }
9680         mtx_unlock(&lun->lun_lock);
9681
9682         /*
9683          * We already have a pending error, return it.
9684          */
9685         if (have_error != 0) {
9686                 /*
9687                  * We report the SCSI status as OK, since the status of the
9688                  * request sense command itself is OK.
9689                  */
9690                 ctsio->scsi_status = SCSI_STATUS_OK;
9691
9692                 /*
9693                  * We report 0 for the sense length, because we aren't doing
9694                  * autosense in this case.  We're reporting sense as
9695                  * parameter data.
9696                  */
9697                 ctsio->sense_len = 0;
9698                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9699                 ctsio->be_move_done = ctl_config_move_done;
9700                 ctl_datamove((union ctl_io *)ctsio);
9701
9702                 return (CTL_RETVAL_COMPLETE);
9703         }
9704
9705 no_sense:
9706
9707         /*
9708          * No sense information to report, so we report that everything is
9709          * okay.
9710          */
9711         ctl_set_sense_data(sense_ptr,
9712                            lun,
9713                            sense_format,
9714                            /*current_error*/ 1,
9715                            /*sense_key*/ SSD_KEY_NO_SENSE,
9716                            /*asc*/ 0x00,
9717                            /*ascq*/ 0x00,
9718                            SSD_ELEM_NONE);
9719
9720         ctsio->scsi_status = SCSI_STATUS_OK;
9721
9722         /*
9723          * We report 0 for the sense length, because we aren't doing
9724          * autosense in this case.  We're reporting sense as parameter data.
9725          */
9726         ctsio->sense_len = 0;
9727         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9728         ctsio->be_move_done = ctl_config_move_done;
9729         ctl_datamove((union ctl_io *)ctsio);
9730
9731         return (CTL_RETVAL_COMPLETE);
9732 }
9733
9734 int
9735 ctl_tur(struct ctl_scsiio *ctsio)
9736 {
9737         struct ctl_lun *lun;
9738
9739         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9740
9741         CTL_DEBUG_PRINT(("ctl_tur\n"));
9742
9743         if (lun == NULL)
9744                 return (EINVAL);
9745
9746         ctsio->scsi_status = SCSI_STATUS_OK;
9747         ctsio->io_hdr.status = CTL_SUCCESS;
9748
9749         ctl_done((union ctl_io *)ctsio);
9750
9751         return (CTL_RETVAL_COMPLETE);
9752 }
9753
9754 #ifdef notyet
9755 static int
9756 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9757 {
9758
9759 }
9760 #endif
9761
9762 static int
9763 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9764 {
9765         struct scsi_vpd_supported_pages *pages;
9766         int sup_page_size;
9767         struct ctl_lun *lun;
9768
9769         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9770
9771         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9772             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9773         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9774         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9775         ctsio->kern_sg_entries = 0;
9776
9777         if (sup_page_size < alloc_len) {
9778                 ctsio->residual = alloc_len - sup_page_size;
9779                 ctsio->kern_data_len = sup_page_size;
9780                 ctsio->kern_total_len = sup_page_size;
9781         } else {
9782                 ctsio->residual = 0;
9783                 ctsio->kern_data_len = alloc_len;
9784                 ctsio->kern_total_len = alloc_len;
9785         }
9786         ctsio->kern_data_resid = 0;
9787         ctsio->kern_rel_offset = 0;
9788         ctsio->kern_sg_entries = 0;
9789
9790         /*
9791          * The control device is always connected.  The disk device, on the
9792          * other hand, may not be online all the time.  Need to change this
9793          * to figure out whether the disk device is actually online or not.
9794          */
9795         if (lun != NULL)
9796                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9797                                 lun->be_lun->lun_type;
9798         else
9799                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9800
9801         pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9802         /* Supported VPD pages */
9803         pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9804         /* Serial Number */
9805         pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9806         /* Device Identification */
9807         pages->page_list[2] = SVPD_DEVICE_ID;
9808         /* SCSI Ports */
9809         pages->page_list[3] = SVPD_SCSI_PORTS;
9810         /* Third-party Copy */
9811         pages->page_list[4] = SVPD_SCSI_TPC;
9812         /* Block limits */
9813         pages->page_list[5] = SVPD_BLOCK_LIMITS;
9814         /* Logical Block Provisioning */
9815         pages->page_list[6] = SVPD_LBP;
9816
9817         ctsio->scsi_status = SCSI_STATUS_OK;
9818
9819         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9820         ctsio->be_move_done = ctl_config_move_done;
9821         ctl_datamove((union ctl_io *)ctsio);
9822
9823         return (CTL_RETVAL_COMPLETE);
9824 }
9825
9826 static int
9827 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9828 {
9829         struct scsi_vpd_unit_serial_number *sn_ptr;
9830         struct ctl_lun *lun;
9831
9832         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9833
9834         ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9835         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9836         ctsio->kern_sg_entries = 0;
9837
9838         if (sizeof(*sn_ptr) < alloc_len) {
9839                 ctsio->residual = alloc_len - sizeof(*sn_ptr);
9840                 ctsio->kern_data_len = sizeof(*sn_ptr);
9841                 ctsio->kern_total_len = sizeof(*sn_ptr);
9842         } else {
9843                 ctsio->residual = 0;
9844                 ctsio->kern_data_len = alloc_len;
9845                 ctsio->kern_total_len = alloc_len;
9846         }
9847         ctsio->kern_data_resid = 0;
9848         ctsio->kern_rel_offset = 0;
9849         ctsio->kern_sg_entries = 0;
9850
9851         /*
9852          * The control device is always connected.  The disk device, on the
9853          * other hand, may not be online all the time.  Need to change this
9854          * to figure out whether the disk device is actually online or not.
9855          */
9856         if (lun != NULL)
9857                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9858                                   lun->be_lun->lun_type;
9859         else
9860                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9861
9862         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9863         sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9864         /*
9865          * If we don't have a LUN, we just leave the serial number as
9866          * all spaces.
9867          */
9868         memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9869         if (lun != NULL) {
9870                 strncpy((char *)sn_ptr->serial_num,
9871                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9872         }
9873         ctsio->scsi_status = SCSI_STATUS_OK;
9874
9875         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9876         ctsio->be_move_done = ctl_config_move_done;
9877         ctl_datamove((union ctl_io *)ctsio);
9878
9879         return (CTL_RETVAL_COMPLETE);
9880 }
9881
9882
9883 static int
9884 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9885 {
9886         struct scsi_vpd_device_id *devid_ptr;
9887         struct scsi_vpd_id_descriptor *desc;
9888         struct ctl_softc *ctl_softc;
9889         struct ctl_lun *lun;
9890         struct ctl_port *port;
9891         int data_len;
9892         uint8_t proto;
9893
9894         ctl_softc = control_softc;
9895
9896         port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9897         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9898
9899         data_len = sizeof(struct scsi_vpd_device_id) +
9900             sizeof(struct scsi_vpd_id_descriptor) +
9901                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9902             sizeof(struct scsi_vpd_id_descriptor) +
9903                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9904         if (lun && lun->lun_devid)
9905                 data_len += lun->lun_devid->len;
9906         if (port->port_devid)
9907                 data_len += port->port_devid->len;
9908         if (port->target_devid)
9909                 data_len += port->target_devid->len;
9910
9911         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9912         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9913         ctsio->kern_sg_entries = 0;
9914
9915         if (data_len < alloc_len) {
9916                 ctsio->residual = alloc_len - data_len;
9917                 ctsio->kern_data_len = data_len;
9918                 ctsio->kern_total_len = data_len;
9919         } else {
9920                 ctsio->residual = 0;
9921                 ctsio->kern_data_len = alloc_len;
9922                 ctsio->kern_total_len = alloc_len;
9923         }
9924         ctsio->kern_data_resid = 0;
9925         ctsio->kern_rel_offset = 0;
9926         ctsio->kern_sg_entries = 0;
9927
9928         /*
9929          * The control device is always connected.  The disk device, on the
9930          * other hand, may not be online all the time.
9931          */
9932         if (lun != NULL)
9933                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9934                                      lun->be_lun->lun_type;
9935         else
9936                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9937         devid_ptr->page_code = SVPD_DEVICE_ID;
9938         scsi_ulto2b(data_len - 4, devid_ptr->length);
9939
9940         if (port->port_type == CTL_PORT_FC)
9941                 proto = SCSI_PROTO_FC << 4;
9942         else if (port->port_type == CTL_PORT_ISCSI)
9943                 proto = SCSI_PROTO_ISCSI << 4;
9944         else
9945                 proto = SCSI_PROTO_SPI << 4;
9946         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9947
9948         /*
9949          * We're using a LUN association here.  i.e., this device ID is a
9950          * per-LUN identifier.
9951          */
9952         if (lun && lun->lun_devid) {
9953                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9954                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9955                     lun->lun_devid->len);
9956         }
9957
9958         /*
9959          * This is for the WWPN which is a port association.
9960          */
9961         if (port->port_devid) {
9962                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9963                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9964                     port->port_devid->len);
9965         }
9966
9967         /*
9968          * This is for the Relative Target Port(type 4h) identifier
9969          */
9970         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9971         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9972             SVPD_ID_TYPE_RELTARG;
9973         desc->length = 4;
9974         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9975         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9976             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9977
9978         /*
9979          * This is for the Target Port Group(type 5h) identifier
9980          */
9981         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9982         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9983             SVPD_ID_TYPE_TPORTGRP;
9984         desc->length = 4;
9985         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9986             &desc->identifier[2]);
9987         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9988             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9989
9990         /*
9991          * This is for the Target identifier
9992          */
9993         if (port->target_devid) {
9994                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9995         }
9996
9997         ctsio->scsi_status = SCSI_STATUS_OK;
9998         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9999         ctsio->be_move_done = ctl_config_move_done;
10000         ctl_datamove((union ctl_io *)ctsio);
10001
10002         return (CTL_RETVAL_COMPLETE);
10003 }
10004
10005 static int
10006 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10007 {
10008         struct ctl_softc *softc = control_softc;
10009         struct scsi_vpd_scsi_ports *sp;
10010         struct scsi_vpd_port_designation *pd;
10011         struct scsi_vpd_port_designation_cont *pdc;
10012         struct ctl_lun *lun;
10013         struct ctl_port *port;
10014         int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10015         int num_target_port_groups, single;
10016
10017         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10018
10019         single = ctl_is_single;
10020         if (single)
10021                 num_target_port_groups = 1;
10022         else
10023                 num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10024         num_target_ports = 0;
10025         iid_len = 0;
10026         id_len = 0;
10027         mtx_lock(&softc->ctl_lock);
10028         STAILQ_FOREACH(port, &softc->port_list, links) {
10029                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10030                         continue;
10031                 if (lun != NULL &&
10032                     ctl_map_lun_back(port->targ_port, lun->lun) >=
10033                     CTL_MAX_LUNS)
10034                         continue;
10035                 num_target_ports++;
10036                 if (port->init_devid)
10037                         iid_len += port->init_devid->len;
10038                 if (port->port_devid)
10039                         id_len += port->port_devid->len;
10040         }
10041         mtx_unlock(&softc->ctl_lock);
10042
10043         data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10044             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10045              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10046         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10047         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10048         ctsio->kern_sg_entries = 0;
10049
10050         if (data_len < alloc_len) {
10051                 ctsio->residual = alloc_len - data_len;
10052                 ctsio->kern_data_len = data_len;
10053                 ctsio->kern_total_len = data_len;
10054         } else {
10055                 ctsio->residual = 0;
10056                 ctsio->kern_data_len = alloc_len;
10057                 ctsio->kern_total_len = alloc_len;
10058         }
10059         ctsio->kern_data_resid = 0;
10060         ctsio->kern_rel_offset = 0;
10061         ctsio->kern_sg_entries = 0;
10062
10063         /*
10064          * The control device is always connected.  The disk device, on the
10065          * other hand, may not be online all the time.  Need to change this
10066          * to figure out whether the disk device is actually online or not.
10067          */
10068         if (lun != NULL)
10069                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10070                                   lun->be_lun->lun_type;
10071         else
10072                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10073
10074         sp->page_code = SVPD_SCSI_PORTS;
10075         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10076             sp->page_length);
10077         pd = &sp->design[0];
10078
10079         mtx_lock(&softc->ctl_lock);
10080         if (softc->flags & CTL_FLAG_MASTER_SHELF)
10081                 pg = 0;
10082         else
10083                 pg = 1;
10084         for (g = 0; g < num_target_port_groups; g++) {
10085                 STAILQ_FOREACH(port, &softc->port_list, links) {
10086                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10087                                 continue;
10088                         if (lun != NULL &&
10089                             ctl_map_lun_back(port->targ_port, lun->lun) >=
10090                             CTL_MAX_LUNS)
10091                                 continue;
10092                         p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10093                         scsi_ulto2b(p, pd->relative_port_id);
10094                         if (port->init_devid && g == pg) {
10095                                 iid_len = port->init_devid->len;
10096                                 memcpy(pd->initiator_transportid,
10097                                     port->init_devid->data, port->init_devid->len);
10098                         } else
10099                                 iid_len = 0;
10100                         scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10101                         pdc = (struct scsi_vpd_port_designation_cont *)
10102                             (&pd->initiator_transportid[iid_len]);
10103                         if (port->port_devid && g == pg) {
10104                                 id_len = port->port_devid->len;
10105                                 memcpy(pdc->target_port_descriptors,
10106                                     port->port_devid->data, port->port_devid->len);
10107                         } else
10108                                 id_len = 0;
10109                         scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10110                         pd = (struct scsi_vpd_port_designation *)
10111                             ((uint8_t *)pdc->target_port_descriptors + id_len);
10112                 }
10113         }
10114         mtx_unlock(&softc->ctl_lock);
10115
10116         ctsio->scsi_status = SCSI_STATUS_OK;
10117         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10118         ctsio->be_move_done = ctl_config_move_done;
10119         ctl_datamove((union ctl_io *)ctsio);
10120
10121         return (CTL_RETVAL_COMPLETE);
10122 }
10123
10124 static int
10125 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10126 {
10127         struct scsi_vpd_block_limits *bl_ptr;
10128         struct ctl_lun *lun;
10129         int bs;
10130
10131         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10132
10133         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10134         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10135         ctsio->kern_sg_entries = 0;
10136
10137         if (sizeof(*bl_ptr) < alloc_len) {
10138                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
10139                 ctsio->kern_data_len = sizeof(*bl_ptr);
10140                 ctsio->kern_total_len = sizeof(*bl_ptr);
10141         } else {
10142                 ctsio->residual = 0;
10143                 ctsio->kern_data_len = alloc_len;
10144                 ctsio->kern_total_len = alloc_len;
10145         }
10146         ctsio->kern_data_resid = 0;
10147         ctsio->kern_rel_offset = 0;
10148         ctsio->kern_sg_entries = 0;
10149
10150         /*
10151          * The control device is always connected.  The disk device, on the
10152          * other hand, may not be online all the time.  Need to change this
10153          * to figure out whether the disk device is actually online or not.
10154          */
10155         if (lun != NULL)
10156                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10157                                   lun->be_lun->lun_type;
10158         else
10159                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10160
10161         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10162         scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
10163         bl_ptr->max_cmp_write_len = 0xff;
10164         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10165         if (lun != NULL) {
10166                 bs = lun->be_lun->blocksize;
10167                 scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10168                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10169                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10170                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10171                 }
10172         }
10173         scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10174
10175         ctsio->scsi_status = SCSI_STATUS_OK;
10176         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10177         ctsio->be_move_done = ctl_config_move_done;
10178         ctl_datamove((union ctl_io *)ctsio);
10179
10180         return (CTL_RETVAL_COMPLETE);
10181 }
10182
10183 static int
10184 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10185 {
10186         struct scsi_vpd_logical_block_prov *lbp_ptr;
10187         struct ctl_lun *lun;
10188
10189         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10190
10191         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10192         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10193         ctsio->kern_sg_entries = 0;
10194
10195         if (sizeof(*lbp_ptr) < alloc_len) {
10196                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10197                 ctsio->kern_data_len = sizeof(*lbp_ptr);
10198                 ctsio->kern_total_len = sizeof(*lbp_ptr);
10199         } else {
10200                 ctsio->residual = 0;
10201                 ctsio->kern_data_len = alloc_len;
10202                 ctsio->kern_total_len = alloc_len;
10203         }
10204         ctsio->kern_data_resid = 0;
10205         ctsio->kern_rel_offset = 0;
10206         ctsio->kern_sg_entries = 0;
10207
10208         /*
10209          * The control device is always connected.  The disk device, on the
10210          * other hand, may not be online all the time.  Need to change this
10211          * to figure out whether the disk device is actually online or not.
10212          */
10213         if (lun != NULL)
10214                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10215                                   lun->be_lun->lun_type;
10216         else
10217                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10218
10219         lbp_ptr->page_code = SVPD_LBP;
10220         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
10221                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10;
10222
10223         ctsio->scsi_status = SCSI_STATUS_OK;
10224         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10225         ctsio->be_move_done = ctl_config_move_done;
10226         ctl_datamove((union ctl_io *)ctsio);
10227
10228         return (CTL_RETVAL_COMPLETE);
10229 }
10230
10231 static int
10232 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10233 {
10234         struct scsi_inquiry *cdb;
10235         struct ctl_lun *lun;
10236         int alloc_len, retval;
10237
10238         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10239         cdb = (struct scsi_inquiry *)ctsio->cdb;
10240
10241         retval = CTL_RETVAL_COMPLETE;
10242
10243         alloc_len = scsi_2btoul(cdb->length);
10244
10245         switch (cdb->page_code) {
10246         case SVPD_SUPPORTED_PAGES:
10247                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10248                 break;
10249         case SVPD_UNIT_SERIAL_NUMBER:
10250                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10251                 break;
10252         case SVPD_DEVICE_ID:
10253                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10254                 break;
10255         case SVPD_SCSI_PORTS:
10256                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10257                 break;
10258         case SVPD_SCSI_TPC:
10259                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10260                 break;
10261         case SVPD_BLOCK_LIMITS:
10262                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10263                 break;
10264         case SVPD_LBP:
10265                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10266                 break;
10267         default:
10268                 ctl_set_invalid_field(ctsio,
10269                                       /*sks_valid*/ 1,
10270                                       /*command*/ 1,
10271                                       /*field*/ 2,
10272                                       /*bit_valid*/ 0,
10273                                       /*bit*/ 0);
10274                 ctl_done((union ctl_io *)ctsio);
10275                 retval = CTL_RETVAL_COMPLETE;
10276                 break;
10277         }
10278
10279         return (retval);
10280 }
10281
10282 static int
10283 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10284 {
10285         struct scsi_inquiry_data *inq_ptr;
10286         struct scsi_inquiry *cdb;
10287         struct ctl_softc *ctl_softc;
10288         struct ctl_lun *lun;
10289         char *val;
10290         uint32_t alloc_len;
10291         ctl_port_type port_type;
10292
10293         ctl_softc = control_softc;
10294
10295         /*
10296          * Figure out whether we're talking to a Fibre Channel port or not.
10297          * We treat the ioctl front end, and any SCSI adapters, as packetized
10298          * SCSI front ends.
10299          */
10300         port_type = ctl_softc->ctl_ports[
10301             ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10302         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10303                 port_type = CTL_PORT_SCSI;
10304
10305         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10306         cdb = (struct scsi_inquiry *)ctsio->cdb;
10307         alloc_len = scsi_2btoul(cdb->length);
10308
10309         /*
10310          * We malloc the full inquiry data size here and fill it
10311          * in.  If the user only asks for less, we'll give him
10312          * that much.
10313          */
10314         ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
10315         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10316         ctsio->kern_sg_entries = 0;
10317         ctsio->kern_data_resid = 0;
10318         ctsio->kern_rel_offset = 0;
10319
10320         if (sizeof(*inq_ptr) < alloc_len) {
10321                 ctsio->residual = alloc_len - sizeof(*inq_ptr);
10322                 ctsio->kern_data_len = sizeof(*inq_ptr);
10323                 ctsio->kern_total_len = sizeof(*inq_ptr);
10324         } else {
10325                 ctsio->residual = 0;
10326                 ctsio->kern_data_len = alloc_len;
10327                 ctsio->kern_total_len = alloc_len;
10328         }
10329
10330         /*
10331          * If we have a LUN configured, report it as connected.  Otherwise,
10332          * report that it is offline or no device is supported, depending
10333          * on the value of inquiry_pq_no_lun.
10334          *
10335          * According to the spec (SPC-4 r34), the peripheral qualifier
10336          * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10337          *
10338          * "A peripheral device having the specified peripheral device type
10339          * is not connected to this logical unit. However, the device
10340          * server is capable of supporting the specified peripheral device
10341          * type on this logical unit."
10342          *
10343          * According to the same spec, the peripheral qualifier
10344          * SID_QUAL_BAD_LU (011b) is used in this scenario:
10345          *
10346          * "The device server is not capable of supporting a peripheral
10347          * device on this logical unit. For this peripheral qualifier the
10348          * peripheral device type shall be set to 1Fh. All other peripheral
10349          * device type values are reserved for this peripheral qualifier."
10350          *
10351          * Given the text, it would seem that we probably want to report that
10352          * the LUN is offline here.  There is no LUN connected, but we can
10353          * support a LUN at the given LUN number.
10354          *
10355          * In the real world, though, it sounds like things are a little
10356          * different:
10357          *
10358          * - Linux, when presented with a LUN with the offline peripheral
10359          *   qualifier, will create an sg driver instance for it.  So when
10360          *   you attach it to CTL, you wind up with a ton of sg driver
10361          *   instances.  (One for every LUN that Linux bothered to probe.)
10362          *   Linux does this despite the fact that it issues a REPORT LUNs
10363          *   to LUN 0 to get the inventory of supported LUNs.
10364          *
10365          * - There is other anecdotal evidence (from Emulex folks) about
10366          *   arrays that use the offline peripheral qualifier for LUNs that
10367          *   are on the "passive" path in an active/passive array.
10368          *
10369          * So the solution is provide a hopefully reasonable default
10370          * (return bad/no LUN) and allow the user to change the behavior
10371          * with a tunable/sysctl variable.
10372          */
10373         if (lun != NULL)
10374                 inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10375                                   lun->be_lun->lun_type;
10376         else if (ctl_softc->inquiry_pq_no_lun == 0)
10377                 inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10378         else
10379                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10380
10381         /* RMB in byte 2 is 0 */
10382         inq_ptr->version = SCSI_REV_SPC4;
10383
10384         /*
10385          * According to SAM-3, even if a device only supports a single
10386          * level of LUN addressing, it should still set the HISUP bit:
10387          *
10388          * 4.9.1 Logical unit numbers overview
10389          *
10390          * All logical unit number formats described in this standard are
10391          * hierarchical in structure even when only a single level in that
10392          * hierarchy is used. The HISUP bit shall be set to one in the
10393          * standard INQUIRY data (see SPC-2) when any logical unit number
10394          * format described in this standard is used.  Non-hierarchical
10395          * formats are outside the scope of this standard.
10396          *
10397          * Therefore we set the HiSup bit here.
10398          *
10399          * The reponse format is 2, per SPC-3.
10400          */
10401         inq_ptr->response_format = SID_HiSup | 2;
10402
10403         inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
10404         CTL_DEBUG_PRINT(("additional_length = %d\n",
10405                          inq_ptr->additional_length));
10406
10407         inq_ptr->spc3_flags = SPC3_SID_3PC;
10408         if (!ctl_is_single)
10409                 inq_ptr->spc3_flags |= SPC3_SID_TPGS_IMPLICIT;
10410         /* 16 bit addressing */
10411         if (port_type == CTL_PORT_SCSI)
10412                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10413         /* XXX set the SID_MultiP bit here if we're actually going to
10414            respond on multiple ports */
10415         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10416
10417         /* 16 bit data bus, synchronous transfers */
10418         if (port_type == CTL_PORT_SCSI)
10419                 inq_ptr->flags = SID_WBus16 | SID_Sync;
10420         /*
10421          * XXX KDM do we want to support tagged queueing on the control
10422          * device at all?
10423          */
10424         if ((lun == NULL)
10425          || (lun->be_lun->lun_type != T_PROCESSOR))
10426                 inq_ptr->flags |= SID_CmdQue;
10427         /*
10428          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10429          * We have 8 bytes for the vendor name, and 16 bytes for the device
10430          * name and 4 bytes for the revision.
10431          */
10432         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10433             "vendor")) == NULL) {
10434                 strcpy(inq_ptr->vendor, CTL_VENDOR);
10435         } else {
10436                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10437                 strncpy(inq_ptr->vendor, val,
10438                     min(sizeof(inq_ptr->vendor), strlen(val)));
10439         }
10440         if (lun == NULL) {
10441                 strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10442         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10443                 switch (lun->be_lun->lun_type) {
10444                 case T_DIRECT:
10445                         strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10446                         break;
10447                 case T_PROCESSOR:
10448                         strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
10449                         break;
10450                 default:
10451                         strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
10452                         break;
10453                 }
10454         } else {
10455                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10456                 strncpy(inq_ptr->product, val,
10457                     min(sizeof(inq_ptr->product), strlen(val)));
10458         }
10459
10460         /*
10461          * XXX make this a macro somewhere so it automatically gets
10462          * incremented when we make changes.
10463          */
10464         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10465             "revision")) == NULL) {
10466                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10467         } else {
10468                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10469                 strncpy(inq_ptr->revision, val,
10470                     min(sizeof(inq_ptr->revision), strlen(val)));
10471         }
10472
10473         /*
10474          * For parallel SCSI, we support double transition and single
10475          * transition clocking.  We also support QAS (Quick Arbitration
10476          * and Selection) and Information Unit transfers on both the
10477          * control and array devices.
10478          */
10479         if (port_type == CTL_PORT_SCSI)
10480                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10481                                     SID_SPI_IUS;
10482
10483         /* SAM-5 (no version claimed) */
10484         scsi_ulto2b(0x00A0, inq_ptr->version1);
10485         /* SPC-4 (no version claimed) */
10486         scsi_ulto2b(0x0460, inq_ptr->version2);
10487         if (port_type == CTL_PORT_FC) {
10488                 /* FCP-2 ANSI INCITS.350:2003 */
10489                 scsi_ulto2b(0x0917, inq_ptr->version3);
10490         } else if (port_type == CTL_PORT_SCSI) {
10491                 /* SPI-4 ANSI INCITS.362:200x */
10492                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10493         } else if (port_type == CTL_PORT_ISCSI) {
10494                 /* iSCSI (no version claimed) */
10495                 scsi_ulto2b(0x0960, inq_ptr->version3);
10496         } else if (port_type == CTL_PORT_SAS) {
10497                 /* SAS (no version claimed) */
10498                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10499         }
10500
10501         if (lun == NULL) {
10502                 /* SBC-3 (no version claimed) */
10503                 scsi_ulto2b(0x04C0, inq_ptr->version4);
10504         } else {
10505                 switch (lun->be_lun->lun_type) {
10506                 case T_DIRECT:
10507                         /* SBC-3 (no version claimed) */
10508                         scsi_ulto2b(0x04C0, inq_ptr->version4);
10509                         break;
10510                 case T_PROCESSOR:
10511                 default:
10512                         break;
10513                 }
10514         }
10515
10516         ctsio->scsi_status = SCSI_STATUS_OK;
10517         if (ctsio->kern_data_len > 0) {
10518                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10519                 ctsio->be_move_done = ctl_config_move_done;
10520                 ctl_datamove((union ctl_io *)ctsio);
10521         } else {
10522                 ctsio->io_hdr.status = CTL_SUCCESS;
10523                 ctl_done((union ctl_io *)ctsio);
10524         }
10525
10526         return (CTL_RETVAL_COMPLETE);
10527 }
10528
10529 int
10530 ctl_inquiry(struct ctl_scsiio *ctsio)
10531 {
10532         struct scsi_inquiry *cdb;
10533         int retval;
10534
10535         cdb = (struct scsi_inquiry *)ctsio->cdb;
10536
10537         retval = 0;
10538
10539         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10540
10541         /*
10542          * Right now, we don't support the CmdDt inquiry information.
10543          * This would be nice to support in the future.  When we do
10544          * support it, we should change this test so that it checks to make
10545          * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10546          */
10547 #ifdef notyet
10548         if (((cdb->byte2 & SI_EVPD)
10549          && (cdb->byte2 & SI_CMDDT)))
10550 #endif
10551         if (cdb->byte2 & SI_CMDDT) {
10552                 /*
10553                  * Point to the SI_CMDDT bit.  We might change this
10554                  * when we support SI_CMDDT, but since both bits would be
10555                  * "wrong", this should probably just stay as-is then.
10556                  */
10557                 ctl_set_invalid_field(ctsio,
10558                                       /*sks_valid*/ 1,
10559                                       /*command*/ 1,
10560                                       /*field*/ 1,
10561                                       /*bit_valid*/ 1,
10562                                       /*bit*/ 1);
10563                 ctl_done((union ctl_io *)ctsio);
10564                 return (CTL_RETVAL_COMPLETE);
10565         }
10566         if (cdb->byte2 & SI_EVPD)
10567                 retval = ctl_inquiry_evpd(ctsio);
10568 #ifdef notyet
10569         else if (cdb->byte2 & SI_CMDDT)
10570                 retval = ctl_inquiry_cmddt(ctsio);
10571 #endif
10572         else
10573                 retval = ctl_inquiry_std(ctsio);
10574
10575         return (retval);
10576 }
10577
10578 /*
10579  * For known CDB types, parse the LBA and length.
10580  */
10581 static int
10582 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10583 {
10584         if (io->io_hdr.io_type != CTL_IO_SCSI)
10585                 return (1);
10586
10587         switch (io->scsiio.cdb[0]) {
10588         case COMPARE_AND_WRITE: {
10589                 struct scsi_compare_and_write *cdb;
10590
10591                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10592
10593                 *lba = scsi_8btou64(cdb->addr);
10594                 *len = cdb->length;
10595                 break;
10596         }
10597         case READ_6:
10598         case WRITE_6: {
10599                 struct scsi_rw_6 *cdb;
10600
10601                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10602
10603                 *lba = scsi_3btoul(cdb->addr);
10604                 /* only 5 bits are valid in the most significant address byte */
10605                 *lba &= 0x1fffff;
10606                 *len = cdb->length;
10607                 break;
10608         }
10609         case READ_10:
10610         case WRITE_10: {
10611                 struct scsi_rw_10 *cdb;
10612
10613                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10614
10615                 *lba = scsi_4btoul(cdb->addr);
10616                 *len = scsi_2btoul(cdb->length);
10617                 break;
10618         }
10619         case WRITE_VERIFY_10: {
10620                 struct scsi_write_verify_10 *cdb;
10621
10622                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10623
10624                 *lba = scsi_4btoul(cdb->addr);
10625                 *len = scsi_2btoul(cdb->length);
10626                 break;
10627         }
10628         case READ_12:
10629         case WRITE_12: {
10630                 struct scsi_rw_12 *cdb;
10631
10632                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10633
10634                 *lba = scsi_4btoul(cdb->addr);
10635                 *len = scsi_4btoul(cdb->length);
10636                 break;
10637         }
10638         case WRITE_VERIFY_12: {
10639                 struct scsi_write_verify_12 *cdb;
10640
10641                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10642
10643                 *lba = scsi_4btoul(cdb->addr);
10644                 *len = scsi_4btoul(cdb->length);
10645                 break;
10646         }
10647         case READ_16:
10648         case WRITE_16: {
10649                 struct scsi_rw_16 *cdb;
10650
10651                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10652
10653                 *lba = scsi_8btou64(cdb->addr);
10654                 *len = scsi_4btoul(cdb->length);
10655                 break;
10656         }
10657         case WRITE_VERIFY_16: {
10658                 struct scsi_write_verify_16 *cdb;
10659
10660                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10661
10662
10663                 *lba = scsi_8btou64(cdb->addr);
10664                 *len = scsi_4btoul(cdb->length);
10665                 break;
10666         }
10667         case WRITE_SAME_10: {
10668                 struct scsi_write_same_10 *cdb;
10669
10670                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10671
10672                 *lba = scsi_4btoul(cdb->addr);
10673                 *len = scsi_2btoul(cdb->length);
10674                 break;
10675         }
10676         case WRITE_SAME_16: {
10677                 struct scsi_write_same_16 *cdb;
10678
10679                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10680
10681                 *lba = scsi_8btou64(cdb->addr);
10682                 *len = scsi_4btoul(cdb->length);
10683                 break;
10684         }
10685         case VERIFY_10: {
10686                 struct scsi_verify_10 *cdb;
10687
10688                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10689
10690                 *lba = scsi_4btoul(cdb->addr);
10691                 *len = scsi_2btoul(cdb->length);
10692                 break;
10693         }
10694         case VERIFY_12: {
10695                 struct scsi_verify_12 *cdb;
10696
10697                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10698
10699                 *lba = scsi_4btoul(cdb->addr);
10700                 *len = scsi_4btoul(cdb->length);
10701                 break;
10702         }
10703         case VERIFY_16: {
10704                 struct scsi_verify_16 *cdb;
10705
10706                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10707
10708                 *lba = scsi_8btou64(cdb->addr);
10709                 *len = scsi_4btoul(cdb->length);
10710                 break;
10711         }
10712         default:
10713                 return (1);
10714                 break; /* NOTREACHED */
10715         }
10716
10717         return (0);
10718 }
10719
10720 static ctl_action
10721 ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10722 {
10723         uint64_t endlba1, endlba2;
10724
10725         endlba1 = lba1 + len1 - 1;
10726         endlba2 = lba2 + len2 - 1;
10727
10728         if ((endlba1 < lba2)
10729          || (endlba2 < lba1))
10730                 return (CTL_ACTION_PASS);
10731         else
10732                 return (CTL_ACTION_BLOCK);
10733 }
10734
10735 static ctl_action
10736 ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10737 {
10738         uint64_t lba1, lba2;
10739         uint32_t len1, len2;
10740         int retval;
10741
10742         retval = ctl_get_lba_len(io1, &lba1, &len1);
10743         if (retval != 0)
10744                 return (CTL_ACTION_ERROR);
10745
10746         retval = ctl_get_lba_len(io2, &lba2, &len2);
10747         if (retval != 0)
10748                 return (CTL_ACTION_ERROR);
10749
10750         return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10751 }
10752
10753 static ctl_action
10754 ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10755 {
10756         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10757         ctl_serialize_action *serialize_row;
10758
10759         /*
10760          * The initiator attempted multiple untagged commands at the same
10761          * time.  Can't do that.
10762          */
10763         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10764          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10765          && ((pending_io->io_hdr.nexus.targ_port ==
10766               ooa_io->io_hdr.nexus.targ_port)
10767           && (pending_io->io_hdr.nexus.initid.id ==
10768               ooa_io->io_hdr.nexus.initid.id))
10769          && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10770                 return (CTL_ACTION_OVERLAP);
10771
10772         /*
10773          * The initiator attempted to send multiple tagged commands with
10774          * the same ID.  (It's fine if different initiators have the same
10775          * tag ID.)
10776          *
10777          * Even if all of those conditions are true, we don't kill the I/O
10778          * if the command ahead of us has been aborted.  We won't end up
10779          * sending it to the FETD, and it's perfectly legal to resend a
10780          * command with the same tag number as long as the previous
10781          * instance of this tag number has been aborted somehow.
10782          */
10783         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10784          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10785          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10786          && ((pending_io->io_hdr.nexus.targ_port ==
10787               ooa_io->io_hdr.nexus.targ_port)
10788           && (pending_io->io_hdr.nexus.initid.id ==
10789               ooa_io->io_hdr.nexus.initid.id))
10790          && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10791                 return (CTL_ACTION_OVERLAP_TAG);
10792
10793         /*
10794          * If we get a head of queue tag, SAM-3 says that we should
10795          * immediately execute it.
10796          *
10797          * What happens if this command would normally block for some other
10798          * reason?  e.g. a request sense with a head of queue tag
10799          * immediately after a write.  Normally that would block, but this
10800          * will result in its getting executed immediately...
10801          *
10802          * We currently return "pass" instead of "skip", so we'll end up
10803          * going through the rest of the queue to check for overlapped tags.
10804          *
10805          * XXX KDM check for other types of blockage first??
10806          */
10807         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10808                 return (CTL_ACTION_PASS);
10809
10810         /*
10811          * Ordered tags have to block until all items ahead of them
10812          * have completed.  If we get called with an ordered tag, we always
10813          * block, if something else is ahead of us in the queue.
10814          */
10815         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10816                 return (CTL_ACTION_BLOCK);
10817
10818         /*
10819          * Simple tags get blocked until all head of queue and ordered tags
10820          * ahead of them have completed.  I'm lumping untagged commands in
10821          * with simple tags here.  XXX KDM is that the right thing to do?
10822          */
10823         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10824           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10825          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10826           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10827                 return (CTL_ACTION_BLOCK);
10828
10829         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio);
10830         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio);
10831
10832         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10833
10834         switch (serialize_row[pending_entry->seridx]) {
10835         case CTL_SER_BLOCK:
10836                 return (CTL_ACTION_BLOCK);
10837                 break; /* NOTREACHED */
10838         case CTL_SER_EXTENT:
10839                 return (ctl_extent_check(pending_io, ooa_io));
10840                 break; /* NOTREACHED */
10841         case CTL_SER_PASS:
10842                 return (CTL_ACTION_PASS);
10843                 break; /* NOTREACHED */
10844         case CTL_SER_SKIP:
10845                 return (CTL_ACTION_SKIP);
10846                 break;
10847         default:
10848                 panic("invalid serialization value %d",
10849                       serialize_row[pending_entry->seridx]);
10850                 break; /* NOTREACHED */
10851         }
10852
10853         return (CTL_ACTION_ERROR);
10854 }
10855
10856 /*
10857  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10858  * Assumptions:
10859  * - pending_io is generally either incoming, or on the blocked queue
10860  * - starting I/O is the I/O we want to start the check with.
10861  */
10862 static ctl_action
10863 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10864               union ctl_io *starting_io)
10865 {
10866         union ctl_io *ooa_io;
10867         ctl_action action;
10868
10869         mtx_assert(&lun->lun_lock, MA_OWNED);
10870
10871         /*
10872          * Run back along the OOA queue, starting with the current
10873          * blocked I/O and going through every I/O before it on the
10874          * queue.  If starting_io is NULL, we'll just end up returning
10875          * CTL_ACTION_PASS.
10876          */
10877         for (ooa_io = starting_io; ooa_io != NULL;
10878              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10879              ooa_links)){
10880
10881                 /*
10882                  * This routine just checks to see whether
10883                  * cur_blocked is blocked by ooa_io, which is ahead
10884                  * of it in the queue.  It doesn't queue/dequeue
10885                  * cur_blocked.
10886                  */
10887                 action = ctl_check_for_blockage(pending_io, ooa_io);
10888                 switch (action) {
10889                 case CTL_ACTION_BLOCK:
10890                 case CTL_ACTION_OVERLAP:
10891                 case CTL_ACTION_OVERLAP_TAG:
10892                 case CTL_ACTION_SKIP:
10893                 case CTL_ACTION_ERROR:
10894                         return (action);
10895                         break; /* NOTREACHED */
10896                 case CTL_ACTION_PASS:
10897                         break;
10898                 default:
10899                         panic("invalid action %d", action);
10900                         break;  /* NOTREACHED */
10901                 }
10902         }
10903
10904         return (CTL_ACTION_PASS);
10905 }
10906
10907 /*
10908  * Assumptions:
10909  * - An I/O has just completed, and has been removed from the per-LUN OOA
10910  *   queue, so some items on the blocked queue may now be unblocked.
10911  */
10912 static int
10913 ctl_check_blocked(struct ctl_lun *lun)
10914 {
10915         union ctl_io *cur_blocked, *next_blocked;
10916
10917         mtx_assert(&lun->lun_lock, MA_OWNED);
10918
10919         /*
10920          * Run forward from the head of the blocked queue, checking each
10921          * entry against the I/Os prior to it on the OOA queue to see if
10922          * there is still any blockage.
10923          *
10924          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10925          * with our removing a variable on it while it is traversing the
10926          * list.
10927          */
10928         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10929              cur_blocked != NULL; cur_blocked = next_blocked) {
10930                 union ctl_io *prev_ooa;
10931                 ctl_action action;
10932
10933                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10934                                                           blocked_links);
10935
10936                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10937                                                       ctl_ooaq, ooa_links);
10938
10939                 /*
10940                  * If cur_blocked happens to be the first item in the OOA
10941                  * queue now, prev_ooa will be NULL, and the action
10942                  * returned will just be CTL_ACTION_PASS.
10943                  */
10944                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10945
10946                 switch (action) {
10947                 case CTL_ACTION_BLOCK:
10948                         /* Nothing to do here, still blocked */
10949                         break;
10950                 case CTL_ACTION_OVERLAP:
10951                 case CTL_ACTION_OVERLAP_TAG:
10952                         /*
10953                          * This shouldn't happen!  In theory we've already
10954                          * checked this command for overlap...
10955                          */
10956                         break;
10957                 case CTL_ACTION_PASS:
10958                 case CTL_ACTION_SKIP: {
10959                         struct ctl_softc *softc;
10960                         const struct ctl_cmd_entry *entry;
10961                         uint32_t initidx;
10962                         int isc_retval;
10963
10964                         /*
10965                          * The skip case shouldn't happen, this transaction
10966                          * should have never made it onto the blocked queue.
10967                          */
10968                         /*
10969                          * This I/O is no longer blocked, we can remove it
10970                          * from the blocked queue.  Since this is a TAILQ
10971                          * (doubly linked list), we can do O(1) removals
10972                          * from any place on the list.
10973                          */
10974                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10975                                      blocked_links);
10976                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10977
10978                         if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10979                                 /*
10980                                  * Need to send IO back to original side to
10981                                  * run
10982                                  */
10983                                 union ctl_ha_msg msg_info;
10984
10985                                 msg_info.hdr.original_sc =
10986                                         cur_blocked->io_hdr.original_sc;
10987                                 msg_info.hdr.serializing_sc = cur_blocked;
10988                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
10989                                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10990                                      &msg_info, sizeof(msg_info), 0)) >
10991                                      CTL_HA_STATUS_SUCCESS) {
10992                                         printf("CTL:Check Blocked error from "
10993                                                "ctl_ha_msg_send %d\n",
10994                                                isc_retval);
10995                                 }
10996                                 break;
10997                         }
10998                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio);
10999                         softc = control_softc;
11000
11001                         initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11002
11003                         /*
11004                          * Check this I/O for LUN state changes that may
11005                          * have happened while this command was blocked.
11006                          * The LUN state may have been changed by a command
11007                          * ahead of us in the queue, so we need to re-check
11008                          * for any states that can be caused by SCSI
11009                          * commands.
11010                          */
11011                         if (ctl_scsiio_lun_check(softc, lun, entry,
11012                                                  &cur_blocked->scsiio) == 0) {
11013                                 cur_blocked->io_hdr.flags |=
11014                                                       CTL_FLAG_IS_WAS_ON_RTR;
11015                                 ctl_enqueue_rtr(cur_blocked);
11016                         } else
11017                                 ctl_done(cur_blocked);
11018                         break;
11019                 }
11020                 default:
11021                         /*
11022                          * This probably shouldn't happen -- we shouldn't
11023                          * get CTL_ACTION_ERROR, or anything else.
11024                          */
11025                         break;
11026                 }
11027         }
11028
11029         return (CTL_RETVAL_COMPLETE);
11030 }
11031
11032 /*
11033  * This routine (with one exception) checks LUN flags that can be set by
11034  * commands ahead of us in the OOA queue.  These flags have to be checked
11035  * when a command initially comes in, and when we pull a command off the
11036  * blocked queue and are preparing to execute it.  The reason we have to
11037  * check these flags for commands on the blocked queue is that the LUN
11038  * state may have been changed by a command ahead of us while we're on the
11039  * blocked queue.
11040  *
11041  * Ordering is somewhat important with these checks, so please pay
11042  * careful attention to the placement of any new checks.
11043  */
11044 static int
11045 ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11046     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11047 {
11048         int retval;
11049
11050         retval = 0;
11051
11052         mtx_assert(&lun->lun_lock, MA_OWNED);
11053
11054         /*
11055          * If this shelf is a secondary shelf controller, we have to reject
11056          * any media access commands.
11057          */
11058 #if 0
11059         /* No longer needed for HA */
11060         if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11061          && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11062                 ctl_set_lun_standby(ctsio);
11063                 retval = 1;
11064                 goto bailout;
11065         }
11066 #endif
11067
11068         /*
11069          * Check for a reservation conflict.  If this command isn't allowed
11070          * even on reserved LUNs, and if this initiator isn't the one who
11071          * reserved us, reject the command with a reservation conflict.
11072          */
11073         if ((lun->flags & CTL_LUN_RESERVED)
11074          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11075                 if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
11076                  || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
11077                  || (ctsio->io_hdr.nexus.targ_target.id !=
11078                      lun->rsv_nexus.targ_target.id)) {
11079                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11080                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
11081                         retval = 1;
11082                         goto bailout;
11083                 }
11084         }
11085
11086         if ( (lun->flags & CTL_LUN_PR_RESERVED)
11087          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11088                 uint32_t residx;
11089
11090                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11091                 /*
11092                  * if we aren't registered or it's a res holder type
11093                  * reservation and this isn't the res holder then set a
11094                  * conflict.
11095                  * NOTE: Commands which might be allowed on write exclusive
11096                  * type reservations are checked in the particular command
11097                  * for a conflict. Read and SSU are the only ones.
11098                  */
11099                 if (!lun->per_res[residx].registered
11100                  || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11101                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11102                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
11103                         retval = 1;
11104                         goto bailout;
11105                 }
11106
11107         }
11108
11109         if ((lun->flags & CTL_LUN_OFFLINE)
11110          && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11111                 ctl_set_lun_not_ready(ctsio);
11112                 retval = 1;
11113                 goto bailout;
11114         }
11115
11116         /*
11117          * If the LUN is stopped, see if this particular command is allowed
11118          * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11119          */
11120         if ((lun->flags & CTL_LUN_STOPPED)
11121          && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11122                 /* "Logical unit not ready, initializing cmd. required" */
11123                 ctl_set_lun_stopped(ctsio);
11124                 retval = 1;
11125                 goto bailout;
11126         }
11127
11128         if ((lun->flags & CTL_LUN_INOPERABLE)
11129          && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11130                 /* "Medium format corrupted" */
11131                 ctl_set_medium_format_corrupted(ctsio);
11132                 retval = 1;
11133                 goto bailout;
11134         }
11135
11136 bailout:
11137         return (retval);
11138
11139 }
11140
11141 static void
11142 ctl_failover_io(union ctl_io *io, int have_lock)
11143 {
11144         ctl_set_busy(&io->scsiio);
11145         ctl_done(io);
11146 }
11147
11148 static void
11149 ctl_failover(void)
11150 {
11151         struct ctl_lun *lun;
11152         struct ctl_softc *ctl_softc;
11153         union ctl_io *next_io, *pending_io;
11154         union ctl_io *io;
11155         int lun_idx;
11156         int i;
11157
11158         ctl_softc = control_softc;
11159
11160         mtx_lock(&ctl_softc->ctl_lock);
11161         /*
11162          * Remove any cmds from the other SC from the rtr queue.  These
11163          * will obviously only be for LUNs for which we're the primary.
11164          * We can't send status or get/send data for these commands.
11165          * Since they haven't been executed yet, we can just remove them.
11166          * We'll either abort them or delete them below, depending on
11167          * which HA mode we're in.
11168          */
11169 #ifdef notyet
11170         mtx_lock(&ctl_softc->queue_lock);
11171         for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11172              io != NULL; io = next_io) {
11173                 next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11174                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11175                         STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11176                                       ctl_io_hdr, links);
11177         }
11178         mtx_unlock(&ctl_softc->queue_lock);
11179 #endif
11180
11181         for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11182                 lun = ctl_softc->ctl_luns[lun_idx];
11183                 if (lun==NULL)
11184                         continue;
11185
11186                 /*
11187                  * Processor LUNs are primary on both sides.
11188                  * XXX will this always be true?
11189                  */
11190                 if (lun->be_lun->lun_type == T_PROCESSOR)
11191                         continue;
11192
11193                 if ((lun->flags & CTL_LUN_PRIMARY_SC)
11194                  && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11195                         printf("FAILOVER: primary lun %d\n", lun_idx);
11196                         /*
11197                          * Remove all commands from the other SC. First from the
11198                          * blocked queue then from the ooa queue. Once we have
11199                          * removed them. Call ctl_check_blocked to see if there
11200                          * is anything that can run.
11201                          */
11202                         for (io = (union ctl_io *)TAILQ_FIRST(
11203                              &lun->blocked_queue); io != NULL; io = next_io) {
11204
11205                                 next_io = (union ctl_io *)TAILQ_NEXT(
11206                                     &io->io_hdr, blocked_links);
11207
11208                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11209                                         TAILQ_REMOVE(&lun->blocked_queue,
11210                                                      &io->io_hdr,blocked_links);
11211                                         io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11212                                         TAILQ_REMOVE(&lun->ooa_queue,
11213                                                      &io->io_hdr, ooa_links);
11214
11215                                         ctl_free_io(io);
11216                                 }
11217                         }
11218
11219                         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11220                              io != NULL; io = next_io) {
11221
11222                                 next_io = (union ctl_io *)TAILQ_NEXT(
11223                                     &io->io_hdr, ooa_links);
11224
11225                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11226
11227                                         TAILQ_REMOVE(&lun->ooa_queue,
11228                                                 &io->io_hdr,
11229                                                 ooa_links);
11230
11231                                         ctl_free_io(io);
11232                                 }
11233                         }
11234                         ctl_check_blocked(lun);
11235                 } else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11236                         && (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11237
11238                         printf("FAILOVER: primary lun %d\n", lun_idx);
11239                         /*
11240                          * Abort all commands from the other SC.  We can't
11241                          * send status back for them now.  These should get
11242                          * cleaned up when they are completed or come out
11243                          * for a datamove operation.
11244                          */
11245                         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11246                              io != NULL; io = next_io) {
11247                                 next_io = (union ctl_io *)TAILQ_NEXT(
11248                                         &io->io_hdr, ooa_links);
11249
11250                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11251                                         io->io_hdr.flags |= CTL_FLAG_ABORT;
11252                         }
11253                 } else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11254                         && (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11255
11256                         printf("FAILOVER: secondary lun %d\n", lun_idx);
11257
11258                         lun->flags |= CTL_LUN_PRIMARY_SC;
11259
11260                         /*
11261                          * We send all I/O that was sent to this controller
11262                          * and redirected to the other side back with
11263                          * busy status, and have the initiator retry it.
11264                          * Figuring out how much data has been transferred,
11265                          * etc. and picking up where we left off would be
11266                          * very tricky.
11267                          *
11268                          * XXX KDM need to remove I/O from the blocked
11269                          * queue as well!
11270                          */
11271                         for (pending_io = (union ctl_io *)TAILQ_FIRST(
11272                              &lun->ooa_queue); pending_io != NULL;
11273                              pending_io = next_io) {
11274
11275                                 next_io =  (union ctl_io *)TAILQ_NEXT(
11276                                         &pending_io->io_hdr, ooa_links);
11277
11278                                 pending_io->io_hdr.flags &=
11279                                         ~CTL_FLAG_SENT_2OTHER_SC;
11280
11281                                 if (pending_io->io_hdr.flags &
11282                                     CTL_FLAG_IO_ACTIVE) {
11283                                         pending_io->io_hdr.flags |=
11284                                                 CTL_FLAG_FAILOVER;
11285                                 } else {
11286                                         ctl_set_busy(&pending_io->scsiio);
11287                                         ctl_done(pending_io);
11288                                 }
11289                         }
11290
11291                         /*
11292                          * Build Unit Attention
11293                          */
11294                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11295                                 lun->pending_ua[i] |=
11296                                                      CTL_UA_ASYM_ACC_CHANGE;
11297                         }
11298                 } else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11299                         && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11300                         printf("FAILOVER: secondary lun %d\n", lun_idx);
11301                         /*
11302                          * if the first io on the OOA is not on the RtR queue
11303                          * add it.
11304                          */
11305                         lun->flags |= CTL_LUN_PRIMARY_SC;
11306
11307                         pending_io = (union ctl_io *)TAILQ_FIRST(
11308                             &lun->ooa_queue);
11309                         if (pending_io==NULL) {
11310                                 printf("Nothing on OOA queue\n");
11311                                 continue;
11312                         }
11313
11314                         pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11315                         if ((pending_io->io_hdr.flags &
11316                              CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11317                                 pending_io->io_hdr.flags |=
11318                                     CTL_FLAG_IS_WAS_ON_RTR;
11319                                 ctl_enqueue_rtr(pending_io);
11320                         }
11321 #if 0
11322                         else
11323                         {
11324                                 printf("Tag 0x%04x is running\n",
11325                                       pending_io->scsiio.tag_num);
11326                         }
11327 #endif
11328
11329                         next_io = (union ctl_io *)TAILQ_NEXT(
11330                             &pending_io->io_hdr, ooa_links);
11331                         for (pending_io=next_io; pending_io != NULL;
11332                              pending_io = next_io) {
11333                                 pending_io->io_hdr.flags &=
11334                                     ~CTL_FLAG_SENT_2OTHER_SC;
11335                                 next_io = (union ctl_io *)TAILQ_NEXT(
11336                                         &pending_io->io_hdr, ooa_links);
11337                                 if (pending_io->io_hdr.flags &
11338                                     CTL_FLAG_IS_WAS_ON_RTR) {
11339 #if 0
11340                                         printf("Tag 0x%04x is running\n",
11341                                                 pending_io->scsiio.tag_num);
11342 #endif
11343                                         continue;
11344                                 }
11345
11346                                 switch (ctl_check_ooa(lun, pending_io,
11347                                     (union ctl_io *)TAILQ_PREV(
11348                                     &pending_io->io_hdr, ctl_ooaq,
11349                                     ooa_links))) {
11350
11351                                 case CTL_ACTION_BLOCK:
11352                                         TAILQ_INSERT_TAIL(&lun->blocked_queue,
11353                                                           &pending_io->io_hdr,
11354                                                           blocked_links);
11355                                         pending_io->io_hdr.flags |=
11356                                             CTL_FLAG_BLOCKED;
11357                                         break;
11358                                 case CTL_ACTION_PASS:
11359                                 case CTL_ACTION_SKIP:
11360                                         pending_io->io_hdr.flags |=
11361                                             CTL_FLAG_IS_WAS_ON_RTR;
11362                                         ctl_enqueue_rtr(pending_io);
11363                                         break;
11364                                 case CTL_ACTION_OVERLAP:
11365                                         ctl_set_overlapped_cmd(
11366                                             (struct ctl_scsiio *)pending_io);
11367                                         ctl_done(pending_io);
11368                                         break;
11369                                 case CTL_ACTION_OVERLAP_TAG:
11370                                         ctl_set_overlapped_tag(
11371                                             (struct ctl_scsiio *)pending_io,
11372                                             pending_io->scsiio.tag_num & 0xff);
11373                                         ctl_done(pending_io);
11374                                         break;
11375                                 case CTL_ACTION_ERROR:
11376                                 default:
11377                                         ctl_set_internal_failure(
11378                                                 (struct ctl_scsiio *)pending_io,
11379                                                 0,  // sks_valid
11380                                                 0); //retry count
11381                                         ctl_done(pending_io);
11382                                         break;
11383                                 }
11384                         }
11385
11386                         /*
11387                          * Build Unit Attention
11388                          */
11389                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11390                                 lun->pending_ua[i] |=
11391                                                      CTL_UA_ASYM_ACC_CHANGE;
11392                         }
11393                 } else {
11394                         panic("Unhandled HA mode failover, LUN flags = %#x, "
11395                               "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11396                 }
11397         }
11398         ctl_pause_rtr = 0;
11399         mtx_unlock(&ctl_softc->ctl_lock);
11400 }
11401
11402 static int
11403 ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11404 {
11405         struct ctl_lun *lun;
11406         const struct ctl_cmd_entry *entry;
11407         uint32_t initidx, targ_lun;
11408         int retval;
11409
11410         retval = 0;
11411
11412         lun = NULL;
11413
11414         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11415         if ((targ_lun < CTL_MAX_LUNS)
11416          && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11417                 lun = ctl_softc->ctl_luns[targ_lun];
11418                 /*
11419                  * If the LUN is invalid, pretend that it doesn't exist.
11420                  * It will go away as soon as all pending I/O has been
11421                  * completed.
11422                  */
11423                 if (lun->flags & CTL_LUN_DISABLED) {
11424                         lun = NULL;
11425                 } else {
11426                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11427                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11428                                 lun->be_lun;
11429                         if (lun->be_lun->lun_type == T_PROCESSOR) {
11430                                 ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11431                         }
11432
11433                         /*
11434                          * Every I/O goes into the OOA queue for a
11435                          * particular LUN, and stays there until completion.
11436                          */
11437                         mtx_lock(&lun->lun_lock);
11438                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11439                             ooa_links);
11440                 }
11441         } else {
11442                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11443                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11444         }
11445
11446         /* Get command entry and return error if it is unsuppotyed. */
11447         entry = ctl_validate_command(ctsio);
11448         if (entry == NULL) {
11449                 if (lun)
11450                         mtx_unlock(&lun->lun_lock);
11451                 return (retval);
11452         }
11453
11454         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11455         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11456
11457         /*
11458          * Check to see whether we can send this command to LUNs that don't
11459          * exist.  This should pretty much only be the case for inquiry
11460          * and request sense.  Further checks, below, really require having
11461          * a LUN, so we can't really check the command anymore.  Just put
11462          * it on the rtr queue.
11463          */
11464         if (lun == NULL) {
11465                 if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11466                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11467                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11468                         return (retval);
11469                 }
11470
11471                 ctl_set_unsupported_lun(ctsio);
11472                 ctl_done((union ctl_io *)ctsio);
11473                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11474                 return (retval);
11475         } else {
11476                 /*
11477                  * Make sure we support this particular command on this LUN.
11478                  * e.g., we don't support writes to the control LUN.
11479                  */
11480                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11481                         mtx_unlock(&lun->lun_lock);
11482                         ctl_set_invalid_opcode(ctsio);
11483                         ctl_done((union ctl_io *)ctsio);
11484                         return (retval);
11485                 }
11486         }
11487
11488         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11489
11490 #ifdef CTL_WITH_CA
11491         /*
11492          * If we've got a request sense, it'll clear the contingent
11493          * allegiance condition.  Otherwise, if we have a CA condition for
11494          * this initiator, clear it, because it sent down a command other
11495          * than request sense.
11496          */
11497         if ((ctsio->cdb[0] != REQUEST_SENSE)
11498          && (ctl_is_set(lun->have_ca, initidx)))
11499                 ctl_clear_mask(lun->have_ca, initidx);
11500 #endif
11501
11502         /*
11503          * If the command has this flag set, it handles its own unit
11504          * attention reporting, we shouldn't do anything.  Otherwise we
11505          * check for any pending unit attentions, and send them back to the
11506          * initiator.  We only do this when a command initially comes in,
11507          * not when we pull it off the blocked queue.
11508          *
11509          * According to SAM-3, section 5.3.2, the order that things get
11510          * presented back to the host is basically unit attentions caused
11511          * by some sort of reset event, busy status, reservation conflicts
11512          * or task set full, and finally any other status.
11513          *
11514          * One issue here is that some of the unit attentions we report
11515          * don't fall into the "reset" category (e.g. "reported luns data
11516          * has changed").  So reporting it here, before the reservation
11517          * check, may be technically wrong.  I guess the only thing to do
11518          * would be to check for and report the reset events here, and then
11519          * check for the other unit attention types after we check for a
11520          * reservation conflict.
11521          *
11522          * XXX KDM need to fix this
11523          */
11524         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11525                 ctl_ua_type ua_type;
11526
11527                 ua_type = lun->pending_ua[initidx];
11528                 if (ua_type != CTL_UA_NONE) {
11529                         scsi_sense_data_type sense_format;
11530
11531                         if (lun != NULL)
11532                                 sense_format = (lun->flags &
11533                                     CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11534                                     SSD_TYPE_FIXED;
11535                         else
11536                                 sense_format = SSD_TYPE_FIXED;
11537
11538                         ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11539                                                sense_format);
11540                         if (ua_type != CTL_UA_NONE) {
11541                                 ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11542                                 ctsio->io_hdr.status = CTL_SCSI_ERROR |
11543                                                        CTL_AUTOSENSE;
11544                                 ctsio->sense_len = SSD_FULL_SIZE;
11545                                 lun->pending_ua[initidx] &= ~ua_type;
11546                                 mtx_unlock(&lun->lun_lock);
11547                                 ctl_done((union ctl_io *)ctsio);
11548                                 return (retval);
11549                         }
11550                 }
11551         }
11552
11553
11554         if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11555                 mtx_unlock(&lun->lun_lock);
11556                 ctl_done((union ctl_io *)ctsio);
11557                 return (retval);
11558         }
11559
11560         /*
11561          * XXX CHD this is where we want to send IO to other side if
11562          * this LUN is secondary on this SC. We will need to make a copy
11563          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11564          * the copy we send as FROM_OTHER.
11565          * We also need to stuff the address of the original IO so we can
11566          * find it easily. Something similar will need be done on the other
11567          * side so when we are done we can find the copy.
11568          */
11569         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11570                 union ctl_ha_msg msg_info;
11571                 int isc_retval;
11572
11573                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11574
11575                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11576                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11577 #if 0
11578                 printf("1. ctsio %p\n", ctsio);
11579 #endif
11580                 msg_info.hdr.serializing_sc = NULL;
11581                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11582                 msg_info.scsi.tag_num = ctsio->tag_num;
11583                 msg_info.scsi.tag_type = ctsio->tag_type;
11584                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11585
11586                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11587
11588                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11589                     (void *)&msg_info, sizeof(msg_info), 0)) >
11590                     CTL_HA_STATUS_SUCCESS) {
11591                         printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11592                                isc_retval);
11593                         printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11594                 } else {
11595 #if 0
11596                         printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11597 #endif
11598                 }
11599
11600                 /*
11601                  * XXX KDM this I/O is off the incoming queue, but hasn't
11602                  * been inserted on any other queue.  We may need to come
11603                  * up with a holding queue while we wait for serialization
11604                  * so that we have an idea of what we're waiting for from
11605                  * the other side.
11606                  */
11607                 mtx_unlock(&lun->lun_lock);
11608                 return (retval);
11609         }
11610
11611         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11612                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11613                               ctl_ooaq, ooa_links))) {
11614         case CTL_ACTION_BLOCK:
11615                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11616                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11617                                   blocked_links);
11618                 mtx_unlock(&lun->lun_lock);
11619                 return (retval);
11620         case CTL_ACTION_PASS:
11621         case CTL_ACTION_SKIP:
11622                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11623                 mtx_unlock(&lun->lun_lock);
11624                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11625                 break;
11626         case CTL_ACTION_OVERLAP:
11627                 mtx_unlock(&lun->lun_lock);
11628                 ctl_set_overlapped_cmd(ctsio);
11629                 ctl_done((union ctl_io *)ctsio);
11630                 break;
11631         case CTL_ACTION_OVERLAP_TAG:
11632                 mtx_unlock(&lun->lun_lock);
11633                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11634                 ctl_done((union ctl_io *)ctsio);
11635                 break;
11636         case CTL_ACTION_ERROR:
11637         default:
11638                 mtx_unlock(&lun->lun_lock);
11639                 ctl_set_internal_failure(ctsio,
11640                                          /*sks_valid*/ 0,
11641                                          /*retry_count*/ 0);
11642                 ctl_done((union ctl_io *)ctsio);
11643                 break;
11644         }
11645         return (retval);
11646 }
11647
11648 const struct ctl_cmd_entry *
11649 ctl_get_cmd_entry(struct ctl_scsiio *ctsio)
11650 {
11651         const struct ctl_cmd_entry *entry;
11652         int service_action;
11653
11654         entry = &ctl_cmd_table[ctsio->cdb[0]];
11655         if (entry->flags & CTL_CMD_FLAG_SA5) {
11656                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11657                 entry = &((const struct ctl_cmd_entry *)
11658                     entry->execute)[service_action];
11659         }
11660         return (entry);
11661 }
11662
11663 const struct ctl_cmd_entry *
11664 ctl_validate_command(struct ctl_scsiio *ctsio)
11665 {
11666         const struct ctl_cmd_entry *entry;
11667         int i;
11668         uint8_t diff;
11669
11670         entry = ctl_get_cmd_entry(ctsio);
11671         if (entry->execute == NULL) {
11672                 ctl_set_invalid_opcode(ctsio);
11673                 ctl_done((union ctl_io *)ctsio);
11674                 return (NULL);
11675         }
11676         KASSERT(entry->length > 0,
11677             ("Not defined length for command 0x%02x/0x%02x",
11678              ctsio->cdb[0], ctsio->cdb[1]));
11679         for (i = 1; i < entry->length; i++) {
11680                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11681                 if (diff == 0)
11682                         continue;
11683                 ctl_set_invalid_field(ctsio,
11684                                       /*sks_valid*/ 1,
11685                                       /*command*/ 1,
11686                                       /*field*/ i,
11687                                       /*bit_valid*/ 1,
11688                                       /*bit*/ fls(diff) - 1);
11689                 ctl_done((union ctl_io *)ctsio);
11690                 return (NULL);
11691         }
11692         return (entry);
11693 }
11694
11695 static int
11696 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11697 {
11698
11699         switch (lun_type) {
11700         case T_PROCESSOR:
11701                 if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11702                     ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11703                         return (0);
11704                 break;
11705         case T_DIRECT:
11706                 if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11707                     ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11708                         return (0);
11709                 break;
11710         default:
11711                 return (0);
11712         }
11713         return (1);
11714 }
11715
11716 static int
11717 ctl_scsiio(struct ctl_scsiio *ctsio)
11718 {
11719         int retval;
11720         const struct ctl_cmd_entry *entry;
11721
11722         retval = CTL_RETVAL_COMPLETE;
11723
11724         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11725
11726         entry = ctl_get_cmd_entry(ctsio);
11727
11728         /*
11729          * If this I/O has been aborted, just send it straight to
11730          * ctl_done() without executing it.
11731          */
11732         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11733                 ctl_done((union ctl_io *)ctsio);
11734                 goto bailout;
11735         }
11736
11737         /*
11738          * All the checks should have been handled by ctl_scsiio_precheck().
11739          * We should be clear now to just execute the I/O.
11740          */
11741         retval = entry->execute(ctsio);
11742
11743 bailout:
11744         return (retval);
11745 }
11746
11747 /*
11748  * Since we only implement one target right now, a bus reset simply resets
11749  * our single target.
11750  */
11751 static int
11752 ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11753 {
11754         return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11755 }
11756
11757 static int
11758 ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11759                  ctl_ua_type ua_type)
11760 {
11761         struct ctl_lun *lun;
11762         int retval;
11763
11764         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11765                 union ctl_ha_msg msg_info;
11766
11767                 io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11768                 msg_info.hdr.nexus = io->io_hdr.nexus;
11769                 if (ua_type==CTL_UA_TARG_RESET)
11770                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11771                 else
11772                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11773                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11774                 msg_info.hdr.original_sc = NULL;
11775                 msg_info.hdr.serializing_sc = NULL;
11776                 if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11777                     (void *)&msg_info, sizeof(msg_info), 0)) {
11778                 }
11779         }
11780         retval = 0;
11781
11782         mtx_lock(&ctl_softc->ctl_lock);
11783         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11784                 retval += ctl_lun_reset(lun, io, ua_type);
11785         mtx_unlock(&ctl_softc->ctl_lock);
11786
11787         return (retval);
11788 }
11789
11790 /*
11791  * The LUN should always be set.  The I/O is optional, and is used to
11792  * distinguish between I/Os sent by this initiator, and by other
11793  * initiators.  We set unit attention for initiators other than this one.
11794  * SAM-3 is vague on this point.  It does say that a unit attention should
11795  * be established for other initiators when a LUN is reset (see section
11796  * 5.7.3), but it doesn't specifically say that the unit attention should
11797  * be established for this particular initiator when a LUN is reset.  Here
11798  * is the relevant text, from SAM-3 rev 8:
11799  *
11800  * 5.7.2 When a SCSI initiator port aborts its own tasks
11801  *
11802  * When a SCSI initiator port causes its own task(s) to be aborted, no
11803  * notification that the task(s) have been aborted shall be returned to
11804  * the SCSI initiator port other than the completion response for the
11805  * command or task management function action that caused the task(s) to
11806  * be aborted and notification(s) associated with related effects of the
11807  * action (e.g., a reset unit attention condition).
11808  *
11809  * XXX KDM for now, we're setting unit attention for all initiators.
11810  */
11811 static int
11812 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11813 {
11814         union ctl_io *xio;
11815 #if 0
11816         uint32_t initindex;
11817 #endif
11818         int i;
11819
11820         mtx_lock(&lun->lun_lock);
11821         /*
11822          * Run through the OOA queue and abort each I/O.
11823          */
11824 #if 0
11825         TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11826 #endif
11827         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11828              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11829                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11830         }
11831
11832         /*
11833          * This version sets unit attention for every
11834          */
11835 #if 0
11836         initindex = ctl_get_initindex(&io->io_hdr.nexus);
11837         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11838                 if (initindex == i)
11839                         continue;
11840                 lun->pending_ua[i] |= ua_type;
11841         }
11842 #endif
11843
11844         /*
11845          * A reset (any kind, really) clears reservations established with
11846          * RESERVE/RELEASE.  It does not clear reservations established
11847          * with PERSISTENT RESERVE OUT, but we don't support that at the
11848          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11849          * reservations made with the RESERVE/RELEASE commands, because
11850          * those commands are obsolete in SPC-3.
11851          */
11852         lun->flags &= ~CTL_LUN_RESERVED;
11853
11854         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11855 #ifdef CTL_WITH_CA
11856                 ctl_clear_mask(lun->have_ca, i);
11857 #endif
11858                 lun->pending_ua[i] |= ua_type;
11859         }
11860         mtx_unlock(&lun->lun_lock);
11861
11862         return (0);
11863 }
11864
11865 static int
11866 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11867     int other_sc)
11868 {
11869         union ctl_io *xio;
11870         int found;
11871
11872         mtx_assert(&lun->lun_lock, MA_OWNED);
11873
11874         /*
11875          * Run through the OOA queue and attempt to find the given I/O.
11876          * The target port, initiator ID, tag type and tag number have to
11877          * match the values that we got from the initiator.  If we have an
11878          * untagged command to abort, simply abort the first untagged command
11879          * we come to.  We only allow one untagged command at a time of course.
11880          */
11881         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11882              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11883
11884                 if ((targ_port == UINT32_MAX ||
11885                      targ_port == xio->io_hdr.nexus.targ_port) &&
11886                     (init_id == UINT32_MAX ||
11887                      init_id == xio->io_hdr.nexus.initid.id)) {
11888                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11889                             init_id != xio->io_hdr.nexus.initid.id)
11890                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11891                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11892                         found = 1;
11893                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11894                                 union ctl_ha_msg msg_info;
11895
11896                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11897                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11898                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11899                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11900                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11901                                 msg_info.hdr.original_sc = NULL;
11902                                 msg_info.hdr.serializing_sc = NULL;
11903                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11904                                     (void *)&msg_info, sizeof(msg_info), 0);
11905                         }
11906                 }
11907         }
11908         return (found);
11909 }
11910
11911 static int
11912 ctl_abort_task_set(union ctl_io *io)
11913 {
11914         struct ctl_softc *softc = control_softc;
11915         struct ctl_lun *lun;
11916         uint32_t targ_lun;
11917
11918         /*
11919          * Look up the LUN.
11920          */
11921         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11922         mtx_lock(&softc->ctl_lock);
11923         if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
11924                 lun = softc->ctl_luns[targ_lun];
11925         else {
11926                 mtx_unlock(&softc->ctl_lock);
11927                 return (1);
11928         }
11929
11930         mtx_lock(&lun->lun_lock);
11931         mtx_unlock(&softc->ctl_lock);
11932         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11933                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11934                     io->io_hdr.nexus.initid.id,
11935                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11936         } else { /* CTL_TASK_CLEAR_TASK_SET */
11937                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11938                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11939         }
11940         mtx_unlock(&lun->lun_lock);
11941         return (0);
11942 }
11943
11944 static int
11945 ctl_i_t_nexus_reset(union ctl_io *io)
11946 {
11947         struct ctl_softc *softc = control_softc;
11948         struct ctl_lun *lun;
11949         uint32_t initindex;
11950
11951         initindex = ctl_get_initindex(&io->io_hdr.nexus);
11952         mtx_lock(&softc->ctl_lock);
11953         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11954                 mtx_lock(&lun->lun_lock);
11955                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11956                     io->io_hdr.nexus.initid.id,
11957                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11958 #ifdef CTL_WITH_CA
11959                 ctl_clear_mask(lun->have_ca, initindex);
11960 #endif
11961                 lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
11962                 mtx_unlock(&lun->lun_lock);
11963         }
11964         mtx_unlock(&softc->ctl_lock);
11965         return (0);
11966 }
11967
11968 static int
11969 ctl_abort_task(union ctl_io *io)
11970 {
11971         union ctl_io *xio;
11972         struct ctl_lun *lun;
11973         struct ctl_softc *ctl_softc;
11974 #if 0
11975         struct sbuf sb;
11976         char printbuf[128];
11977 #endif
11978         int found;
11979         uint32_t targ_lun;
11980
11981         ctl_softc = control_softc;
11982         found = 0;
11983
11984         /*
11985          * Look up the LUN.
11986          */
11987         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11988         mtx_lock(&ctl_softc->ctl_lock);
11989         if ((targ_lun < CTL_MAX_LUNS)
11990          && (ctl_softc->ctl_luns[targ_lun] != NULL))
11991                 lun = ctl_softc->ctl_luns[targ_lun];
11992         else {
11993                 mtx_unlock(&ctl_softc->ctl_lock);
11994                 return (1);
11995         }
11996
11997 #if 0
11998         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11999                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12000 #endif
12001
12002         mtx_lock(&lun->lun_lock);
12003         mtx_unlock(&ctl_softc->ctl_lock);
12004         /*
12005          * Run through the OOA queue and attempt to find the given I/O.
12006          * The target port, initiator ID, tag type and tag number have to
12007          * match the values that we got from the initiator.  If we have an
12008          * untagged command to abort, simply abort the first untagged command
12009          * we come to.  We only allow one untagged command at a time of course.
12010          */
12011 #if 0
12012         TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12013 #endif
12014         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12015              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12016 #if 0
12017                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12018
12019                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12020                             lun->lun, xio->scsiio.tag_num,
12021                             xio->scsiio.tag_type,
12022                             (xio->io_hdr.blocked_links.tqe_prev
12023                             == NULL) ? "" : " BLOCKED",
12024                             (xio->io_hdr.flags &
12025                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12026                             (xio->io_hdr.flags &
12027                             CTL_FLAG_ABORT) ? " ABORT" : "",
12028                             (xio->io_hdr.flags &
12029                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12030                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12031                 sbuf_finish(&sb);
12032                 printf("%s\n", sbuf_data(&sb));
12033 #endif
12034
12035                 if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12036                  && (xio->io_hdr.nexus.initid.id ==
12037                      io->io_hdr.nexus.initid.id)) {
12038                         /*
12039                          * If the abort says that the task is untagged, the
12040                          * task in the queue must be untagged.  Otherwise,
12041                          * we just check to see whether the tag numbers
12042                          * match.  This is because the QLogic firmware
12043                          * doesn't pass back the tag type in an abort
12044                          * request.
12045                          */
12046 #if 0
12047                         if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12048                           && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12049                          || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12050 #endif
12051                         /*
12052                          * XXX KDM we've got problems with FC, because it
12053                          * doesn't send down a tag type with aborts.  So we
12054                          * can only really go by the tag number...
12055                          * This may cause problems with parallel SCSI.
12056                          * Need to figure that out!!
12057                          */
12058                         if (xio->scsiio.tag_num == io->taskio.tag_num) {
12059                                 xio->io_hdr.flags |= CTL_FLAG_ABORT;
12060                                 found = 1;
12061                                 if ((io->io_hdr.flags &
12062                                      CTL_FLAG_FROM_OTHER_SC) == 0 &&
12063                                     !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12064                                         union ctl_ha_msg msg_info;
12065
12066                                         io->io_hdr.flags |=
12067                                                         CTL_FLAG_SENT_2OTHER_SC;
12068                                         msg_info.hdr.nexus = io->io_hdr.nexus;
12069                                         msg_info.task.task_action =
12070                                                 CTL_TASK_ABORT_TASK;
12071                                         msg_info.task.tag_num =
12072                                                 io->taskio.tag_num;
12073                                         msg_info.task.tag_type =
12074                                                 io->taskio.tag_type;
12075                                         msg_info.hdr.msg_type =
12076                                                 CTL_MSG_MANAGE_TASKS;
12077                                         msg_info.hdr.original_sc = NULL;
12078                                         msg_info.hdr.serializing_sc = NULL;
12079 #if 0
12080                                         printf("Sent Abort to other side\n");
12081 #endif
12082                                         if (CTL_HA_STATUS_SUCCESS !=
12083                                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12084                                                 (void *)&msg_info,
12085                                                 sizeof(msg_info), 0)) {
12086                                         }
12087                                 }
12088 #if 0
12089                                 printf("ctl_abort_task: found I/O to abort\n");
12090 #endif
12091                                 break;
12092                         }
12093                 }
12094         }
12095         mtx_unlock(&lun->lun_lock);
12096
12097         if (found == 0) {
12098                 /*
12099                  * This isn't really an error.  It's entirely possible for
12100                  * the abort and command completion to cross on the wire.
12101                  * This is more of an informative/diagnostic error.
12102                  */
12103 #if 0
12104                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12105                        "%d:%d:%d:%d tag %d type %d\n",
12106                        io->io_hdr.nexus.initid.id,
12107                        io->io_hdr.nexus.targ_port,
12108                        io->io_hdr.nexus.targ_target.id,
12109                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12110                        io->taskio.tag_type);
12111 #endif
12112         }
12113         return (0);
12114 }
12115
12116 static void
12117 ctl_run_task(union ctl_io *io)
12118 {
12119         struct ctl_softc *ctl_softc = control_softc;
12120         int retval = 1;
12121         const char *task_desc;
12122
12123         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12124
12125         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12126             ("ctl_run_task: Unextected io_type %d\n",
12127              io->io_hdr.io_type));
12128
12129         task_desc = ctl_scsi_task_string(&io->taskio);
12130         if (task_desc != NULL) {
12131 #ifdef NEEDTOPORT
12132                 csevent_log(CSC_CTL | CSC_SHELF_SW |
12133                             CTL_TASK_REPORT,
12134                             csevent_LogType_Trace,
12135                             csevent_Severity_Information,
12136                             csevent_AlertLevel_Green,
12137                             csevent_FRU_Firmware,
12138                             csevent_FRU_Unknown,
12139                             "CTL: received task: %s",task_desc);
12140 #endif
12141         } else {
12142 #ifdef NEEDTOPORT
12143                 csevent_log(CSC_CTL | CSC_SHELF_SW |
12144                             CTL_TASK_REPORT,
12145                             csevent_LogType_Trace,
12146                             csevent_Severity_Information,
12147                             csevent_AlertLevel_Green,
12148                             csevent_FRU_Firmware,
12149                             csevent_FRU_Unknown,
12150                             "CTL: received unknown task "
12151                             "type: %d (%#x)",
12152                             io->taskio.task_action,
12153                             io->taskio.task_action);
12154 #endif
12155         }
12156         switch (io->taskio.task_action) {
12157         case CTL_TASK_ABORT_TASK:
12158                 retval = ctl_abort_task(io);
12159                 break;
12160         case CTL_TASK_ABORT_TASK_SET:
12161         case CTL_TASK_CLEAR_TASK_SET:
12162                 retval = ctl_abort_task_set(io);
12163                 break;
12164         case CTL_TASK_CLEAR_ACA:
12165                 break;
12166         case CTL_TASK_I_T_NEXUS_RESET:
12167                 retval = ctl_i_t_nexus_reset(io);
12168                 break;
12169         case CTL_TASK_LUN_RESET: {
12170                 struct ctl_lun *lun;
12171                 uint32_t targ_lun;
12172
12173                 targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12174                 mtx_lock(&ctl_softc->ctl_lock);
12175                 if ((targ_lun < CTL_MAX_LUNS)
12176                  && (ctl_softc->ctl_luns[targ_lun] != NULL))
12177                         lun = ctl_softc->ctl_luns[targ_lun];
12178                 else {
12179                         mtx_unlock(&ctl_softc->ctl_lock);
12180                         retval = 1;
12181                         break;
12182                 }
12183
12184                 if (!(io->io_hdr.flags &
12185                     CTL_FLAG_FROM_OTHER_SC)) {
12186                         union ctl_ha_msg msg_info;
12187
12188                         io->io_hdr.flags |=
12189                                 CTL_FLAG_SENT_2OTHER_SC;
12190                         msg_info.hdr.msg_type =
12191                                 CTL_MSG_MANAGE_TASKS;
12192                         msg_info.hdr.nexus = io->io_hdr.nexus;
12193                         msg_info.task.task_action =
12194                                 CTL_TASK_LUN_RESET;
12195                         msg_info.hdr.original_sc = NULL;
12196                         msg_info.hdr.serializing_sc = NULL;
12197                         if (CTL_HA_STATUS_SUCCESS !=
12198                             ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12199                             (void *)&msg_info,
12200                             sizeof(msg_info), 0)) {
12201                         }
12202                 }
12203
12204                 retval = ctl_lun_reset(lun, io,
12205                                        CTL_UA_LUN_RESET);
12206                 mtx_unlock(&ctl_softc->ctl_lock);
12207                 break;
12208         }
12209         case CTL_TASK_TARGET_RESET:
12210                 retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12211                 break;
12212         case CTL_TASK_BUS_RESET:
12213                 retval = ctl_bus_reset(ctl_softc, io);
12214                 break;
12215         case CTL_TASK_PORT_LOGIN:
12216                 break;
12217         case CTL_TASK_PORT_LOGOUT:
12218                 break;
12219         default:
12220                 printf("ctl_run_task: got unknown task management event %d\n",
12221                        io->taskio.task_action);
12222                 break;
12223         }
12224         if (retval == 0)
12225                 io->io_hdr.status = CTL_SUCCESS;
12226         else
12227                 io->io_hdr.status = CTL_ERROR;
12228         ctl_done(io);
12229 }
12230
12231 /*
12232  * For HA operation.  Handle commands that come in from the other
12233  * controller.
12234  */
12235 static void
12236 ctl_handle_isc(union ctl_io *io)
12237 {
12238         int free_io;
12239         struct ctl_lun *lun;
12240         struct ctl_softc *ctl_softc;
12241         uint32_t targ_lun;
12242
12243         ctl_softc = control_softc;
12244
12245         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12246         lun = ctl_softc->ctl_luns[targ_lun];
12247
12248         switch (io->io_hdr.msg_type) {
12249         case CTL_MSG_SERIALIZE:
12250                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12251                 break;
12252         case CTL_MSG_R2R: {
12253                 const struct ctl_cmd_entry *entry;
12254
12255                 /*
12256                  * This is only used in SER_ONLY mode.
12257                  */
12258                 free_io = 0;
12259                 entry = ctl_get_cmd_entry(&io->scsiio);
12260                 mtx_lock(&lun->lun_lock);
12261                 if (ctl_scsiio_lun_check(ctl_softc, lun,
12262                     entry, (struct ctl_scsiio *)io) != 0) {
12263                         mtx_unlock(&lun->lun_lock);
12264                         ctl_done(io);
12265                         break;
12266                 }
12267                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12268                 mtx_unlock(&lun->lun_lock);
12269                 ctl_enqueue_rtr(io);
12270                 break;
12271         }
12272         case CTL_MSG_FINISH_IO:
12273                 if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12274                         free_io = 0;
12275                         ctl_done(io);
12276                 } else {
12277                         free_io = 1;
12278                         mtx_lock(&lun->lun_lock);
12279                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12280                                      ooa_links);
12281                         ctl_check_blocked(lun);
12282                         mtx_unlock(&lun->lun_lock);
12283                 }
12284                 break;
12285         case CTL_MSG_PERS_ACTION:
12286                 ctl_hndl_per_res_out_on_other_sc(
12287                         (union ctl_ha_msg *)&io->presio.pr_msg);
12288                 free_io = 1;
12289                 break;
12290         case CTL_MSG_BAD_JUJU:
12291                 free_io = 0;
12292                 ctl_done(io);
12293                 break;
12294         case CTL_MSG_DATAMOVE:
12295                 /* Only used in XFER mode */
12296                 free_io = 0;
12297                 ctl_datamove_remote(io);
12298                 break;
12299         case CTL_MSG_DATAMOVE_DONE:
12300                 /* Only used in XFER mode */
12301                 free_io = 0;
12302                 io->scsiio.be_move_done(io);
12303                 break;
12304         default:
12305                 free_io = 1;
12306                 printf("%s: Invalid message type %d\n",
12307                        __func__, io->io_hdr.msg_type);
12308                 break;
12309         }
12310         if (free_io)
12311                 ctl_free_io(io);
12312
12313 }
12314
12315
12316 /*
12317  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12318  * there is no match.
12319  */
12320 static ctl_lun_error_pattern
12321 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12322 {
12323         const struct ctl_cmd_entry *entry;
12324         ctl_lun_error_pattern filtered_pattern, pattern;
12325
12326         pattern = desc->error_pattern;
12327
12328         /*
12329          * XXX KDM we need more data passed into this function to match a
12330          * custom pattern, and we actually need to implement custom pattern
12331          * matching.
12332          */
12333         if (pattern & CTL_LUN_PAT_CMD)
12334                 return (CTL_LUN_PAT_CMD);
12335
12336         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12337                 return (CTL_LUN_PAT_ANY);
12338
12339         entry = ctl_get_cmd_entry(ctsio);
12340
12341         filtered_pattern = entry->pattern & pattern;
12342
12343         /*
12344          * If the user requested specific flags in the pattern (e.g.
12345          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12346          * flags.
12347          *
12348          * If the user did not specify any flags, it doesn't matter whether
12349          * or not the command supports the flags.
12350          */
12351         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12352              (pattern & ~CTL_LUN_PAT_MASK))
12353                 return (CTL_LUN_PAT_NONE);
12354
12355         /*
12356          * If the user asked for a range check, see if the requested LBA
12357          * range overlaps with this command's LBA range.
12358          */
12359         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12360                 uint64_t lba1;
12361                 uint32_t len1;
12362                 ctl_action action;
12363                 int retval;
12364
12365                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12366                 if (retval != 0)
12367                         return (CTL_LUN_PAT_NONE);
12368
12369                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12370                                               desc->lba_range.len);
12371                 /*
12372                  * A "pass" means that the LBA ranges don't overlap, so
12373                  * this doesn't match the user's range criteria.
12374                  */
12375                 if (action == CTL_ACTION_PASS)
12376                         return (CTL_LUN_PAT_NONE);
12377         }
12378
12379         return (filtered_pattern);
12380 }
12381
12382 static void
12383 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12384 {
12385         struct ctl_error_desc *desc, *desc2;
12386
12387         mtx_assert(&lun->lun_lock, MA_OWNED);
12388
12389         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12390                 ctl_lun_error_pattern pattern;
12391                 /*
12392                  * Check to see whether this particular command matches
12393                  * the pattern in the descriptor.
12394                  */
12395                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12396                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12397                         continue;
12398
12399                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12400                 case CTL_LUN_INJ_ABORTED:
12401                         ctl_set_aborted(&io->scsiio);
12402                         break;
12403                 case CTL_LUN_INJ_MEDIUM_ERR:
12404                         ctl_set_medium_error(&io->scsiio);
12405                         break;
12406                 case CTL_LUN_INJ_UA:
12407                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12408                          * OCCURRED */
12409                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12410                         break;
12411                 case CTL_LUN_INJ_CUSTOM:
12412                         /*
12413                          * We're assuming the user knows what he is doing.
12414                          * Just copy the sense information without doing
12415                          * checks.
12416                          */
12417                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12418                               ctl_min(sizeof(desc->custom_sense),
12419                                       sizeof(io->scsiio.sense_data)));
12420                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12421                         io->scsiio.sense_len = SSD_FULL_SIZE;
12422                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12423                         break;
12424                 case CTL_LUN_INJ_NONE:
12425                 default:
12426                         /*
12427                          * If this is an error injection type we don't know
12428                          * about, clear the continuous flag (if it is set)
12429                          * so it will get deleted below.
12430                          */
12431                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12432                         break;
12433                 }
12434                 /*
12435                  * By default, each error injection action is a one-shot
12436                  */
12437                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12438                         continue;
12439
12440                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12441
12442                 free(desc, M_CTL);
12443         }
12444 }
12445
12446 #ifdef CTL_IO_DELAY
12447 static void
12448 ctl_datamove_timer_wakeup(void *arg)
12449 {
12450         union ctl_io *io;
12451
12452         io = (union ctl_io *)arg;
12453
12454         ctl_datamove(io);
12455 }
12456 #endif /* CTL_IO_DELAY */
12457
12458 void
12459 ctl_datamove(union ctl_io *io)
12460 {
12461         void (*fe_datamove)(union ctl_io *io);
12462
12463         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12464
12465         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12466
12467 #ifdef CTL_TIME_IO
12468         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12469                 char str[256];
12470                 char path_str[64];
12471                 struct sbuf sb;
12472
12473                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12474                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12475
12476                 sbuf_cat(&sb, path_str);
12477                 switch (io->io_hdr.io_type) {
12478                 case CTL_IO_SCSI:
12479                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12480                         sbuf_printf(&sb, "\n");
12481                         sbuf_cat(&sb, path_str);
12482                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12483                                     io->scsiio.tag_num, io->scsiio.tag_type);
12484                         break;
12485                 case CTL_IO_TASK:
12486                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12487                                     "Tag Type: %d\n", io->taskio.task_action,
12488                                     io->taskio.tag_num, io->taskio.tag_type);
12489                         break;
12490                 default:
12491                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12492                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12493                         break;
12494                 }
12495                 sbuf_cat(&sb, path_str);
12496                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12497                             (intmax_t)time_uptime - io->io_hdr.start_time);
12498                 sbuf_finish(&sb);
12499                 printf("%s", sbuf_data(&sb));
12500         }
12501 #endif /* CTL_TIME_IO */
12502
12503 #ifdef CTL_IO_DELAY
12504         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12505                 struct ctl_lun *lun;
12506
12507                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12508
12509                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12510         } else {
12511                 struct ctl_lun *lun;
12512
12513                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12514                 if ((lun != NULL)
12515                  && (lun->delay_info.datamove_delay > 0)) {
12516                         struct callout *callout;
12517
12518                         callout = (struct callout *)&io->io_hdr.timer_bytes;
12519                         callout_init(callout, /*mpsafe*/ 1);
12520                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12521                         callout_reset(callout,
12522                                       lun->delay_info.datamove_delay * hz,
12523                                       ctl_datamove_timer_wakeup, io);
12524                         if (lun->delay_info.datamove_type ==
12525                             CTL_DELAY_TYPE_ONESHOT)
12526                                 lun->delay_info.datamove_delay = 0;
12527                         return;
12528                 }
12529         }
12530 #endif
12531
12532         /*
12533          * This command has been aborted.  Set the port status, so we fail
12534          * the data move.
12535          */
12536         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12537                 printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12538                        io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12539                        io->io_hdr.nexus.targ_port,
12540                        (uintmax_t)io->io_hdr.nexus.targ_target.id,
12541                        io->io_hdr.nexus.targ_lun);
12542                 io->io_hdr.port_status = 31337;
12543                 /*
12544                  * Note that the backend, in this case, will get the
12545                  * callback in its context.  In other cases it may get
12546                  * called in the frontend's interrupt thread context.
12547                  */
12548                 io->scsiio.be_move_done(io);
12549                 return;
12550         }
12551
12552         /*
12553          * If we're in XFER mode and this I/O is from the other shelf
12554          * controller, we need to send the DMA to the other side to
12555          * actually transfer the data to/from the host.  In serialize only
12556          * mode the transfer happens below CTL and ctl_datamove() is only
12557          * called on the machine that originally received the I/O.
12558          */
12559         if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12560          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12561                 union ctl_ha_msg msg;
12562                 uint32_t sg_entries_sent;
12563                 int do_sg_copy;
12564                 int i;
12565
12566                 memset(&msg, 0, sizeof(msg));
12567                 msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12568                 msg.hdr.original_sc = io->io_hdr.original_sc;
12569                 msg.hdr.serializing_sc = io;
12570                 msg.hdr.nexus = io->io_hdr.nexus;
12571                 msg.dt.flags = io->io_hdr.flags;
12572                 /*
12573                  * We convert everything into a S/G list here.  We can't
12574                  * pass by reference, only by value between controllers.
12575                  * So we can't pass a pointer to the S/G list, only as many
12576                  * S/G entries as we can fit in here.  If it's possible for
12577                  * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12578                  * then we need to break this up into multiple transfers.
12579                  */
12580                 if (io->scsiio.kern_sg_entries == 0) {
12581                         msg.dt.kern_sg_entries = 1;
12582                         /*
12583                          * If this is in cached memory, flush the cache
12584                          * before we send the DMA request to the other
12585                          * controller.  We want to do this in either the
12586                          * read or the write case.  The read case is
12587                          * straightforward.  In the write case, we want to
12588                          * make sure nothing is in the local cache that
12589                          * could overwrite the DMAed data.
12590                          */
12591                         if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12592                                 /*
12593                                  * XXX KDM use bus_dmamap_sync() here.
12594                                  */
12595                         }
12596
12597                         /*
12598                          * Convert to a physical address if this is a
12599                          * virtual address.
12600                          */
12601                         if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12602                                 msg.dt.sg_list[0].addr =
12603                                         io->scsiio.kern_data_ptr;
12604                         } else {
12605                                 /*
12606                                  * XXX KDM use busdma here!
12607                                  */
12608 #if 0
12609                                 msg.dt.sg_list[0].addr = (void *)
12610                                         vtophys(io->scsiio.kern_data_ptr);
12611 #endif
12612                         }
12613
12614                         msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12615                         do_sg_copy = 0;
12616                 } else {
12617                         struct ctl_sg_entry *sgl;
12618
12619                         do_sg_copy = 1;
12620                         msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12621                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12622                         if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12623                                 /*
12624                                  * XXX KDM use bus_dmamap_sync() here.
12625                                  */
12626                         }
12627                 }
12628
12629                 msg.dt.kern_data_len = io->scsiio.kern_data_len;
12630                 msg.dt.kern_total_len = io->scsiio.kern_total_len;
12631                 msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12632                 msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12633                 msg.dt.sg_sequence = 0;
12634
12635                 /*
12636                  * Loop until we've sent all of the S/G entries.  On the
12637                  * other end, we'll recompose these S/G entries into one
12638                  * contiguous list before passing it to the
12639                  */
12640                 for (sg_entries_sent = 0; sg_entries_sent <
12641                      msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12642                         msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12643                                 sizeof(msg.dt.sg_list[0])),
12644                                 msg.dt.kern_sg_entries - sg_entries_sent);
12645
12646                         if (do_sg_copy != 0) {
12647                                 struct ctl_sg_entry *sgl;
12648                                 int j;
12649
12650                                 sgl = (struct ctl_sg_entry *)
12651                                         io->scsiio.kern_data_ptr;
12652                                 /*
12653                                  * If this is in cached memory, flush the cache
12654                                  * before we send the DMA request to the other
12655                                  * controller.  We want to do this in either
12656                                  * the * read or the write case.  The read
12657                                  * case is straightforward.  In the write
12658                                  * case, we want to make sure nothing is
12659                                  * in the local cache that could overwrite
12660                                  * the DMAed data.
12661                                  */
12662
12663                                 for (i = sg_entries_sent, j = 0;
12664                                      i < msg.dt.cur_sg_entries; i++, j++) {
12665                                         if ((io->io_hdr.flags &
12666                                              CTL_FLAG_NO_DATASYNC) == 0) {
12667                                                 /*
12668                                                  * XXX KDM use bus_dmamap_sync()
12669                                                  */
12670                                         }
12671                                         if ((io->io_hdr.flags &
12672                                              CTL_FLAG_BUS_ADDR) == 0) {
12673                                                 /*
12674                                                  * XXX KDM use busdma.
12675                                                  */
12676 #if 0
12677                                                 msg.dt.sg_list[j].addr =(void *)
12678                                                        vtophys(sgl[i].addr);
12679 #endif
12680                                         } else {
12681                                                 msg.dt.sg_list[j].addr =
12682                                                         sgl[i].addr;
12683                                         }
12684                                         msg.dt.sg_list[j].len = sgl[i].len;
12685                                 }
12686                         }
12687
12688                         sg_entries_sent += msg.dt.cur_sg_entries;
12689                         if (sg_entries_sent >= msg.dt.kern_sg_entries)
12690                                 msg.dt.sg_last = 1;
12691                         else
12692                                 msg.dt.sg_last = 0;
12693
12694                         /*
12695                          * XXX KDM drop and reacquire the lock here?
12696                          */
12697                         if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12698                             sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12699                                 /*
12700                                  * XXX do something here.
12701                                  */
12702                         }
12703
12704                         msg.dt.sent_sg_entries = sg_entries_sent;
12705                 }
12706                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12707                 if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12708                         ctl_failover_io(io, /*have_lock*/ 0);
12709
12710         } else {
12711
12712                 /*
12713                  * Lookup the fe_datamove() function for this particular
12714                  * front end.
12715                  */
12716                 fe_datamove =
12717                     control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12718
12719                 fe_datamove(io);
12720         }
12721 }
12722
12723 static void
12724 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12725 {
12726         union ctl_ha_msg msg;
12727         int isc_status;
12728
12729         memset(&msg, 0, sizeof(msg));
12730
12731         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12732         msg.hdr.original_sc = io;
12733         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12734         msg.hdr.nexus = io->io_hdr.nexus;
12735         msg.hdr.status = io->io_hdr.status;
12736         msg.scsi.tag_num = io->scsiio.tag_num;
12737         msg.scsi.tag_type = io->scsiio.tag_type;
12738         msg.scsi.scsi_status = io->scsiio.scsi_status;
12739         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12740                sizeof(io->scsiio.sense_data));
12741         msg.scsi.sense_len = io->scsiio.sense_len;
12742         msg.scsi.sense_residual = io->scsiio.sense_residual;
12743         msg.scsi.fetd_status = io->io_hdr.port_status;
12744         msg.scsi.residual = io->scsiio.residual;
12745         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12746
12747         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12748                 ctl_failover_io(io, /*have_lock*/ have_lock);
12749                 return;
12750         }
12751
12752         isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12753         if (isc_status > CTL_HA_STATUS_SUCCESS) {
12754                 /* XXX do something if this fails */
12755         }
12756
12757 }
12758
12759 /*
12760  * The DMA to the remote side is done, now we need to tell the other side
12761  * we're done so it can continue with its data movement.
12762  */
12763 static void
12764 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12765 {
12766         union ctl_io *io;
12767
12768         io = rq->context;
12769
12770         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12771                 printf("%s: ISC DMA write failed with error %d", __func__,
12772                        rq->ret);
12773                 ctl_set_internal_failure(&io->scsiio,
12774                                          /*sks_valid*/ 1,
12775                                          /*retry_count*/ rq->ret);
12776         }
12777
12778         ctl_dt_req_free(rq);
12779
12780         /*
12781          * In this case, we had to malloc the memory locally.  Free it.
12782          */
12783         if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12784                 int i;
12785                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12786                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12787         }
12788         /*
12789          * The data is in local and remote memory, so now we need to send
12790          * status (good or back) back to the other side.
12791          */
12792         ctl_send_datamove_done(io, /*have_lock*/ 0);
12793 }
12794
12795 /*
12796  * We've moved the data from the host/controller into local memory.  Now we
12797  * need to push it over to the remote controller's memory.
12798  */
12799 static int
12800 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12801 {
12802         int retval;
12803
12804         retval = 0;
12805
12806         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12807                                           ctl_datamove_remote_write_cb);
12808
12809         return (retval);
12810 }
12811
12812 static void
12813 ctl_datamove_remote_write(union ctl_io *io)
12814 {
12815         int retval;
12816         void (*fe_datamove)(union ctl_io *io);
12817
12818         /*
12819          * - Get the data from the host/HBA into local memory.
12820          * - DMA memory from the local controller to the remote controller.
12821          * - Send status back to the remote controller.
12822          */
12823
12824         retval = ctl_datamove_remote_sgl_setup(io);
12825         if (retval != 0)
12826                 return;
12827
12828         /* Switch the pointer over so the FETD knows what to do */
12829         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12830
12831         /*
12832          * Use a custom move done callback, since we need to send completion
12833          * back to the other controller, not to the backend on this side.
12834          */
12835         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12836
12837         fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12838
12839         fe_datamove(io);
12840
12841         return;
12842
12843 }
12844
12845 static int
12846 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12847 {
12848 #if 0
12849         char str[256];
12850         char path_str[64];
12851         struct sbuf sb;
12852 #endif
12853
12854         /*
12855          * In this case, we had to malloc the memory locally.  Free it.
12856          */
12857         if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12858                 int i;
12859                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12860                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12861         }
12862
12863 #if 0
12864         scsi_path_string(io, path_str, sizeof(path_str));
12865         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12866         sbuf_cat(&sb, path_str);
12867         scsi_command_string(&io->scsiio, NULL, &sb);
12868         sbuf_printf(&sb, "\n");
12869         sbuf_cat(&sb, path_str);
12870         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12871                     io->scsiio.tag_num, io->scsiio.tag_type);
12872         sbuf_cat(&sb, path_str);
12873         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12874                     io->io_hdr.flags, io->io_hdr.status);
12875         sbuf_finish(&sb);
12876         printk("%s", sbuf_data(&sb));
12877 #endif
12878
12879
12880         /*
12881          * The read is done, now we need to send status (good or bad) back
12882          * to the other side.
12883          */
12884         ctl_send_datamove_done(io, /*have_lock*/ 0);
12885
12886         return (0);
12887 }
12888
12889 static void
12890 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12891 {
12892         union ctl_io *io;
12893         void (*fe_datamove)(union ctl_io *io);
12894
12895         io = rq->context;
12896
12897         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12898                 printf("%s: ISC DMA read failed with error %d", __func__,
12899                        rq->ret);
12900                 ctl_set_internal_failure(&io->scsiio,
12901                                          /*sks_valid*/ 1,
12902                                          /*retry_count*/ rq->ret);
12903         }
12904
12905         ctl_dt_req_free(rq);
12906
12907         /* Switch the pointer over so the FETD knows what to do */
12908         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12909
12910         /*
12911          * Use a custom move done callback, since we need to send completion
12912          * back to the other controller, not to the backend on this side.
12913          */
12914         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12915
12916         /* XXX KDM add checks like the ones in ctl_datamove? */
12917
12918         fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12919
12920         fe_datamove(io);
12921 }
12922
12923 static int
12924 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12925 {
12926         struct ctl_sg_entry *local_sglist, *remote_sglist;
12927         struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12928         struct ctl_softc *softc;
12929         int retval;
12930         int i;
12931
12932         retval = 0;
12933         softc = control_softc;
12934
12935         local_sglist = io->io_hdr.local_sglist;
12936         local_dma_sglist = io->io_hdr.local_dma_sglist;
12937         remote_sglist = io->io_hdr.remote_sglist;
12938         remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12939
12940         if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12941                 for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12942                         local_sglist[i].len = remote_sglist[i].len;
12943
12944                         /*
12945                          * XXX Detect the situation where the RS-level I/O
12946                          * redirector on the other side has already read the
12947                          * data off of the AOR RS on this side, and
12948                          * transferred it to remote (mirror) memory on the
12949                          * other side.  Since we already have the data in
12950                          * memory here, we just need to use it.
12951                          *
12952                          * XXX KDM this can probably be removed once we
12953                          * get the cache device code in and take the
12954                          * current AOR implementation out.
12955                          */
12956 #ifdef NEEDTOPORT
12957                         if ((remote_sglist[i].addr >=
12958                              (void *)vtophys(softc->mirr->addr))
12959                          && (remote_sglist[i].addr <
12960                              ((void *)vtophys(softc->mirr->addr) +
12961                              CacheMirrorOffset))) {
12962                                 local_sglist[i].addr = remote_sglist[i].addr -
12963                                         CacheMirrorOffset;
12964                                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12965                                      CTL_FLAG_DATA_IN)
12966                                         io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12967                         } else {
12968                                 local_sglist[i].addr = remote_sglist[i].addr +
12969                                         CacheMirrorOffset;
12970                         }
12971 #endif
12972 #if 0
12973                         printf("%s: local %p, remote %p, len %d\n",
12974                                __func__, local_sglist[i].addr,
12975                                remote_sglist[i].addr, local_sglist[i].len);
12976 #endif
12977                 }
12978         } else {
12979                 uint32_t len_to_go;
12980
12981                 /*
12982                  * In this case, we don't have automatically allocated
12983                  * memory for this I/O on this controller.  This typically
12984                  * happens with internal CTL I/O -- e.g. inquiry, mode
12985                  * sense, etc.  Anything coming from RAIDCore will have
12986                  * a mirror area available.
12987                  */
12988                 len_to_go = io->scsiio.kern_data_len;
12989
12990                 /*
12991                  * Clear the no datasync flag, we have to use malloced
12992                  * buffers.
12993                  */
12994                 io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12995
12996                 /*
12997                  * The difficult thing here is that the size of the various
12998                  * S/G segments may be different than the size from the
12999                  * remote controller.  That'll make it harder when DMAing
13000                  * the data back to the other side.
13001                  */
13002                 for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13003                      sizeof(io->io_hdr.remote_sglist[0])) &&
13004                      (len_to_go > 0); i++) {
13005                         local_sglist[i].len = ctl_min(len_to_go, 131072);
13006                         CTL_SIZE_8B(local_dma_sglist[i].len,
13007                                     local_sglist[i].len);
13008                         local_sglist[i].addr =
13009                                 malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13010
13011                         local_dma_sglist[i].addr = local_sglist[i].addr;
13012
13013                         if (local_sglist[i].addr == NULL) {
13014                                 int j;
13015
13016                                 printf("malloc failed for %zd bytes!",
13017                                        local_dma_sglist[i].len);
13018                                 for (j = 0; j < i; j++) {
13019                                         free(local_sglist[j].addr, M_CTL);
13020                                 }
13021                                 ctl_set_internal_failure(&io->scsiio,
13022                                                          /*sks_valid*/ 1,
13023                                                          /*retry_count*/ 4857);
13024                                 retval = 1;
13025                                 goto bailout_error;
13026
13027                         }
13028                         /* XXX KDM do we need a sync here? */
13029
13030                         len_to_go -= local_sglist[i].len;
13031                 }
13032                 /*
13033                  * Reset the number of S/G entries accordingly.  The
13034                  * original number of S/G entries is available in
13035                  * rem_sg_entries.
13036                  */
13037                 io->scsiio.kern_sg_entries = i;
13038
13039 #if 0
13040                 printf("%s: kern_sg_entries = %d\n", __func__,
13041                        io->scsiio.kern_sg_entries);
13042                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13043                         printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13044                                local_sglist[i].addr, local_sglist[i].len,
13045                                local_dma_sglist[i].len);
13046 #endif
13047         }
13048
13049
13050         return (retval);
13051
13052 bailout_error:
13053
13054         ctl_send_datamove_done(io, /*have_lock*/ 0);
13055
13056         return (retval);
13057 }
13058
13059 static int
13060 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13061                          ctl_ha_dt_cb callback)
13062 {
13063         struct ctl_ha_dt_req *rq;
13064         struct ctl_sg_entry *remote_sglist, *local_sglist;
13065         struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13066         uint32_t local_used, remote_used, total_used;
13067         int retval;
13068         int i, j;
13069
13070         retval = 0;
13071
13072         rq = ctl_dt_req_alloc();
13073
13074         /*
13075          * If we failed to allocate the request, and if the DMA didn't fail
13076          * anyway, set busy status.  This is just a resource allocation
13077          * failure.
13078          */
13079         if ((rq == NULL)
13080          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13081                 ctl_set_busy(&io->scsiio);
13082
13083         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13084
13085                 if (rq != NULL)
13086                         ctl_dt_req_free(rq);
13087
13088                 /*
13089                  * The data move failed.  We need to return status back
13090                  * to the other controller.  No point in trying to DMA
13091                  * data to the remote controller.
13092                  */
13093
13094                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13095
13096                 retval = 1;
13097
13098                 goto bailout;
13099         }
13100
13101         local_sglist = io->io_hdr.local_sglist;
13102         local_dma_sglist = io->io_hdr.local_dma_sglist;
13103         remote_sglist = io->io_hdr.remote_sglist;
13104         remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13105         local_used = 0;
13106         remote_used = 0;
13107         total_used = 0;
13108
13109         if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13110                 rq->ret = CTL_HA_STATUS_SUCCESS;
13111                 rq->context = io;
13112                 callback(rq);
13113                 goto bailout;
13114         }
13115
13116         /*
13117          * Pull/push the data over the wire from/to the other controller.
13118          * This takes into account the possibility that the local and
13119          * remote sglists may not be identical in terms of the size of
13120          * the elements and the number of elements.
13121          *
13122          * One fundamental assumption here is that the length allocated for
13123          * both the local and remote sglists is identical.  Otherwise, we've
13124          * essentially got a coding error of some sort.
13125          */
13126         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13127                 int isc_ret;
13128                 uint32_t cur_len, dma_length;
13129                 uint8_t *tmp_ptr;
13130
13131                 rq->id = CTL_HA_DATA_CTL;
13132                 rq->command = command;
13133                 rq->context = io;
13134
13135                 /*
13136                  * Both pointers should be aligned.  But it is possible
13137                  * that the allocation length is not.  They should both
13138                  * also have enough slack left over at the end, though,
13139                  * to round up to the next 8 byte boundary.
13140                  */
13141                 cur_len = ctl_min(local_sglist[i].len - local_used,
13142                                   remote_sglist[j].len - remote_used);
13143
13144                 /*
13145                  * In this case, we have a size issue and need to decrease
13146                  * the size, except in the case where we actually have less
13147                  * than 8 bytes left.  In that case, we need to increase
13148                  * the DMA length to get the last bit.
13149                  */
13150                 if ((cur_len & 0x7) != 0) {
13151                         if (cur_len > 0x7) {
13152                                 cur_len = cur_len - (cur_len & 0x7);
13153                                 dma_length = cur_len;
13154                         } else {
13155                                 CTL_SIZE_8B(dma_length, cur_len);
13156                         }
13157
13158                 } else
13159                         dma_length = cur_len;
13160
13161                 /*
13162                  * If we had to allocate memory for this I/O, instead of using
13163                  * the non-cached mirror memory, we'll need to flush the cache
13164                  * before trying to DMA to the other controller.
13165                  *
13166                  * We could end up doing this multiple times for the same
13167                  * segment if we have a larger local segment than remote
13168                  * segment.  That shouldn't be an issue.
13169                  */
13170                 if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13171                         /*
13172                          * XXX KDM use bus_dmamap_sync() here.
13173                          */
13174                 }
13175
13176                 rq->size = dma_length;
13177
13178                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
13179                 tmp_ptr += local_used;
13180
13181                 /* Use physical addresses when talking to ISC hardware */
13182                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13183                         /* XXX KDM use busdma */
13184 #if 0
13185                         rq->local = vtophys(tmp_ptr);
13186 #endif
13187                 } else
13188                         rq->local = tmp_ptr;
13189
13190                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13191                 tmp_ptr += remote_used;
13192                 rq->remote = tmp_ptr;
13193
13194                 rq->callback = NULL;
13195
13196                 local_used += cur_len;
13197                 if (local_used >= local_sglist[i].len) {
13198                         i++;
13199                         local_used = 0;
13200                 }
13201
13202                 remote_used += cur_len;
13203                 if (remote_used >= remote_sglist[j].len) {
13204                         j++;
13205                         remote_used = 0;
13206                 }
13207                 total_used += cur_len;
13208
13209                 if (total_used >= io->scsiio.kern_data_len)
13210                         rq->callback = callback;
13211
13212                 if ((rq->size & 0x7) != 0) {
13213                         printf("%s: warning: size %d is not on 8b boundary\n",
13214                                __func__, rq->size);
13215                 }
13216                 if (((uintptr_t)rq->local & 0x7) != 0) {
13217                         printf("%s: warning: local %p not on 8b boundary\n",
13218                                __func__, rq->local);
13219                 }
13220                 if (((uintptr_t)rq->remote & 0x7) != 0) {
13221                         printf("%s: warning: remote %p not on 8b boundary\n",
13222                                __func__, rq->local);
13223                 }
13224 #if 0
13225                 printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13226                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13227                        rq->local, rq->remote, rq->size);
13228 #endif
13229
13230                 isc_ret = ctl_dt_single(rq);
13231                 if (isc_ret == CTL_HA_STATUS_WAIT)
13232                         continue;
13233
13234                 if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13235                         rq->ret = CTL_HA_STATUS_SUCCESS;
13236                 } else {
13237                         rq->ret = isc_ret;
13238                 }
13239                 callback(rq);
13240                 goto bailout;
13241         }
13242
13243 bailout:
13244         return (retval);
13245
13246 }
13247
13248 static void
13249 ctl_datamove_remote_read(union ctl_io *io)
13250 {
13251         int retval;
13252         int i;
13253
13254         /*
13255          * This will send an error to the other controller in the case of a
13256          * failure.
13257          */
13258         retval = ctl_datamove_remote_sgl_setup(io);
13259         if (retval != 0)
13260                 return;
13261
13262         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13263                                           ctl_datamove_remote_read_cb);
13264         if ((retval != 0)
13265          && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13266                 /*
13267                  * Make sure we free memory if there was an error..  The
13268                  * ctl_datamove_remote_xfer() function will send the
13269                  * datamove done message, or call the callback with an
13270                  * error if there is a problem.
13271                  */
13272                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13273                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13274         }
13275
13276         return;
13277 }
13278
13279 /*
13280  * Process a datamove request from the other controller.  This is used for
13281  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13282  * first.  Once that is complete, the data gets DMAed into the remote
13283  * controller's memory.  For reads, we DMA from the remote controller's
13284  * memory into our memory first, and then move it out to the FETD.
13285  */
13286 static void
13287 ctl_datamove_remote(union ctl_io *io)
13288 {
13289         struct ctl_softc *softc;
13290
13291         softc = control_softc;
13292
13293         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13294
13295         /*
13296          * Note that we look for an aborted I/O here, but don't do some of
13297          * the other checks that ctl_datamove() normally does.
13298          * We don't need to run the datamove delay code, since that should
13299          * have been done if need be on the other controller.
13300          */
13301         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13302                 printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13303                        io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13304                        io->io_hdr.nexus.targ_port,
13305                        io->io_hdr.nexus.targ_target.id,
13306                        io->io_hdr.nexus.targ_lun);
13307                 io->io_hdr.port_status = 31338;
13308                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13309                 return;
13310         }
13311
13312         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13313                 ctl_datamove_remote_write(io);
13314         } else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13315                 ctl_datamove_remote_read(io);
13316         } else {
13317                 union ctl_ha_msg msg;
13318                 struct scsi_sense_data *sense;
13319                 uint8_t sks[3];
13320                 int retry_count;
13321
13322                 memset(&msg, 0, sizeof(msg));
13323
13324                 msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13325                 msg.hdr.status = CTL_SCSI_ERROR;
13326                 msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13327
13328                 retry_count = 4243;
13329
13330                 sense = &msg.scsi.sense_data;
13331                 sks[0] = SSD_SCS_VALID;
13332                 sks[1] = (retry_count >> 8) & 0xff;
13333                 sks[2] = retry_count & 0xff;
13334
13335                 /* "Internal target failure" */
13336                 scsi_set_sense_data(sense,
13337                                     /*sense_format*/ SSD_TYPE_NONE,
13338                                     /*current_error*/ 1,
13339                                     /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13340                                     /*asc*/ 0x44,
13341                                     /*ascq*/ 0x00,
13342                                     /*type*/ SSD_ELEM_SKS,
13343                                     /*size*/ sizeof(sks),
13344                                     /*data*/ sks,
13345                                     SSD_ELEM_NONE);
13346
13347                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13348                 if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13349                         ctl_failover_io(io, /*have_lock*/ 1);
13350                         return;
13351                 }
13352
13353                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13354                     CTL_HA_STATUS_SUCCESS) {
13355                         /* XXX KDM what to do if this fails? */
13356                 }
13357                 return;
13358         }
13359
13360 }
13361
13362 static int
13363 ctl_process_done(union ctl_io *io)
13364 {
13365         struct ctl_lun *lun;
13366         struct ctl_softc *ctl_softc;
13367         void (*fe_done)(union ctl_io *io);
13368         uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13369
13370         CTL_DEBUG_PRINT(("ctl_process_done\n"));
13371
13372         fe_done =
13373             control_softc->ctl_ports[targ_port]->fe_done;
13374
13375 #ifdef CTL_TIME_IO
13376         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13377                 char str[256];
13378                 char path_str[64];
13379                 struct sbuf sb;
13380
13381                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
13382                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13383
13384                 sbuf_cat(&sb, path_str);
13385                 switch (io->io_hdr.io_type) {
13386                 case CTL_IO_SCSI:
13387                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13388                         sbuf_printf(&sb, "\n");
13389                         sbuf_cat(&sb, path_str);
13390                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13391                                     io->scsiio.tag_num, io->scsiio.tag_type);
13392                         break;
13393                 case CTL_IO_TASK:
13394                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13395                                     "Tag Type: %d\n", io->taskio.task_action,
13396                                     io->taskio.tag_num, io->taskio.tag_type);
13397                         break;
13398                 default:
13399                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13400                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13401                         break;
13402                 }
13403                 sbuf_cat(&sb, path_str);
13404                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13405                             (intmax_t)time_uptime - io->io_hdr.start_time);
13406                 sbuf_finish(&sb);
13407                 printf("%s", sbuf_data(&sb));
13408         }
13409 #endif /* CTL_TIME_IO */
13410
13411         switch (io->io_hdr.io_type) {
13412         case CTL_IO_SCSI:
13413                 break;
13414         case CTL_IO_TASK:
13415                 if (bootverbose || verbose > 0)
13416                         ctl_io_error_print(io, NULL);
13417                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13418                         ctl_free_io(io);
13419                 else
13420                         fe_done(io);
13421                 return (CTL_RETVAL_COMPLETE);
13422                 break;
13423         default:
13424                 printf("ctl_process_done: invalid io type %d\n",
13425                        io->io_hdr.io_type);
13426                 panic("ctl_process_done: invalid io type %d\n",
13427                       io->io_hdr.io_type);
13428                 break; /* NOTREACHED */
13429         }
13430
13431         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13432         if (lun == NULL) {
13433                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13434                                  io->io_hdr.nexus.targ_mapped_lun));
13435                 fe_done(io);
13436                 goto bailout;
13437         }
13438         ctl_softc = lun->ctl_softc;
13439
13440         mtx_lock(&lun->lun_lock);
13441
13442         /*
13443          * Check to see if we have any errors to inject here.  We only
13444          * inject errors for commands that don't already have errors set.
13445          */
13446         if ((STAILQ_FIRST(&lun->error_list) != NULL)
13447          && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13448                 ctl_inject_error(lun, io);
13449
13450         /*
13451          * XXX KDM how do we treat commands that aren't completed
13452          * successfully?
13453          *
13454          * XXX KDM should we also track I/O latency?
13455          */
13456         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13457             io->io_hdr.io_type == CTL_IO_SCSI) {
13458 #ifdef CTL_TIME_IO
13459                 struct bintime cur_bt;
13460 #endif
13461                 int type;
13462
13463                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13464                     CTL_FLAG_DATA_IN)
13465                         type = CTL_STATS_READ;
13466                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13467                     CTL_FLAG_DATA_OUT)
13468                         type = CTL_STATS_WRITE;
13469                 else
13470                         type = CTL_STATS_NO_IO;
13471
13472                 lun->stats.ports[targ_port].bytes[type] +=
13473                     io->scsiio.kern_total_len;
13474                 lun->stats.ports[targ_port].operations[type]++;
13475 #ifdef CTL_TIME_IO
13476                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13477                    &io->io_hdr.dma_bt);
13478                 lun->stats.ports[targ_port].num_dmas[type] +=
13479                     io->io_hdr.num_dmas;
13480                 getbintime(&cur_bt);
13481                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13482                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13483 #endif
13484         }
13485
13486         /*
13487          * Remove this from the OOA queue.
13488          */
13489         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13490
13491         /*
13492          * Run through the blocked queue on this LUN and see if anything
13493          * has become unblocked, now that this transaction is done.
13494          */
13495         ctl_check_blocked(lun);
13496
13497         /*
13498          * If the LUN has been invalidated, free it if there is nothing
13499          * left on its OOA queue.
13500          */
13501         if ((lun->flags & CTL_LUN_INVALID)
13502          && TAILQ_EMPTY(&lun->ooa_queue)) {
13503                 mtx_unlock(&lun->lun_lock);
13504                 mtx_lock(&ctl_softc->ctl_lock);
13505                 ctl_free_lun(lun);
13506                 mtx_unlock(&ctl_softc->ctl_lock);
13507         } else
13508                 mtx_unlock(&lun->lun_lock);
13509
13510         /*
13511          * If this command has been aborted, make sure we set the status
13512          * properly.  The FETD is responsible for freeing the I/O and doing
13513          * whatever it needs to do to clean up its state.
13514          */
13515         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13516                 ctl_set_task_aborted(&io->scsiio);
13517
13518         /*
13519          * We print out status for every task management command.  For SCSI
13520          * commands, we filter out any unit attention errors; they happen
13521          * on every boot, and would clutter up the log.  Note:  task
13522          * management commands aren't printed here, they are printed above,
13523          * since they should never even make it down here.
13524          */
13525         switch (io->io_hdr.io_type) {
13526         case CTL_IO_SCSI: {
13527                 int error_code, sense_key, asc, ascq;
13528
13529                 sense_key = 0;
13530
13531                 if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13532                  && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13533                         /*
13534                          * Since this is just for printing, no need to
13535                          * show errors here.
13536                          */
13537                         scsi_extract_sense_len(&io->scsiio.sense_data,
13538                                                io->scsiio.sense_len,
13539                                                &error_code,
13540                                                &sense_key,
13541                                                &asc,
13542                                                &ascq,
13543                                                /*show_errors*/ 0);
13544                 }
13545
13546                 if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13547                  && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13548                   || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13549                   || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13550
13551                         if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13552                                 ctl_softc->skipped_prints++;
13553                         } else {
13554                                 uint32_t skipped_prints;
13555
13556                                 skipped_prints = ctl_softc->skipped_prints;
13557
13558                                 ctl_softc->skipped_prints = 0;
13559                                 ctl_softc->last_print_jiffies = time_uptime;
13560
13561                                 if (skipped_prints > 0) {
13562 #ifdef NEEDTOPORT
13563                                         csevent_log(CSC_CTL | CSC_SHELF_SW |
13564                                             CTL_ERROR_REPORT,
13565                                             csevent_LogType_Trace,
13566                                             csevent_Severity_Information,
13567                                             csevent_AlertLevel_Green,
13568                                             csevent_FRU_Firmware,
13569                                             csevent_FRU_Unknown,
13570                                             "High CTL error volume, %d prints "
13571                                             "skipped", skipped_prints);
13572 #endif
13573                                 }
13574                                 if (bootverbose || verbose > 0)
13575                                         ctl_io_error_print(io, NULL);
13576                         }
13577                 }
13578                 break;
13579         }
13580         case CTL_IO_TASK:
13581                 if (bootverbose || verbose > 0)
13582                         ctl_io_error_print(io, NULL);
13583                 break;
13584         default:
13585                 break;
13586         }
13587
13588         /*
13589          * Tell the FETD or the other shelf controller we're done with this
13590          * command.  Note that only SCSI commands get to this point.  Task
13591          * management commands are completed above.
13592          *
13593          * We only send status to the other controller if we're in XFER
13594          * mode.  In SER_ONLY mode, the I/O is done on the controller that
13595          * received the I/O (from CTL's perspective), and so the status is
13596          * generated there.
13597          *
13598          * XXX KDM if we hold the lock here, we could cause a deadlock
13599          * if the frontend comes back in in this context to queue
13600          * something.
13601          */
13602         if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13603          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13604                 union ctl_ha_msg msg;
13605
13606                 memset(&msg, 0, sizeof(msg));
13607                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13608                 msg.hdr.original_sc = io->io_hdr.original_sc;
13609                 msg.hdr.nexus = io->io_hdr.nexus;
13610                 msg.hdr.status = io->io_hdr.status;
13611                 msg.scsi.scsi_status = io->scsiio.scsi_status;
13612                 msg.scsi.tag_num = io->scsiio.tag_num;
13613                 msg.scsi.tag_type = io->scsiio.tag_type;
13614                 msg.scsi.sense_len = io->scsiio.sense_len;
13615                 msg.scsi.sense_residual = io->scsiio.sense_residual;
13616                 msg.scsi.residual = io->scsiio.residual;
13617                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13618                        sizeof(io->scsiio.sense_data));
13619                 /*
13620                  * We copy this whether or not this is an I/O-related
13621                  * command.  Otherwise, we'd have to go and check to see
13622                  * whether it's a read/write command, and it really isn't
13623                  * worth it.
13624                  */
13625                 memcpy(&msg.scsi.lbalen,
13626                        &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13627                        sizeof(msg.scsi.lbalen));
13628
13629                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13630                                 sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13631                         /* XXX do something here */
13632                 }
13633
13634                 ctl_free_io(io);
13635         } else
13636                 fe_done(io);
13637
13638 bailout:
13639
13640         return (CTL_RETVAL_COMPLETE);
13641 }
13642
13643 #ifdef CTL_WITH_CA
13644 /*
13645  * Front end should call this if it doesn't do autosense.  When the request
13646  * sense comes back in from the initiator, we'll dequeue this and send it.
13647  */
13648 int
13649 ctl_queue_sense(union ctl_io *io)
13650 {
13651         struct ctl_lun *lun;
13652         struct ctl_softc *ctl_softc;
13653         uint32_t initidx, targ_lun;
13654
13655         ctl_softc = control_softc;
13656
13657         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13658
13659         /*
13660          * LUN lookup will likely move to the ctl_work_thread() once we
13661          * have our new queueing infrastructure (that doesn't put things on
13662          * a per-LUN queue initially).  That is so that we can handle
13663          * things like an INQUIRY to a LUN that we don't have enabled.  We
13664          * can't deal with that right now.
13665          */
13666         mtx_lock(&ctl_softc->ctl_lock);
13667
13668         /*
13669          * If we don't have a LUN for this, just toss the sense
13670          * information.
13671          */
13672         targ_lun = io->io_hdr.nexus.targ_lun;
13673         targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13674         if ((targ_lun < CTL_MAX_LUNS)
13675          && (ctl_softc->ctl_luns[targ_lun] != NULL))
13676                 lun = ctl_softc->ctl_luns[targ_lun];
13677         else
13678                 goto bailout;
13679
13680         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13681
13682         mtx_lock(&lun->lun_lock);
13683         /*
13684          * Already have CA set for this LUN...toss the sense information.
13685          */
13686         if (ctl_is_set(lun->have_ca, initidx)) {
13687                 mtx_unlock(&lun->lun_lock);
13688                 goto bailout;
13689         }
13690
13691         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13692                ctl_min(sizeof(lun->pending_sense[initidx]),
13693                sizeof(io->scsiio.sense_data)));
13694         ctl_set_mask(lun->have_ca, initidx);
13695         mtx_unlock(&lun->lun_lock);
13696
13697 bailout:
13698         mtx_unlock(&ctl_softc->ctl_lock);
13699
13700         ctl_free_io(io);
13701
13702         return (CTL_RETVAL_COMPLETE);
13703 }
13704 #endif
13705
13706 /*
13707  * Primary command inlet from frontend ports.  All SCSI and task I/O
13708  * requests must go through this function.
13709  */
13710 int
13711 ctl_queue(union ctl_io *io)
13712 {
13713         struct ctl_softc *ctl_softc;
13714
13715         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13716
13717         ctl_softc = control_softc;
13718
13719 #ifdef CTL_TIME_IO
13720         io->io_hdr.start_time = time_uptime;
13721         getbintime(&io->io_hdr.start_bt);
13722 #endif /* CTL_TIME_IO */
13723
13724         /* Map FE-specific LUN ID into global one. */
13725         io->io_hdr.nexus.targ_mapped_lun =
13726             ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13727
13728         switch (io->io_hdr.io_type) {
13729         case CTL_IO_SCSI:
13730         case CTL_IO_TASK:
13731                 ctl_enqueue_incoming(io);
13732                 break;
13733         default:
13734                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13735                 return (EINVAL);
13736         }
13737
13738         return (CTL_RETVAL_COMPLETE);
13739 }
13740
13741 #ifdef CTL_IO_DELAY
13742 static void
13743 ctl_done_timer_wakeup(void *arg)
13744 {
13745         union ctl_io *io;
13746
13747         io = (union ctl_io *)arg;
13748         ctl_done(io);
13749 }
13750 #endif /* CTL_IO_DELAY */
13751
13752 void
13753 ctl_done(union ctl_io *io)
13754 {
13755         struct ctl_softc *ctl_softc;
13756
13757         ctl_softc = control_softc;
13758
13759         /*
13760          * Enable this to catch duplicate completion issues.
13761          */
13762 #if 0
13763         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13764                 printf("%s: type %d msg %d cdb %x iptl: "
13765                        "%d:%d:%d:%d tag 0x%04x "
13766                        "flag %#x status %x\n",
13767                         __func__,
13768                         io->io_hdr.io_type,
13769                         io->io_hdr.msg_type,
13770                         io->scsiio.cdb[0],
13771                         io->io_hdr.nexus.initid.id,
13772                         io->io_hdr.nexus.targ_port,
13773                         io->io_hdr.nexus.targ_target.id,
13774                         io->io_hdr.nexus.targ_lun,
13775                         (io->io_hdr.io_type ==
13776                         CTL_IO_TASK) ?
13777                         io->taskio.tag_num :
13778                         io->scsiio.tag_num,
13779                         io->io_hdr.flags,
13780                         io->io_hdr.status);
13781         } else
13782                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13783 #endif
13784
13785         /*
13786          * This is an internal copy of an I/O, and should not go through
13787          * the normal done processing logic.
13788          */
13789         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13790                 return;
13791
13792         /*
13793          * We need to send a msg to the serializing shelf to finish the IO
13794          * as well.  We don't send a finish message to the other shelf if
13795          * this is a task management command.  Task management commands
13796          * aren't serialized in the OOA queue, but rather just executed on
13797          * both shelf controllers for commands that originated on that
13798          * controller.
13799          */
13800         if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13801          && (io->io_hdr.io_type != CTL_IO_TASK)) {
13802                 union ctl_ha_msg msg_io;
13803
13804                 msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13805                 msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13806                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13807                     sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13808                 }
13809                 /* continue on to finish IO */
13810         }
13811 #ifdef CTL_IO_DELAY
13812         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13813                 struct ctl_lun *lun;
13814
13815                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13816
13817                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13818         } else {
13819                 struct ctl_lun *lun;
13820
13821                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13822
13823                 if ((lun != NULL)
13824                  && (lun->delay_info.done_delay > 0)) {
13825                         struct callout *callout;
13826
13827                         callout = (struct callout *)&io->io_hdr.timer_bytes;
13828                         callout_init(callout, /*mpsafe*/ 1);
13829                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13830                         callout_reset(callout,
13831                                       lun->delay_info.done_delay * hz,
13832                                       ctl_done_timer_wakeup, io);
13833                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13834                                 lun->delay_info.done_delay = 0;
13835                         return;
13836                 }
13837         }
13838 #endif /* CTL_IO_DELAY */
13839
13840         ctl_enqueue_done(io);
13841 }
13842
13843 int
13844 ctl_isc(struct ctl_scsiio *ctsio)
13845 {
13846         struct ctl_lun *lun;
13847         int retval;
13848
13849         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13850
13851         CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13852
13853         CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13854
13855         retval = lun->backend->data_submit((union ctl_io *)ctsio);
13856
13857         return (retval);
13858 }
13859
13860
13861 static void
13862 ctl_work_thread(void *arg)
13863 {
13864         struct ctl_thread *thr = (struct ctl_thread *)arg;
13865         struct ctl_softc *softc = thr->ctl_softc;
13866         union ctl_io *io;
13867         int retval;
13868
13869         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13870
13871         for (;;) {
13872                 retval = 0;
13873
13874                 /*
13875                  * We handle the queues in this order:
13876                  * - ISC
13877                  * - done queue (to free up resources, unblock other commands)
13878                  * - RtR queue
13879                  * - incoming queue
13880                  *
13881                  * If those queues are empty, we break out of the loop and
13882                  * go to sleep.
13883                  */
13884                 mtx_lock(&thr->queue_lock);
13885                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13886                 if (io != NULL) {
13887                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13888                         mtx_unlock(&thr->queue_lock);
13889                         ctl_handle_isc(io);
13890                         continue;
13891                 }
13892                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13893                 if (io != NULL) {
13894                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13895                         /* clear any blocked commands, call fe_done */
13896                         mtx_unlock(&thr->queue_lock);
13897                         retval = ctl_process_done(io);
13898                         continue;
13899                 }
13900                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13901                 if (io != NULL) {
13902                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13903                         mtx_unlock(&thr->queue_lock);
13904                         if (io->io_hdr.io_type == CTL_IO_TASK)
13905                                 ctl_run_task(io);
13906                         else
13907                                 ctl_scsiio_precheck(softc, &io->scsiio);
13908                         continue;
13909                 }
13910                 if (!ctl_pause_rtr) {
13911                         io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13912                         if (io != NULL) {
13913                                 STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13914                                 mtx_unlock(&thr->queue_lock);
13915                                 retval = ctl_scsiio(&io->scsiio);
13916                                 if (retval != CTL_RETVAL_COMPLETE)
13917                                         CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13918                                 continue;
13919                         }
13920                 }
13921
13922                 /* Sleep until we have something to do. */
13923                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13924         }
13925 }
13926
13927 static void
13928 ctl_lun_thread(void *arg)
13929 {
13930         struct ctl_softc *softc = (struct ctl_softc *)arg;
13931         struct ctl_be_lun *be_lun;
13932         int retval;
13933
13934         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13935
13936         for (;;) {
13937                 retval = 0;
13938                 mtx_lock(&softc->ctl_lock);
13939                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13940                 if (be_lun != NULL) {
13941                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13942                         mtx_unlock(&softc->ctl_lock);
13943                         ctl_create_lun(be_lun);
13944                         continue;
13945                 }
13946
13947                 /* Sleep until we have something to do. */
13948                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13949                     PDROP | PRIBIO, "-", 0);
13950         }
13951 }
13952
13953 static void
13954 ctl_enqueue_incoming(union ctl_io *io)
13955 {
13956         struct ctl_softc *softc = control_softc;
13957         struct ctl_thread *thr;
13958         u_int idx;
13959
13960         idx = (io->io_hdr.nexus.targ_port * 127 +
13961                io->io_hdr.nexus.initid.id) % worker_threads;
13962         thr = &softc->threads[idx];
13963         mtx_lock(&thr->queue_lock);
13964         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13965         mtx_unlock(&thr->queue_lock);
13966         wakeup(thr);
13967 }
13968
13969 static void
13970 ctl_enqueue_rtr(union ctl_io *io)
13971 {
13972         struct ctl_softc *softc = control_softc;
13973         struct ctl_thread *thr;
13974
13975         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13976         mtx_lock(&thr->queue_lock);
13977         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13978         mtx_unlock(&thr->queue_lock);
13979         wakeup(thr);
13980 }
13981
13982 static void
13983 ctl_enqueue_done(union ctl_io *io)
13984 {
13985         struct ctl_softc *softc = control_softc;
13986         struct ctl_thread *thr;
13987
13988         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13989         mtx_lock(&thr->queue_lock);
13990         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13991         mtx_unlock(&thr->queue_lock);
13992         wakeup(thr);
13993 }
13994
13995 static void
13996 ctl_enqueue_isc(union ctl_io *io)
13997 {
13998         struct ctl_softc *softc = control_softc;
13999         struct ctl_thread *thr;
14000
14001         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14002         mtx_lock(&thr->queue_lock);
14003         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14004         mtx_unlock(&thr->queue_lock);
14005         wakeup(thr);
14006 }
14007
14008 /* Initialization and failover */
14009
14010 void
14011 ctl_init_isc_msg(void)
14012 {
14013         printf("CTL: Still calling this thing\n");
14014 }
14015
14016 /*
14017  * Init component
14018  *      Initializes component into configuration defined by bootMode
14019  *      (see hasc-sv.c)
14020  *      returns hasc_Status:
14021  *              OK
14022  *              ERROR - fatal error
14023  */
14024 static ctl_ha_comp_status
14025 ctl_isc_init(struct ctl_ha_component *c)
14026 {
14027         ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14028
14029         c->status = ret;
14030         return ret;
14031 }
14032
14033 /* Start component
14034  *      Starts component in state requested. If component starts successfully,
14035  *      it must set its own state to the requestrd state
14036  *      When requested state is HASC_STATE_HA, the component may refine it
14037  *      by adding _SLAVE or _MASTER flags.
14038  *      Currently allowed state transitions are:
14039  *      UNKNOWN->HA             - initial startup
14040  *      UNKNOWN->SINGLE - initial startup when no parter detected
14041  *      HA->SINGLE              - failover
14042  * returns ctl_ha_comp_status:
14043  *              OK      - component successfully started in requested state
14044  *              FAILED  - could not start the requested state, failover may
14045  *                        be possible
14046  *              ERROR   - fatal error detected, no future startup possible
14047  */
14048 static ctl_ha_comp_status
14049 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14050 {
14051         ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14052
14053         printf("%s: go\n", __func__);
14054
14055         // UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14056         if (c->state == CTL_HA_STATE_UNKNOWN ) {
14057                 ctl_is_single = 0;
14058                 if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14059                     != CTL_HA_STATUS_SUCCESS) {
14060                         printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14061                         ret = CTL_HA_COMP_STATUS_ERROR;
14062                 }
14063         } else if (CTL_HA_STATE_IS_HA(c->state)
14064                 && CTL_HA_STATE_IS_SINGLE(state)){
14065                 // HA->SINGLE transition
14066                 ctl_failover();
14067                 ctl_is_single = 1;
14068         } else {
14069                 printf("ctl_isc_start:Invalid state transition %X->%X\n",
14070                        c->state, state);
14071                 ret = CTL_HA_COMP_STATUS_ERROR;
14072         }
14073         if (CTL_HA_STATE_IS_SINGLE(state))
14074                 ctl_is_single = 1;
14075
14076         c->state = state;
14077         c->status = ret;
14078         return ret;
14079 }
14080
14081 /*
14082  * Quiesce component
14083  * The component must clear any error conditions (set status to OK) and
14084  * prepare itself to another Start call
14085  * returns ctl_ha_comp_status:
14086  *      OK
14087  *      ERROR
14088  */
14089 static ctl_ha_comp_status
14090 ctl_isc_quiesce(struct ctl_ha_component *c)
14091 {
14092         int ret = CTL_HA_COMP_STATUS_OK;
14093
14094         ctl_pause_rtr = 1;
14095         c->status = ret;
14096         return ret;
14097 }
14098
14099 struct ctl_ha_component ctl_ha_component_ctlisc =
14100 {
14101         .name = "CTL ISC",
14102         .state = CTL_HA_STATE_UNKNOWN,
14103         .init = ctl_isc_init,
14104         .start = ctl_isc_start,
14105         .quiesce = ctl_isc_quiesce
14106 };
14107
14108 /*
14109  *  vim: ts=8
14110  */