sys/cam/ctl/ctl.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause
   3  *
   4  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
   5  * Copyright (c) 2012 The FreeBSD Foundation
   6  * Copyright (c) 2014-2017 Alexander Motin <mav@FreeBSD.org>
   7  * Copyright (c) 2017 Jakub Wojciech Klama <jceel@FreeBSD.org>
   8  * Copyright (c) 2018 Marcelo Araujo <araujo@FreeBSD.org>
   9  * All rights reserved.
  10  *
  11  * Portions of this software were developed by Edward Tomasz Napierala
  12  * under sponsorship from the FreeBSD Foundation.
  13  *
  14  * Redistribution and use in source and binary forms, with or without
  15  * modification, are permitted provided that the following conditions
  16  * are met:
  17  * 1. Redistributions of source code must retain the above copyright
  18  *    notice, this list of conditions, and the following disclaimer,
  19  *    without modification.
  20  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
  21  *    substantially similar to the "NO WARRANTY" disclaimer below
  22  *    ("Disclaimer") and any redistribution must be conditioned upon
  23  *    including a substantially similar Disclaimer requirement for further
  24  *    binary redistribution.
  25  *
  26  * NO WARRANTY
  27  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  28  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  29  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
  30  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  31  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  35  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
  36  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  37  * POSSIBILITY OF SUCH DAMAGES.
  38  *
  39  * $Id$
  40  */
  41 /*
  42  * CAM Target Layer, a SCSI device emulation subsystem.
  43  *
  44  * Author: Ken Merry <ken@FreeBSD.org>
  45  */
  46
  47 #include <sys/param.h>
  48 #include <sys/systm.h>
  49 #include <sys/ctype.h>
  50 #include <sys/kernel.h>
  51 #include <sys/types.h>
  52 #include <sys/kthread.h>
  53 #include <sys/bio.h>
  54 #include <sys/fcntl.h>
  55 #include <sys/lock.h>
  56 #include <sys/module.h>
  57 #include <sys/mutex.h>
  58 #include <sys/condvar.h>
  59 #include <sys/malloc.h>
  60 #include <sys/conf.h>
  61 #include <sys/ioccom.h>
  62 #include <sys/queue.h>
  63 #include <sys/sbuf.h>
  64 #include <sys/smp.h>
  65 #include <sys/endian.h>
  66 #include <sys/proc.h>
  67 #include <sys/sched.h>
  68 #include <sys/sysctl.h>
  69 #include <sys/nv.h>
  70 #include <sys/dnv.h>
  71 #include <vm/uma.h>
  72
  73 #include <cam/cam.h>
  74 #include <cam/scsi/scsi_all.h>
  75 #include <cam/scsi/scsi_cd.h>
  76 #include <cam/scsi/scsi_da.h>
  77 #include <cam/ctl/ctl_io.h>
  78 #include <cam/ctl/ctl.h>
  79 #include <cam/ctl/ctl_frontend.h>
  80 #include <cam/ctl/ctl_util.h>
  81 #include <cam/ctl/ctl_backend.h>
  82 #include <cam/ctl/ctl_ioctl.h>
  83 #include <cam/ctl/ctl_ha.h>
  84 #include <cam/ctl/ctl_private.h>
  85 #include <cam/ctl/ctl_debug.h>
  86 #include <cam/ctl/ctl_scsi_all.h>
  87 #include <cam/ctl/ctl_error.h>
  88
  89 struct ctl_softc *control_softc = NULL;
  90
  91 /*
  92  * Template mode pages.
  93  */
  94
  95 /*
  96  * Note that these are default values only.  The actual values will be
  97  * filled in when the user does a mode sense.
  98  */
  99 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
 100         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 101         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 102         /*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
 103         /*read_retry_count*/0,
 104         /*correction_span*/0,
 105         /*head_offset_count*/0,
 106         /*data_strobe_offset_cnt*/0,
 107         /*byte8*/SMS_RWER_LBPERE,
 108         /*write_retry_count*/0,
 109         /*reserved2*/0,
 110         /*recovery_time_limit*/{0, 0},
 111 };
 112
 113 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
 114         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 115         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 116         /*byte3*/SMS_RWER_PER,
 117         /*read_retry_count*/0,
 118         /*correction_span*/0,
 119         /*head_offset_count*/0,
 120         /*data_strobe_offset_cnt*/0,
 121         /*byte8*/SMS_RWER_LBPERE,
 122         /*write_retry_count*/0,
 123         /*reserved2*/0,
 124         /*recovery_time_limit*/{0, 0},
 125 };
 126
 127 const static struct scsi_da_verify_recovery_page verify_er_page_default = {
 128         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 129         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 130         /*byte3*/0,
 131         /*read_retry_count*/0,
 132         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 133         /*recovery_time_limit*/{0, 0},
 134 };
 135
 136 const static struct scsi_da_verify_recovery_page verify_er_page_changeable = {
 137         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 138         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 139         /*byte3*/SMS_VER_PER,
 140         /*read_retry_count*/0,
 141         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 142         /*recovery_time_limit*/{0, 0},
 143 };
 144
 145 const static struct scsi_caching_page caching_page_default = {
 146         /*page_code*/SMS_CACHING_PAGE,
 147         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 148         /*flags1*/ SCP_DISC | SCP_WCE,
 149         /*ret_priority*/ 0,
 150         /*disable_pf_transfer_len*/ {0xff, 0xff},
 151         /*min_prefetch*/ {0, 0},
 152         /*max_prefetch*/ {0xff, 0xff},
 153         /*max_pf_ceiling*/ {0xff, 0xff},
 154         /*flags2*/ 0,
 155         /*cache_segments*/ 0,
 156         /*cache_seg_size*/ {0, 0},
 157         /*reserved*/ 0,
 158         /*non_cache_seg_size*/ {0, 0, 0}
 159 };
 160
 161 const static struct scsi_caching_page caching_page_changeable = {
 162         /*page_code*/SMS_CACHING_PAGE,
 163         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 164         /*flags1*/ SCP_WCE | SCP_RCD,
 165         /*ret_priority*/ 0,
 166         /*disable_pf_transfer_len*/ {0, 0},
 167         /*min_prefetch*/ {0, 0},
 168         /*max_prefetch*/ {0, 0},
 169         /*max_pf_ceiling*/ {0, 0},
 170         /*flags2*/ 0,
 171         /*cache_segments*/ 0,
 172         /*cache_seg_size*/ {0, 0},
 173         /*reserved*/ 0,
 174         /*non_cache_seg_size*/ {0, 0, 0}
 175 };
 176
 177 const static struct scsi_control_page control_page_default = {
 178         /*page_code*/SMS_CONTROL_MODE_PAGE,
 179         /*page_length*/sizeof(struct scsi_control_page) - 2,
 180         /*rlec*/0,
 181         /*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
 182         /*eca_and_aen*/0,
 183         /*flags4*/SCP_TAS,
 184         /*aen_holdoff_period*/{0, 0},
 185         /*busy_timeout_period*/{0, 0},
 186         /*extended_selftest_completion_time*/{0, 0}
 187 };
 188
 189 const static struct scsi_control_page control_page_changeable = {
 190         /*page_code*/SMS_CONTROL_MODE_PAGE,
 191         /*page_length*/sizeof(struct scsi_control_page) - 2,
 192         /*rlec*/SCP_DSENSE,
 193         /*queue_flags*/SCP_QUEUE_ALG_MASK | SCP_NUAR,
 194         /*eca_and_aen*/SCP_SWP,
 195         /*flags4*/0,
 196         /*aen_holdoff_period*/{0, 0},
 197         /*busy_timeout_period*/{0, 0},
 198         /*extended_selftest_completion_time*/{0, 0}
 199 };
 200
 201 #define CTL_CEM_LEN     (sizeof(struct scsi_control_ext_page) - 4)
 202
 203 const static struct scsi_control_ext_page control_ext_page_default = {
 204         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 205         /*subpage_code*/0x01,
 206         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 207         /*flags*/0,
 208         /*prio*/0,
 209         /*max_sense*/0
 210 };
 211
 212 const static struct scsi_control_ext_page control_ext_page_changeable = {
 213         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 214         /*subpage_code*/0x01,
 215         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 216         /*flags*/0,
 217         /*prio*/0,
 218         /*max_sense*/0xff
 219 };
 220
 221 const static struct scsi_info_exceptions_page ie_page_default = {
 222         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 223         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 224         /*info_flags*/SIEP_FLAGS_EWASC,
 225         /*mrie*/SIEP_MRIE_NO,
 226         /*interval_timer*/{0, 0, 0, 0},
 227         /*report_count*/{0, 0, 0, 1}
 228 };
 229
 230 const static struct scsi_info_exceptions_page ie_page_changeable = {
 231         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 232         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 233         /*info_flags*/SIEP_FLAGS_EWASC | SIEP_FLAGS_DEXCPT | SIEP_FLAGS_TEST |
 234             SIEP_FLAGS_LOGERR,
 235         /*mrie*/0x0f,
 236         /*interval_timer*/{0xff, 0xff, 0xff, 0xff},
 237         /*report_count*/{0xff, 0xff, 0xff, 0xff}
 238 };
 239
 240 #define CTL_LBPM_LEN    (sizeof(struct ctl_logical_block_provisioning_page) - 4)
 241
 242 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
 243         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 244         /*subpage_code*/0x02,
 245         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 246         /*flags*/0,
 247         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 248         /*descr*/{}},
 249         {{/*flags*/0,
 250           /*resource*/0x01,
 251           /*reserved*/{0, 0},
 252           /*count*/{0, 0, 0, 0}},
 253          {/*flags*/0,
 254           /*resource*/0x02,
 255           /*reserved*/{0, 0},
 256           /*count*/{0, 0, 0, 0}},
 257          {/*flags*/0,
 258           /*resource*/0xf1,
 259           /*reserved*/{0, 0},
 260           /*count*/{0, 0, 0, 0}},
 261          {/*flags*/0,
 262           /*resource*/0xf2,
 263           /*reserved*/{0, 0},
 264           /*count*/{0, 0, 0, 0}}
 265         }
 266 };
 267
 268 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
 269         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 270         /*subpage_code*/0x02,
 271         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 272         /*flags*/SLBPP_SITUA,
 273         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 274         /*descr*/{}},
 275         {{/*flags*/0,
 276           /*resource*/0,
 277           /*reserved*/{0, 0},
 278           /*count*/{0, 0, 0, 0}},
 279          {/*flags*/0,
 280           /*resource*/0,
 281           /*reserved*/{0, 0},
 282           /*count*/{0, 0, 0, 0}},
 283          {/*flags*/0,
 284           /*resource*/0,
 285           /*reserved*/{0, 0},
 286           /*count*/{0, 0, 0, 0}},
 287          {/*flags*/0,
 288           /*resource*/0,
 289           /*reserved*/{0, 0},
 290           /*count*/{0, 0, 0, 0}}
 291         }
 292 };
 293
 294 const static struct scsi_cddvd_capabilities_page cddvd_page_default = {
 295         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 296         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 297         /*caps1*/0x3f,
 298         /*caps2*/0x00,
 299         /*caps3*/0xf0,
 300         /*caps4*/0x00,
 301         /*caps5*/0x29,
 302         /*caps6*/0x00,
 303         /*obsolete*/{0, 0},
 304         /*nvol_levels*/{0, 0},
 305         /*buffer_size*/{8, 0},
 306         /*obsolete2*/{0, 0},
 307         /*reserved*/0,
 308         /*digital*/0,
 309         /*obsolete3*/0,
 310         /*copy_management*/0,
 311         /*reserved2*/0,
 312         /*rotation_control*/0,
 313         /*cur_write_speed*/0,
 314         /*num_speed_descr*/0,
 315 };
 316
 317 const static struct scsi_cddvd_capabilities_page cddvd_page_changeable = {
 318         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 319         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 320         /*caps1*/0,
 321         /*caps2*/0,
 322         /*caps3*/0,
 323         /*caps4*/0,
 324         /*caps5*/0,
 325         /*caps6*/0,
 326         /*obsolete*/{0, 0},
 327         /*nvol_levels*/{0, 0},
 328         /*buffer_size*/{0, 0},
 329         /*obsolete2*/{0, 0},
 330         /*reserved*/0,
 331         /*digital*/0,
 332         /*obsolete3*/0,
 333         /*copy_management*/0,
 334         /*reserved2*/0,
 335         /*rotation_control*/0,
 336         /*cur_write_speed*/0,
 337         /*num_speed_descr*/0,
 338 };
 339
 340 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
 341     "CAM Target Layer");
 342 static int worker_threads = -1;
 343 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 344     &worker_threads, 1, "Number of worker threads");
 345 static int ctl_debug = CTL_DEBUG_NONE;
 346 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
 347     &ctl_debug, 0, "Enabled debug flags");
 348 static int ctl_lun_map_size = 1024;
 349 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, lun_map_size, CTLFLAG_RWTUN,
 350     &ctl_lun_map_size, 0, "Size of per-port LUN map (max LUN + 1)");
 351 #ifdef  CTL_TIME_IO
 352 static int ctl_time_io_secs = CTL_TIME_IO_DEFAULT_SECS;
 353 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, time_io_secs, CTLFLAG_RWTUN,
 354     &ctl_time_io_secs, 0, "Log requests taking more seconds");
 355 #endif
 356
 357 /*
 358  * Maximum number of LUNs we support.  MUST be a power of 2.
 359  */
 360 #define CTL_DEFAULT_MAX_LUNS    1024
 361 static int ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
 362 TUNABLE_INT("kern.cam.ctl.max_luns", &ctl_max_luns);
 363 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_luns, CTLFLAG_RDTUN,
 364     &ctl_max_luns, CTL_DEFAULT_MAX_LUNS, "Maximum number of LUNs");
 365
 366 /*
 367  * Maximum number of ports registered at one time.
 368  */
 369 #define CTL_DEFAULT_MAX_PORTS           1024
 370 static int ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
 371 TUNABLE_INT("kern.cam.ctl.max_ports", &ctl_max_ports);
 372 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_ports, CTLFLAG_RDTUN,
 373     &ctl_max_ports, CTL_DEFAULT_MAX_LUNS, "Maximum number of ports");
 374
 375 /*
 376  * Maximum number of initiators we support.
 377  */
 378 #define CTL_MAX_INITIATORS      (CTL_MAX_INIT_PER_PORT * ctl_max_ports)
 379
 380 /*
 381  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 382  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 383  * SCSI Ports (0x88), Third-party Copy (0x8F), SCSI Feature Sets (0x92),
 384  * Block limits (0xB0), Block Device Characteristics (0xB1) and
 385  * Logical Block Provisioning (0xB2)
 386  */
 387 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   11
 388
 389 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 390                                   int param);
 391 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 392 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
 393 static int ctl_init(void);
 394 static int ctl_shutdown(void);
 395 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 396 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 397 static void ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 398 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 399                               struct ctl_ooa *ooa_hdr,
 400                               struct ctl_ooa_entry *kern_entries);
 401 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 402                      struct thread *td);
 403 static int ctl_enable_lun(struct ctl_lun *lun);
 404 static int ctl_disable_lun(struct ctl_lun *lun);
 405 static int ctl_free_lun(struct ctl_lun *lun);
 406
 407 static int ctl_do_mode_select(union ctl_io *io);
 408 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 409                            uint64_t res_key, uint64_t sa_res_key,
 410                            uint8_t type, uint32_t residx,
 411                            struct ctl_scsiio *ctsio,
 412                            struct scsi_per_res_out *cdb,
 413                            struct scsi_per_res_out_parms* param);
 414 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 415                                   union ctl_ha_msg *msg);
 416 static void ctl_hndl_per_res_out_on_other_sc(union ctl_io *io);
 417 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 418 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 419 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 420 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 421 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 422 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 423                                          int alloc_len);
 424 static int ctl_inquiry_evpd_sfs(struct ctl_scsiio *ctsio, int alloc_len);
 425 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 426                                          int alloc_len);
 427 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 428 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 429 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 430 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 431 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 432 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
 433     bool seq);
 434 static ctl_action ctl_seq_check(union ctl_io *io1, union ctl_io *io2);
 435 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 436     union ctl_io *pending_io, const uint8_t *serialize_row,
 437     union ctl_io *ooa_io);
 438 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 439                                 union ctl_io **starting_io);
 440 static void ctl_try_unblock_io(struct ctl_lun *lun, union ctl_io *io,
 441     bool skip);
 442 static void ctl_try_unblock_others(struct ctl_lun *lun, union ctl_io *io,
 443     bool skip);
 444 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
 445                                 const struct ctl_cmd_entry *entry,
 446                                 struct ctl_scsiio *ctsio);
 447 static void ctl_failover_lun(union ctl_io *io);
 448 static void ctl_scsiio_precheck(struct ctl_scsiio *ctsio);
 449 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 450
 451 static int ctl_target_reset(union ctl_io *io);
 452 static void ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx,
 453                          ctl_ua_type ua_type);
 454 static int ctl_lun_reset(union ctl_io *io);
 455 static int ctl_abort_task(union ctl_io *io);
 456 static int ctl_abort_task_set(union ctl_io *io);
 457 static int ctl_query_task(union ctl_io *io, int task_set);
 458 static void ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
 459                               ctl_ua_type ua_type);
 460 static int ctl_i_t_nexus_reset(union ctl_io *io);
 461 static int ctl_query_async_event(union ctl_io *io);
 462 static void ctl_run_task(union ctl_io *io);
 463 #ifdef CTL_IO_DELAY
 464 static void ctl_datamove_timer_wakeup(void *arg);
 465 static void ctl_done_timer_wakeup(void *arg);
 466 #endif /* CTL_IO_DELAY */
 467
 468 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 469 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 470 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io, bool samethr);
 471 static void ctl_datamove_remote_write(union ctl_io *io);
 472 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io, bool samethr);
 473 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 474 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 475 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 476                                     ctl_ha_dt_cb callback);
 477 static void ctl_datamove_remote_read(union ctl_io *io);
 478 static void ctl_datamove_remote(union ctl_io *io);
 479 static void ctl_process_done(union ctl_io *io);
 480 static void ctl_thresh_thread(void *arg);
 481 static void ctl_work_thread(void *arg);
 482 static void ctl_enqueue_incoming(union ctl_io *io);
 483 static void ctl_enqueue_rtr(union ctl_io *io);
 484 static void ctl_enqueue_done(union ctl_io *io);
 485 static void ctl_enqueue_isc(union ctl_io *io);
 486 static const struct ctl_cmd_entry *
 487     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 488 static const struct ctl_cmd_entry *
 489     ctl_validate_command(struct ctl_scsiio *ctsio);
 490 static int ctl_cmd_applicable(uint8_t lun_type,
 491     const struct ctl_cmd_entry *entry);
 492 static int ctl_ha_init(void);
 493 static int ctl_ha_shutdown(void);
 494
 495 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
 496 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
 497 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
 498 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
 499
 500 /*
 501  * Load the serialization table.  This isn't very pretty, but is probably
 502  * the easiest way to do it.
 503  */
 504 #include "ctl_ser_table.c"
 505
 506 /*
 507  * We only need to define open, close and ioctl routines for this driver.
 508  */
 509 static struct cdevsw ctl_cdevsw = {
 510         .d_version =    D_VERSION,
 511         .d_flags =      0,
 512         .d_open =       ctl_open,
 513         .d_close =      ctl_close,
 514         .d_ioctl =      ctl_ioctl,
 515         .d_name =       "ctl",
 516 };
 517
 518 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 519
 520 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 521
 522 static moduledata_t ctl_moduledata = {
 523         "ctl",
 524         ctl_module_event_handler,
 525         NULL
 526 };
 527
 528 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 529 MODULE_VERSION(ctl, 1);
 530
 531 static struct ctl_frontend ha_frontend =
 532 {
 533         .name = "ha",
 534         .init = ctl_ha_init,
 535         .shutdown = ctl_ha_shutdown,
 536 };
 537
 538 static int
 539 ctl_ha_init(void)
 540 {
 541         struct ctl_softc *softc = control_softc;
 542
 543         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
 544                             &softc->othersc_pool) != 0)
 545                 return (ENOMEM);
 546         if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
 547                 ctl_pool_free(softc->othersc_pool);
 548                 return (EIO);
 549         }
 550         if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
 551             != CTL_HA_STATUS_SUCCESS) {
 552                 ctl_ha_msg_destroy(softc);
 553                 ctl_pool_free(softc->othersc_pool);
 554                 return (EIO);
 555         }
 556         return (0);
 557 };
 558
 559 static int
 560 ctl_ha_shutdown(void)
 561 {
 562         struct ctl_softc *softc = control_softc;
 563         struct ctl_port *port;
 564
 565         ctl_ha_msg_shutdown(softc);
 566         if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL) != CTL_HA_STATUS_SUCCESS)
 567                 return (EIO);
 568         if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
 569                 return (EIO);
 570         ctl_pool_free(softc->othersc_pool);
 571         while ((port = STAILQ_FIRST(&ha_frontend.port_list)) != NULL) {
 572                 ctl_port_deregister(port);
 573                 free(port->port_name, M_CTL);
 574                 free(port, M_CTL);
 575         }
 576         return (0);
 577 };
 578
 579 static void
 580 ctl_ha_datamove(union ctl_io *io)
 581 {
 582         struct ctl_lun *lun = CTL_LUN(io);
 583         struct ctl_sg_entry *sgl;
 584         union ctl_ha_msg msg;
 585         uint32_t sg_entries_sent;
 586         int do_sg_copy, i, j;
 587
 588         memset(&msg.dt, 0, sizeof(msg.dt));
 589         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
 590         msg.hdr.original_sc = io->io_hdr.remote_io;
 591         msg.hdr.serializing_sc = io;
 592         msg.hdr.nexus = io->io_hdr.nexus;
 593         msg.hdr.status = io->io_hdr.status;
 594         msg.dt.flags = io->io_hdr.flags;
 595
 596         /*
 597          * We convert everything into a S/G list here.  We can't
 598          * pass by reference, only by value between controllers.
 599          * So we can't pass a pointer to the S/G list, only as many
 600          * S/G entries as we can fit in here.  If it's possible for
 601          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
 602          * then we need to break this up into multiple transfers.
 603          */
 604         if (io->scsiio.kern_sg_entries == 0) {
 605                 msg.dt.kern_sg_entries = 1;
 606 #if 0
 607                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 608                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 609                 } else {
 610                         /* XXX KDM use busdma here! */
 611                         msg.dt.sg_list[0].addr =
 612                             (void *)vtophys(io->scsiio.kern_data_ptr);
 613                 }
 614 #else
 615                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
 616                     ("HA does not support BUS_ADDR"));
 617                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 618 #endif
 619                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
 620                 do_sg_copy = 0;
 621         } else {
 622                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
 623                 do_sg_copy = 1;
 624         }
 625
 626         msg.dt.kern_data_len = io->scsiio.kern_data_len;
 627         msg.dt.kern_total_len = io->scsiio.kern_total_len;
 628         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
 629         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
 630         msg.dt.sg_sequence = 0;
 631
 632         /*
 633          * Loop until we've sent all of the S/G entries.  On the
 634          * other end, we'll recompose these S/G entries into one
 635          * contiguous list before processing.
 636          */
 637         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
 638             msg.dt.sg_sequence++) {
 639                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
 640                     sizeof(msg.dt.sg_list[0])),
 641                     msg.dt.kern_sg_entries - sg_entries_sent);
 642                 if (do_sg_copy != 0) {
 643                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
 644                         for (i = sg_entries_sent, j = 0;
 645                              i < msg.dt.cur_sg_entries; i++, j++) {
 646 #if 0
 647                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 648                                         msg.dt.sg_list[j].addr = sgl[i].addr;
 649                                 } else {
 650                                         /* XXX KDM use busdma here! */
 651                                         msg.dt.sg_list[j].addr =
 652                                             (void *)vtophys(sgl[i].addr);
 653                                 }
 654 #else
 655                                 KASSERT((io->io_hdr.flags &
 656                                     CTL_FLAG_BUS_ADDR) == 0,
 657                                     ("HA does not support BUS_ADDR"));
 658                                 msg.dt.sg_list[j].addr = sgl[i].addr;
 659 #endif
 660                                 msg.dt.sg_list[j].len = sgl[i].len;
 661                         }
 662                 }
 663
 664                 sg_entries_sent += msg.dt.cur_sg_entries;
 665                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
 666                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 667                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
 668                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
 669                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
 670                         io->io_hdr.port_status = 31341;
 671                         ctl_datamove_done(io, true);
 672                         return;
 673                 }
 674                 msg.dt.sent_sg_entries = sg_entries_sent;
 675         }
 676
 677         /*
 678          * Officially handover the request from us to peer.
 679          * If failover has just happened, then we must return error.
 680          * If failover happen just after, then it is not our problem.
 681          */
 682         if (lun)
 683                 mtx_lock(&lun->lun_lock);
 684         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
 685                 if (lun)
 686                         mtx_unlock(&lun->lun_lock);
 687                 io->io_hdr.port_status = 31342;
 688                 ctl_datamove_done(io, true);
 689                 return;
 690         }
 691         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
 692         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
 693         if (lun)
 694                 mtx_unlock(&lun->lun_lock);
 695 }
 696
 697 static void
 698 ctl_ha_done(union ctl_io *io)
 699 {
 700         union ctl_ha_msg msg;
 701
 702         if (io->io_hdr.io_type == CTL_IO_SCSI) {
 703                 memset(&msg, 0, sizeof(msg));
 704                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
 705                 msg.hdr.original_sc = io->io_hdr.remote_io;
 706                 msg.hdr.nexus = io->io_hdr.nexus;
 707                 msg.hdr.status = io->io_hdr.status;
 708                 msg.scsi.scsi_status = io->scsiio.scsi_status;
 709                 msg.scsi.tag_num = io->scsiio.tag_num;
 710                 msg.scsi.tag_type = io->scsiio.tag_type;
 711                 msg.scsi.sense_len = io->scsiio.sense_len;
 712                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
 713                     io->scsiio.sense_len);
 714                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 715                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
 716                     msg.scsi.sense_len, M_WAITOK);
 717         }
 718         ctl_free_io(io);
 719 }
 720
 721 static void
 722 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 723                             union ctl_ha_msg *msg_info)
 724 {
 725         struct ctl_scsiio *ctsio;
 726
 727         if (msg_info->hdr.original_sc == NULL) {
 728                 printf("%s: original_sc == NULL!\n", __func__);
 729                 /* XXX KDM now what? */
 730                 return;
 731         }
 732
 733         ctsio = &msg_info->hdr.original_sc->scsiio;
 734         ctsio->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
 735         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 736         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 737         ctsio->io_hdr.status = msg_info->hdr.status;
 738         ctsio->scsi_status = msg_info->scsi.scsi_status;
 739         ctsio->sense_len = msg_info->scsi.sense_len;
 740         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 741                msg_info->scsi.sense_len);
 742         ctl_enqueue_isc((union ctl_io *)ctsio);
 743 }
 744
 745 static void
 746 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 747                                 union ctl_ha_msg *msg_info)
 748 {
 749         struct ctl_scsiio *ctsio;
 750
 751         if (msg_info->hdr.serializing_sc == NULL) {
 752                 printf("%s: serializing_sc == NULL!\n", __func__);
 753                 /* XXX KDM now what? */
 754                 return;
 755         }
 756
 757         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 758         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 759         ctl_enqueue_isc((union ctl_io *)ctsio);
 760 }
 761
 762 void
 763 ctl_isc_announce_lun(struct ctl_lun *lun)
 764 {
 765         struct ctl_softc *softc = lun->ctl_softc;
 766         union ctl_ha_msg *msg;
 767         struct ctl_ha_msg_lun_pr_key pr_key;
 768         int i, k;
 769
 770         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 771                 return;
 772         mtx_lock(&lun->lun_lock);
 773         i = sizeof(msg->lun);
 774         if (lun->lun_devid)
 775                 i += lun->lun_devid->len;
 776         i += sizeof(pr_key) * lun->pr_key_count;
 777 alloc:
 778         mtx_unlock(&lun->lun_lock);
 779         msg = malloc(i, M_CTL, M_WAITOK);
 780         mtx_lock(&lun->lun_lock);
 781         k = sizeof(msg->lun);
 782         if (lun->lun_devid)
 783                 k += lun->lun_devid->len;
 784         k += sizeof(pr_key) * lun->pr_key_count;
 785         if (i < k) {
 786                 free(msg, M_CTL);
 787                 i = k;
 788                 goto alloc;
 789         }
 790         bzero(&msg->lun, sizeof(msg->lun));
 791         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 792         msg->hdr.nexus.targ_lun = lun->lun;
 793         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 794         msg->lun.flags = lun->flags;
 795         msg->lun.pr_generation = lun->pr_generation;
 796         msg->lun.pr_res_idx = lun->pr_res_idx;
 797         msg->lun.pr_res_type = lun->pr_res_type;
 798         msg->lun.pr_key_count = lun->pr_key_count;
 799         i = 0;
 800         if (lun->lun_devid) {
 801                 msg->lun.lun_devid_len = lun->lun_devid->len;
 802                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 803                     msg->lun.lun_devid_len);
 804                 i += msg->lun.lun_devid_len;
 805         }
 806         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 807                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 808                         continue;
 809                 pr_key.pr_iid = k;
 810                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 811                 i += sizeof(pr_key);
 812         }
 813         mtx_unlock(&lun->lun_lock);
 814         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->lun, sizeof(msg->lun) + i,
 815             M_WAITOK);
 816         free(msg, M_CTL);
 817
 818         if (lun->flags & CTL_LUN_PRIMARY_SC) {
 819                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 820                         ctl_isc_announce_mode(lun, -1,
 821                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
 822                             lun->mode_pages.index[i].subpage);
 823                 }
 824         }
 825 }
 826
 827 void
 828 ctl_isc_announce_port(struct ctl_port *port)
 829 {
 830         struct ctl_softc *softc = port->ctl_softc;
 831         union ctl_ha_msg *msg;
 832         int i;
 833
 834         if (port->targ_port < softc->port_min ||
 835             port->targ_port >= softc->port_max ||
 836             softc->ha_link != CTL_HA_LINK_ONLINE)
 837                 return;
 838         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 839         if (port->lun_map)
 840                 i += port->lun_map_size * sizeof(uint32_t);
 841         if (port->port_devid)
 842                 i += port->port_devid->len;
 843         if (port->target_devid)
 844                 i += port->target_devid->len;
 845         if (port->init_devid)
 846                 i += port->init_devid->len;
 847         msg = malloc(i, M_CTL, M_WAITOK);
 848         bzero(&msg->port, sizeof(msg->port));
 849         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 850         msg->hdr.nexus.targ_port = port->targ_port;
 851         msg->port.port_type = port->port_type;
 852         msg->port.physical_port = port->physical_port;
 853         msg->port.virtual_port = port->virtual_port;
 854         msg->port.status = port->status;
 855         i = 0;
 856         msg->port.name_len = sprintf(&msg->port.data[i],
 857             "%d:%s", softc->ha_id, port->port_name) + 1;
 858         i += msg->port.name_len;
 859         if (port->lun_map) {
 860                 msg->port.lun_map_len = port->lun_map_size * sizeof(uint32_t);
 861                 memcpy(&msg->port.data[i], port->lun_map,
 862                     msg->port.lun_map_len);
 863                 i += msg->port.lun_map_len;
 864         }
 865         if (port->port_devid) {
 866                 msg->port.port_devid_len = port->port_devid->len;
 867                 memcpy(&msg->port.data[i], port->port_devid->data,
 868                     msg->port.port_devid_len);
 869                 i += msg->port.port_devid_len;
 870         }
 871         if (port->target_devid) {
 872                 msg->port.target_devid_len = port->target_devid->len;
 873                 memcpy(&msg->port.data[i], port->target_devid->data,
 874                     msg->port.target_devid_len);
 875                 i += msg->port.target_devid_len;
 876         }
 877         if (port->init_devid) {
 878                 msg->port.init_devid_len = port->init_devid->len;
 879                 memcpy(&msg->port.data[i], port->init_devid->data,
 880                     msg->port.init_devid_len);
 881                 i += msg->port.init_devid_len;
 882         }
 883         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 884             M_WAITOK);
 885         free(msg, M_CTL);
 886 }
 887
 888 void
 889 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 890 {
 891         struct ctl_softc *softc = port->ctl_softc;
 892         union ctl_ha_msg *msg;
 893         int i, l;
 894
 895         if (port->targ_port < softc->port_min ||
 896             port->targ_port >= softc->port_max ||
 897             softc->ha_link != CTL_HA_LINK_ONLINE)
 898                 return;
 899         mtx_lock(&softc->ctl_lock);
 900         i = sizeof(msg->iid);
 901         l = 0;
 902         if (port->wwpn_iid[iid].name)
 903                 l = strlen(port->wwpn_iid[iid].name) + 1;
 904         i += l;
 905         msg = malloc(i, M_CTL, M_NOWAIT);
 906         if (msg == NULL) {
 907                 mtx_unlock(&softc->ctl_lock);
 908                 return;
 909         }
 910         bzero(&msg->iid, sizeof(msg->iid));
 911         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 912         msg->hdr.nexus.targ_port = port->targ_port;
 913         msg->hdr.nexus.initid = iid;
 914         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 915         msg->iid.name_len = l;
 916         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 917         if (port->wwpn_iid[iid].name)
 918                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 919         mtx_unlock(&softc->ctl_lock);
 920         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 921         free(msg, M_CTL);
 922 }
 923
 924 void
 925 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
 926     uint8_t page, uint8_t subpage)
 927 {
 928         struct ctl_softc *softc = lun->ctl_softc;
 929         union ctl_ha_msg *msg;
 930         u_int i, l;
 931
 932         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 933                 return;
 934         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 935                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
 936                     page && lun->mode_pages.index[i].subpage == subpage)
 937                         break;
 938         }
 939         if (i == CTL_NUM_MODE_PAGES)
 940                 return;
 941
 942         /* Don't try to replicate pages not present on this device. */
 943         if (lun->mode_pages.index[i].page_data == NULL)
 944                 return;
 945
 946         l = sizeof(msg->mode) + lun->mode_pages.index[i].page_len;
 947         msg = malloc(l, M_CTL, M_WAITOK | M_ZERO);
 948         msg->hdr.msg_type = CTL_MSG_MODE_SYNC;
 949         msg->hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
 950         msg->hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
 951         msg->hdr.nexus.targ_lun = lun->lun;
 952         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 953         msg->mode.page_code = page;
 954         msg->mode.subpage = subpage;
 955         msg->mode.page_len = lun->mode_pages.index[i].page_len;
 956         memcpy(msg->mode.data, lun->mode_pages.index[i].page_data,
 957             msg->mode.page_len);
 958         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->mode, l, M_WAITOK);
 959         free(msg, M_CTL);
 960 }
 961
 962 static void
 963 ctl_isc_ha_link_up(struct ctl_softc *softc)
 964 {
 965         struct ctl_port *port;
 966         struct ctl_lun *lun;
 967         union ctl_ha_msg msg;
 968         int i;
 969
 970         /* Announce this node parameters to peer for validation. */
 971         msg.login.msg_type = CTL_MSG_LOGIN;
 972         msg.login.version = CTL_HA_VERSION;
 973         msg.login.ha_mode = softc->ha_mode;
 974         msg.login.ha_id = softc->ha_id;
 975         msg.login.max_luns = ctl_max_luns;
 976         msg.login.max_ports = ctl_max_ports;
 977         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
 978         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
 979             M_WAITOK);
 980
 981         STAILQ_FOREACH(port, &softc->port_list, links) {
 982                 ctl_isc_announce_port(port);
 983                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
 984                         if (port->wwpn_iid[i].in_use)
 985                                 ctl_isc_announce_iid(port, i);
 986                 }
 987         }
 988         STAILQ_FOREACH(lun, &softc->lun_list, links)
 989                 ctl_isc_announce_lun(lun);
 990 }
 991
 992 static void
 993 ctl_isc_ha_link_down(struct ctl_softc *softc)
 994 {
 995         struct ctl_port *port;
 996         struct ctl_lun *lun;
 997         union ctl_io *io;
 998         int i;
 999
1000         mtx_lock(&softc->ctl_lock);
1001         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1002                 mtx_lock(&lun->lun_lock);
1003                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
1004                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1005                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1006                 }
1007                 mtx_unlock(&lun->lun_lock);
1008
1009                 mtx_unlock(&softc->ctl_lock);
1010                 io = ctl_alloc_io(softc->othersc_pool);
1011                 mtx_lock(&softc->ctl_lock);
1012                 ctl_zero_io(io);
1013                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1014                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1015                 ctl_enqueue_isc(io);
1016         }
1017
1018         STAILQ_FOREACH(port, &softc->port_list, links) {
1019                 if (port->targ_port >= softc->port_min &&
1020                     port->targ_port < softc->port_max)
1021                         continue;
1022                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1023                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1024                         port->wwpn_iid[i].in_use = 0;
1025                         free(port->wwpn_iid[i].name, M_CTL);
1026                         port->wwpn_iid[i].name = NULL;
1027                 }
1028         }
1029         mtx_unlock(&softc->ctl_lock);
1030 }
1031
1032 static void
1033 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1034 {
1035         struct ctl_lun *lun;
1036         uint32_t iid;
1037
1038         if (len < sizeof(msg->ua)) {
1039                 printf("%s: Received truncated message %d < %zu\n",
1040                     __func__, len, sizeof(msg->ua));
1041                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1042                 return;
1043         }
1044
1045         mtx_lock(&softc->ctl_lock);
1046         if (msg->hdr.nexus.targ_mapped_lun >= ctl_max_luns ||
1047             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) == NULL) {
1048                 mtx_unlock(&softc->ctl_lock);
1049                 return;
1050         }
1051         mtx_lock(&lun->lun_lock);
1052         mtx_unlock(&softc->ctl_lock);
1053         if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES && msg->ua.ua_set)
1054                 memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1055         iid = ctl_get_initindex(&msg->hdr.nexus);
1056         if (msg->ua.ua_all) {
1057                 if (msg->ua.ua_set)
1058                         ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1059                 else
1060                         ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1061         } else {
1062                 if (msg->ua.ua_set)
1063                         ctl_est_ua(lun, iid, msg->ua.ua_type);
1064                 else
1065                         ctl_clr_ua(lun, iid, msg->ua.ua_type);
1066         }
1067         mtx_unlock(&lun->lun_lock);
1068 }
1069
1070 static void
1071 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1072 {
1073         struct ctl_lun *lun;
1074         struct ctl_ha_msg_lun_pr_key pr_key;
1075         int i, k;
1076         ctl_lun_flags oflags;
1077         uint32_t targ_lun;
1078
1079         if (len < offsetof(struct ctl_ha_msg_lun, data[0])) {
1080                 printf("%s: Received truncated message %d < %zu\n",
1081                     __func__, len, offsetof(struct ctl_ha_msg_lun, data[0]));
1082                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1083                 return;
1084         }
1085         i = msg->lun.lun_devid_len + msg->lun.pr_key_count * sizeof(pr_key);
1086         if (len < offsetof(struct ctl_ha_msg_lun, data[i])) {
1087                 printf("%s: Received truncated message data %d < %zu\n",
1088                     __func__, len, offsetof(struct ctl_ha_msg_lun, data[i]));
1089                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1090                 return;
1091         }
1092
1093         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1094         mtx_lock(&softc->ctl_lock);
1095         if (targ_lun >= ctl_max_luns ||
1096             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1097                 mtx_unlock(&softc->ctl_lock);
1098                 return;
1099         }
1100         mtx_lock(&lun->lun_lock);
1101         mtx_unlock(&softc->ctl_lock);
1102         if (lun->flags & CTL_LUN_DISABLED) {
1103                 mtx_unlock(&lun->lun_lock);
1104                 return;
1105         }
1106         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1107         if (msg->lun.lun_devid_len != i || (i > 0 &&
1108             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1109                 mtx_unlock(&lun->lun_lock);
1110                 printf("%s: Received conflicting HA LUN %d\n",
1111                     __func__, targ_lun);
1112                 return;
1113         } else {
1114                 /* Record whether peer is primary. */
1115                 oflags = lun->flags;
1116                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1117                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1118                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1119                 else
1120                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1121                 if (oflags != lun->flags)
1122                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1123
1124                 /* If peer is primary and we are not -- use data */
1125                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1126                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1127                         lun->pr_generation = msg->lun.pr_generation;
1128                         lun->pr_res_idx = msg->lun.pr_res_idx;
1129                         lun->pr_res_type = msg->lun.pr_res_type;
1130                         lun->pr_key_count = msg->lun.pr_key_count;
1131                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1132                                 ctl_clr_prkey(lun, k);
1133                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1134                                 memcpy(&pr_key, &msg->lun.data[i],
1135                                     sizeof(pr_key));
1136                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1137                                 ctl_set_prkey(lun, pr_key.pr_iid,
1138                                     pr_key.pr_key);
1139                                 i += sizeof(pr_key);
1140                         }
1141                 }
1142
1143                 mtx_unlock(&lun->lun_lock);
1144                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1145                     __func__, targ_lun,
1146                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1147                     "primary" : "secondary"));
1148
1149                 /* If we are primary but peer doesn't know -- notify */
1150                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1151                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1152                         ctl_isc_announce_lun(lun);
1153         }
1154 }
1155
1156 static void
1157 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1158 {
1159         struct ctl_port *port;
1160         struct ctl_lun *lun;
1161         int i, new;
1162
1163         if (len < offsetof(struct ctl_ha_msg_port, data[0])) {
1164                 printf("%s: Received truncated message %d < %zu\n",
1165                     __func__, len, offsetof(struct ctl_ha_msg_port, data[0]));
1166                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1167                 return;
1168         }
1169         i = msg->port.name_len + msg->port.lun_map_len +
1170             msg->port.port_devid_len + msg->port.target_devid_len +
1171             msg->port.init_devid_len;
1172         if (len < offsetof(struct ctl_ha_msg_port, data[i])) {
1173                 printf("%s: Received truncated message data %d < %zu\n",
1174                     __func__, len, offsetof(struct ctl_ha_msg_port, data[i]));
1175                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1176                 return;
1177         }
1178
1179         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1180         if (port == NULL) {
1181                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1182                     msg->hdr.nexus.targ_port));
1183                 new = 1;
1184                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1185                 port->frontend = &ha_frontend;
1186                 port->targ_port = msg->hdr.nexus.targ_port;
1187                 port->fe_datamove = ctl_ha_datamove;
1188                 port->fe_done = ctl_ha_done;
1189         } else if (port->frontend == &ha_frontend) {
1190                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1191                     msg->hdr.nexus.targ_port));
1192                 new = 0;
1193         } else {
1194                 printf("%s: Received conflicting HA port %d\n",
1195                     __func__, msg->hdr.nexus.targ_port);
1196                 return;
1197         }
1198         port->port_type = msg->port.port_type;
1199         port->physical_port = msg->port.physical_port;
1200         port->virtual_port = msg->port.virtual_port;
1201         port->status = msg->port.status;
1202         i = 0;
1203         free(port->port_name, M_CTL);
1204         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1205             M_CTL);
1206         i += msg->port.name_len;
1207         if (msg->port.lun_map_len != 0) {
1208                 if (port->lun_map == NULL ||
1209                     port->lun_map_size * sizeof(uint32_t) <
1210                     msg->port.lun_map_len) {
1211                         port->lun_map_size = 0;
1212                         free(port->lun_map, M_CTL);
1213                         port->lun_map = malloc(msg->port.lun_map_len,
1214                             M_CTL, M_WAITOK);
1215                 }
1216                 memcpy(port->lun_map, &msg->port.data[i], msg->port.lun_map_len);
1217                 port->lun_map_size = msg->port.lun_map_len / sizeof(uint32_t);
1218                 i += msg->port.lun_map_len;
1219         } else {
1220                 port->lun_map_size = 0;
1221                 free(port->lun_map, M_CTL);
1222                 port->lun_map = NULL;
1223         }
1224         if (msg->port.port_devid_len != 0) {
1225                 if (port->port_devid == NULL ||
1226                     port->port_devid->len < msg->port.port_devid_len) {
1227                         free(port->port_devid, M_CTL);
1228                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1229                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1230                 }
1231                 memcpy(port->port_devid->data, &msg->port.data[i],
1232                     msg->port.port_devid_len);
1233                 port->port_devid->len = msg->port.port_devid_len;
1234                 i += msg->port.port_devid_len;
1235         } else {
1236                 free(port->port_devid, M_CTL);
1237                 port->port_devid = NULL;
1238         }
1239         if (msg->port.target_devid_len != 0) {
1240                 if (port->target_devid == NULL ||
1241                     port->target_devid->len < msg->port.target_devid_len) {
1242                         free(port->target_devid, M_CTL);
1243                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1244                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1245                 }
1246                 memcpy(port->target_devid->data, &msg->port.data[i],
1247                     msg->port.target_devid_len);
1248                 port->target_devid->len = msg->port.target_devid_len;
1249                 i += msg->port.target_devid_len;
1250         } else {
1251                 free(port->target_devid, M_CTL);
1252                 port->target_devid = NULL;
1253         }
1254         if (msg->port.init_devid_len != 0) {
1255                 if (port->init_devid == NULL ||
1256                     port->init_devid->len < msg->port.init_devid_len) {
1257                         free(port->init_devid, M_CTL);
1258                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1259                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1260                 }
1261                 memcpy(port->init_devid->data, &msg->port.data[i],
1262                     msg->port.init_devid_len);
1263                 port->init_devid->len = msg->port.init_devid_len;
1264                 i += msg->port.init_devid_len;
1265         } else {
1266                 free(port->init_devid, M_CTL);
1267                 port->init_devid = NULL;
1268         }
1269         if (new) {
1270                 if (ctl_port_register(port) != 0) {
1271                         printf("%s: ctl_port_register() failed with error\n",
1272                             __func__);
1273                 }
1274         }
1275         mtx_lock(&softc->ctl_lock);
1276         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1277                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
1278                         continue;
1279                 mtx_lock(&lun->lun_lock);
1280                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1281                 mtx_unlock(&lun->lun_lock);
1282         }
1283         mtx_unlock(&softc->ctl_lock);
1284 }
1285
1286 static void
1287 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1288 {
1289         struct ctl_port *port;
1290         int i, iid;
1291
1292         if (len < offsetof(struct ctl_ha_msg_iid, data[0])) {
1293                 printf("%s: Received truncated message %d < %zu\n",
1294                     __func__, len, offsetof(struct ctl_ha_msg_iid, data[0]));
1295                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1296                 return;
1297         }
1298         i = msg->iid.name_len;
1299         if (len < offsetof(struct ctl_ha_msg_iid, data[i])) {
1300                 printf("%s: Received truncated message data %d < %zu\n",
1301                     __func__, len, offsetof(struct ctl_ha_msg_iid, data[i]));
1302                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1303                 return;
1304         }
1305
1306         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1307         if (port == NULL) {
1308                 printf("%s: Received IID for unknown port %d\n",
1309                     __func__, msg->hdr.nexus.targ_port);
1310                 return;
1311         }
1312         iid = msg->hdr.nexus.initid;
1313         if (port->wwpn_iid[iid].in_use != 0 &&
1314             msg->iid.in_use == 0)
1315                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
1316         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1317         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1318         free(port->wwpn_iid[iid].name, M_CTL);
1319         if (msg->iid.name_len) {
1320                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1321                     msg->iid.name_len, M_CTL);
1322         } else
1323                 port->wwpn_iid[iid].name = NULL;
1324 }
1325
1326 static void
1327 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1328 {
1329
1330         if (len < sizeof(msg->login)) {
1331                 printf("%s: Received truncated message %d < %zu\n",
1332                     __func__, len, sizeof(msg->login));
1333                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1334                 return;
1335         }
1336
1337         if (msg->login.version != CTL_HA_VERSION) {
1338                 printf("CTL HA peers have different versions %d != %d\n",
1339                     msg->login.version, CTL_HA_VERSION);
1340                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1341                 return;
1342         }
1343         if (msg->login.ha_mode != softc->ha_mode) {
1344                 printf("CTL HA peers have different ha_mode %d != %d\n",
1345                     msg->login.ha_mode, softc->ha_mode);
1346                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1347                 return;
1348         }
1349         if (msg->login.ha_id == softc->ha_id) {
1350                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1351                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1352                 return;
1353         }
1354         if (msg->login.max_luns != ctl_max_luns ||
1355             msg->login.max_ports != ctl_max_ports ||
1356             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1357                 printf("CTL HA peers have different limits\n");
1358                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1359                 return;
1360         }
1361 }
1362
1363 static void
1364 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1365 {
1366         struct ctl_lun *lun;
1367         u_int i;
1368         uint32_t initidx, targ_lun;
1369
1370         if (len < offsetof(struct ctl_ha_msg_mode, data[0])) {
1371                 printf("%s: Received truncated message %d < %zu\n",
1372                     __func__, len, offsetof(struct ctl_ha_msg_mode, data[0]));
1373                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1374                 return;
1375         }
1376         i = msg->mode.page_len;
1377         if (len < offsetof(struct ctl_ha_msg_mode, data[i])) {
1378                 printf("%s: Received truncated message data %d < %zu\n",
1379                     __func__, len, offsetof(struct ctl_ha_msg_mode, data[i]));
1380                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1381                 return;
1382         }
1383
1384         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1385         mtx_lock(&softc->ctl_lock);
1386         if (targ_lun >= ctl_max_luns ||
1387             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1388                 mtx_unlock(&softc->ctl_lock);
1389                 return;
1390         }
1391         mtx_lock(&lun->lun_lock);
1392         mtx_unlock(&softc->ctl_lock);
1393         if (lun->flags & CTL_LUN_DISABLED) {
1394                 mtx_unlock(&lun->lun_lock);
1395                 return;
1396         }
1397         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1398                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1399                     msg->mode.page_code &&
1400                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1401                         break;
1402         }
1403         if (i == CTL_NUM_MODE_PAGES) {
1404                 mtx_unlock(&lun->lun_lock);
1405                 return;
1406         }
1407         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1408             min(lun->mode_pages.index[i].page_len, msg->mode.page_len));
1409         initidx = ctl_get_initindex(&msg->hdr.nexus);
1410         if (initidx != -1)
1411                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1412         mtx_unlock(&lun->lun_lock);
1413 }
1414
1415 /*
1416  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1417  * subsystem come in here.
1418  */
1419 static void
1420 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1421 {
1422         struct ctl_softc *softc = control_softc;
1423         union ctl_io *io;
1424         struct ctl_prio *presio;
1425         ctl_ha_status isc_status;
1426
1427         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1428         if (event == CTL_HA_EVT_MSG_RECV) {
1429                 union ctl_ha_msg *msg, msgbuf;
1430
1431                 if (param > sizeof(msgbuf))
1432                         msg = malloc(param, M_CTL, M_WAITOK);
1433                 else
1434                         msg = &msgbuf;
1435                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1436                     M_WAITOK);
1437                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1438                         printf("%s: Error receiving message: %d\n",
1439                             __func__, isc_status);
1440                         if (msg != &msgbuf)
1441                                 free(msg, M_CTL);
1442                         return;
1443                 }
1444
1445                 CTL_DEBUG_PRINT(("CTL: msg_type %d len %d\n",
1446                     msg->hdr.msg_type, param));
1447                 switch (msg->hdr.msg_type) {
1448                 case CTL_MSG_SERIALIZE:
1449                         io = ctl_alloc_io(softc->othersc_pool);
1450                         ctl_zero_io(io);
1451                         // populate ctsio from msg
1452                         io->io_hdr.io_type = CTL_IO_SCSI;
1453                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1454                         io->io_hdr.remote_io = msg->hdr.original_sc;
1455                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1456                                             CTL_FLAG_IO_ACTIVE;
1457                         /*
1458                          * If we're in serialization-only mode, we don't
1459                          * want to go through full done processing.  Thus
1460                          * the COPY flag.
1461                          *
1462                          * XXX KDM add another flag that is more specific.
1463                          */
1464                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1465                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1466                         io->io_hdr.nexus = msg->hdr.nexus;
1467                         io->scsiio.priority = msg->scsi.priority;
1468                         io->scsiio.tag_num = msg->scsi.tag_num;
1469                         io->scsiio.tag_type = msg->scsi.tag_type;
1470 #ifdef CTL_TIME_IO
1471                         io->io_hdr.start_time = time_uptime;
1472                         getbinuptime(&io->io_hdr.start_bt);
1473 #endif /* CTL_TIME_IO */
1474                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1475                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1476                                CTL_MAX_CDBLEN);
1477                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1478                                 const struct ctl_cmd_entry *entry;
1479
1480                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1481                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1482                                 io->io_hdr.flags |=
1483                                         entry->flags & CTL_FLAG_DATA_MASK;
1484                         }
1485                         ctl_enqueue_isc(io);
1486                         break;
1487
1488                 /* Performed on the Originating SC, XFER mode only */
1489                 case CTL_MSG_DATAMOVE: {
1490                         struct ctl_sg_entry *sgl;
1491                         int i, j;
1492
1493                         io = msg->hdr.original_sc;
1494                         if (io == NULL) {
1495                                 printf("%s: original_sc == NULL!\n", __func__);
1496                                 /* XXX KDM do something here */
1497                                 break;
1498                         }
1499                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1500                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1501                         /*
1502                          * Keep track of this, we need to send it back over
1503                          * when the datamove is complete.
1504                          */
1505                         io->io_hdr.remote_io = msg->hdr.serializing_sc;
1506                         if (msg->hdr.status == CTL_SUCCESS)
1507                                 io->io_hdr.status = msg->hdr.status;
1508
1509                         if (msg->dt.sg_sequence == 0) {
1510 #ifdef CTL_TIME_IO
1511                                 getbinuptime(&io->io_hdr.dma_start_bt);
1512 #endif
1513                                 i = msg->dt.kern_sg_entries +
1514                                     msg->dt.kern_data_len /
1515                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1516                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1517                                     M_WAITOK | M_ZERO);
1518                                 CTL_RSGL(io) = sgl;
1519                                 CTL_LSGL(io) = &sgl[msg->dt.kern_sg_entries];
1520
1521                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1522
1523                                 io->scsiio.kern_sg_entries =
1524                                         msg->dt.kern_sg_entries;
1525                                 io->scsiio.rem_sg_entries =
1526                                         msg->dt.kern_sg_entries;
1527                                 io->scsiio.kern_data_len =
1528                                         msg->dt.kern_data_len;
1529                                 io->scsiio.kern_total_len =
1530                                         msg->dt.kern_total_len;
1531                                 io->scsiio.kern_data_resid =
1532                                         msg->dt.kern_data_resid;
1533                                 io->scsiio.kern_rel_offset =
1534                                         msg->dt.kern_rel_offset;
1535                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1536                                 io->io_hdr.flags |= msg->dt.flags &
1537                                     CTL_FLAG_BUS_ADDR;
1538                         } else
1539                                 sgl = (struct ctl_sg_entry *)
1540                                         io->scsiio.kern_data_ptr;
1541
1542                         for (i = msg->dt.sent_sg_entries, j = 0;
1543                              i < (msg->dt.sent_sg_entries +
1544                              msg->dt.cur_sg_entries); i++, j++) {
1545                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1546                                 sgl[i].len = msg->dt.sg_list[j].len;
1547                         }
1548
1549                         /*
1550                          * If this is the last piece of the I/O, we've got
1551                          * the full S/G list.  Queue processing in the thread.
1552                          * Otherwise wait for the next piece.
1553                          */
1554                         if (msg->dt.sg_last != 0)
1555                                 ctl_enqueue_isc(io);
1556                         break;
1557                 }
1558                 /* Performed on the Serializing (primary) SC, XFER mode only */
1559                 case CTL_MSG_DATAMOVE_DONE: {
1560                         if (msg->hdr.serializing_sc == NULL) {
1561                                 printf("%s: serializing_sc == NULL!\n",
1562                                        __func__);
1563                                 /* XXX KDM now what? */
1564                                 break;
1565                         }
1566                         /*
1567                          * We grab the sense information here in case
1568                          * there was a failure, so we can return status
1569                          * back to the initiator.
1570                          */
1571                         io = msg->hdr.serializing_sc;
1572                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1573                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1574                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1575                         io->io_hdr.port_status = msg->scsi.port_status;
1576                         io->scsiio.kern_data_resid = msg->scsi.kern_data_resid;
1577                         if (msg->hdr.status != CTL_STATUS_NONE) {
1578                                 io->io_hdr.status = msg->hdr.status;
1579                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1580                                 io->scsiio.sense_len = msg->scsi.sense_len;
1581                                 memcpy(&io->scsiio.sense_data,
1582                                     &msg->scsi.sense_data,
1583                                     msg->scsi.sense_len);
1584                                 if (msg->hdr.status == CTL_SUCCESS)
1585                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1586                         }
1587                         ctl_enqueue_isc(io);
1588                         break;
1589                 }
1590
1591                 /* Preformed on Originating SC, SER_ONLY mode */
1592                 case CTL_MSG_R2R:
1593                         io = msg->hdr.original_sc;
1594                         if (io == NULL) {
1595                                 printf("%s: original_sc == NULL!\n",
1596                                     __func__);
1597                                 break;
1598                         }
1599                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1600                         io->io_hdr.msg_type = CTL_MSG_R2R;
1601                         io->io_hdr.remote_io = msg->hdr.serializing_sc;
1602                         ctl_enqueue_isc(io);
1603                         break;
1604
1605                 /*
1606                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1607                  * mode.
1608                  * Performed on the Originating (i.e. secondary) SC in XFER
1609                  * mode
1610                  */
1611                 case CTL_MSG_FINISH_IO:
1612                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1613                                 ctl_isc_handler_finish_xfer(softc, msg);
1614                         else
1615                                 ctl_isc_handler_finish_ser_only(softc, msg);
1616                         break;
1617
1618                 /* Preformed on Originating SC */
1619                 case CTL_MSG_BAD_JUJU:
1620                         io = msg->hdr.original_sc;
1621                         if (io == NULL) {
1622                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1623                                        __func__);
1624                                 break;
1625                         }
1626                         ctl_copy_sense_data(msg, io);
1627                         /*
1628                          * IO should have already been cleaned up on other
1629                          * SC so clear this flag so we won't send a message
1630                          * back to finish the IO there.
1631                          */
1632                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1633                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1634
1635                         /* io = msg->hdr.serializing_sc; */
1636                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1637                         ctl_enqueue_isc(io);
1638                         break;
1639
1640                 /* Handle resets sent from the other side */
1641                 case CTL_MSG_MANAGE_TASKS: {
1642                         struct ctl_taskio *taskio;
1643                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1644                             softc->othersc_pool);
1645                         ctl_zero_io((union ctl_io *)taskio);
1646                         taskio->io_hdr.io_type = CTL_IO_TASK;
1647                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1648                         taskio->io_hdr.nexus = msg->hdr.nexus;
1649                         taskio->task_action = msg->task.task_action;
1650                         taskio->tag_num = msg->task.tag_num;
1651                         taskio->tag_type = msg->task.tag_type;
1652 #ifdef CTL_TIME_IO
1653                         taskio->io_hdr.start_time = time_uptime;
1654                         getbinuptime(&taskio->io_hdr.start_bt);
1655 #endif /* CTL_TIME_IO */
1656                         ctl_run_task((union ctl_io *)taskio);
1657                         break;
1658                 }
1659                 /* Persistent Reserve action which needs attention */
1660                 case CTL_MSG_PERS_ACTION:
1661                         presio = (struct ctl_prio *)ctl_alloc_io(
1662                             softc->othersc_pool);
1663                         ctl_zero_io((union ctl_io *)presio);
1664                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1665                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1666                         presio->io_hdr.nexus = msg->hdr.nexus;
1667                         presio->pr_msg = msg->pr;
1668                         ctl_enqueue_isc((union ctl_io *)presio);
1669                         break;
1670                 case CTL_MSG_UA:
1671                         ctl_isc_ua(softc, msg, param);
1672                         break;
1673                 case CTL_MSG_PORT_SYNC:
1674                         ctl_isc_port_sync(softc, msg, param);
1675                         break;
1676                 case CTL_MSG_LUN_SYNC:
1677                         ctl_isc_lun_sync(softc, msg, param);
1678                         break;
1679                 case CTL_MSG_IID_SYNC:
1680                         ctl_isc_iid_sync(softc, msg, param);
1681                         break;
1682                 case CTL_MSG_LOGIN:
1683                         ctl_isc_login(softc, msg, param);
1684                         break;
1685                 case CTL_MSG_MODE_SYNC:
1686                         ctl_isc_mode_sync(softc, msg, param);
1687                         break;
1688                 default:
1689                         printf("Received HA message of unknown type %d\n",
1690                             msg->hdr.msg_type);
1691                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1692                         break;
1693                 }
1694                 if (msg != &msgbuf)
1695                         free(msg, M_CTL);
1696         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1697                 printf("CTL: HA link status changed from %d to %d\n",
1698                     softc->ha_link, param);
1699                 if (param == softc->ha_link)
1700                         return;
1701                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1702                         softc->ha_link = param;
1703                         ctl_isc_ha_link_down(softc);
1704                 } else {
1705                         softc->ha_link = param;
1706                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1707                                 ctl_isc_ha_link_up(softc);
1708                 }
1709                 return;
1710         } else {
1711                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1712                 return;
1713         }
1714 }
1715
1716 static void
1717 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1718 {
1719
1720         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1721             src->scsi.sense_len);
1722         dest->scsiio.scsi_status = src->scsi.scsi_status;
1723         dest->scsiio.sense_len = src->scsi.sense_len;
1724         dest->io_hdr.status = src->hdr.status;
1725 }
1726
1727 static void
1728 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1729 {
1730
1731         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1732             src->scsiio.sense_len);
1733         dest->scsi.scsi_status = src->scsiio.scsi_status;
1734         dest->scsi.sense_len = src->scsiio.sense_len;
1735         dest->hdr.status = src->io_hdr.status;
1736 }
1737
1738 void
1739 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1740 {
1741         struct ctl_softc *softc = lun->ctl_softc;
1742         ctl_ua_type *pu;
1743
1744         if (initidx < softc->init_min || initidx >= softc->init_max)
1745                 return;
1746         mtx_assert(&lun->lun_lock, MA_OWNED);
1747         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1748         if (pu == NULL)
1749                 return;
1750         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1751 }
1752
1753 void
1754 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1755 {
1756         int i;
1757
1758         mtx_assert(&lun->lun_lock, MA_OWNED);
1759         if (lun->pending_ua[port] == NULL)
1760                 return;
1761         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1762                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1763                         continue;
1764                 lun->pending_ua[port][i] |= ua;
1765         }
1766 }
1767
1768 void
1769 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1770 {
1771         struct ctl_softc *softc = lun->ctl_softc;
1772         int i;
1773
1774         mtx_assert(&lun->lun_lock, MA_OWNED);
1775         for (i = softc->port_min; i < softc->port_max; i++)
1776                 ctl_est_ua_port(lun, i, except, ua);
1777 }
1778
1779 void
1780 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1781 {
1782         struct ctl_softc *softc = lun->ctl_softc;
1783         ctl_ua_type *pu;
1784
1785         if (initidx < softc->init_min || initidx >= softc->init_max)
1786                 return;
1787         mtx_assert(&lun->lun_lock, MA_OWNED);
1788         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1789         if (pu == NULL)
1790                 return;
1791         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1792 }
1793
1794 void
1795 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1796 {
1797         struct ctl_softc *softc = lun->ctl_softc;
1798         int i, j;
1799
1800         mtx_assert(&lun->lun_lock, MA_OWNED);
1801         for (i = softc->port_min; i < softc->port_max; i++) {
1802                 if (lun->pending_ua[i] == NULL)
1803                         continue;
1804                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1805                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1806                                 continue;
1807                         lun->pending_ua[i][j] &= ~ua;
1808                 }
1809         }
1810 }
1811
1812 void
1813 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1814     ctl_ua_type ua_type)
1815 {
1816         struct ctl_lun *lun;
1817
1818         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1819         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1820                 mtx_lock(&lun->lun_lock);
1821                 ctl_clr_ua(lun, initidx, ua_type);
1822                 mtx_unlock(&lun->lun_lock);
1823         }
1824 }
1825
1826 static int
1827 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1828 {
1829         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1830         struct ctl_lun *lun;
1831         struct ctl_lun_req ireq;
1832         int error, value;
1833
1834         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1835         error = sysctl_handle_int(oidp, &value, 0, req);
1836         if ((error != 0) || (req->newptr == NULL))
1837                 return (error);
1838
1839         mtx_lock(&softc->ctl_lock);
1840         if (value == 0)
1841                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1842         else
1843                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1844         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1845                 mtx_unlock(&softc->ctl_lock);
1846                 bzero(&ireq, sizeof(ireq));
1847                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1848                 ireq.reqdata.modify.lun_id = lun->lun;
1849                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1850                     curthread);
1851                 if (ireq.status != CTL_LUN_OK) {
1852                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1853                             __func__, ireq.status, ireq.error_str);
1854                 }
1855                 mtx_lock(&softc->ctl_lock);
1856         }
1857         mtx_unlock(&softc->ctl_lock);
1858         return (0);
1859 }
1860
1861 static int
1862 ctl_init(void)
1863 {
1864         struct make_dev_args args;
1865         struct ctl_softc *softc;
1866         int i, error;
1867
1868         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1869                                M_WAITOK | M_ZERO);
1870
1871         make_dev_args_init(&args);
1872         args.mda_devsw = &ctl_cdevsw;
1873         args.mda_uid = UID_ROOT;
1874         args.mda_gid = GID_OPERATOR;
1875         args.mda_mode = 0600;
1876         args.mda_si_drv1 = softc;
1877         args.mda_si_drv2 = NULL;
1878         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1879         if (error != 0) {
1880                 free(softc, M_DEVBUF);
1881                 control_softc = NULL;
1882                 return (error);
1883         }
1884
1885         sysctl_ctx_init(&softc->sysctl_ctx);
1886         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1887                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1888                 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "CAM Target Layer");
1889
1890         if (softc->sysctl_tree == NULL) {
1891                 printf("%s: unable to allocate sysctl tree\n", __func__);
1892                 destroy_dev(softc->dev);
1893                 free(softc, M_DEVBUF);
1894                 control_softc = NULL;
1895                 return (ENOMEM);
1896         }
1897
1898         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1899         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1900             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1901         softc->flags = 0;
1902
1903         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1904             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1905             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1906
1907         if (ctl_max_luns <= 0 || powerof2(ctl_max_luns) == 0) {
1908                 printf("Bad value %d for kern.cam.ctl.max_luns, must be a power of two, using %d\n",
1909                     ctl_max_luns, CTL_DEFAULT_MAX_LUNS);
1910                 ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
1911         }
1912         softc->ctl_luns = malloc(sizeof(struct ctl_lun *) * ctl_max_luns,
1913             M_DEVBUF, M_WAITOK | M_ZERO);
1914         softc->ctl_lun_mask = malloc(sizeof(uint32_t) *
1915             ((ctl_max_luns + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1916         if (ctl_max_ports <= 0 || powerof2(ctl_max_ports) == 0) {
1917                 printf("Bad value %d for kern.cam.ctl.max_ports, must be a power of two, using %d\n",
1918                     ctl_max_ports, CTL_DEFAULT_MAX_PORTS);
1919                 ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
1920         }
1921         softc->ctl_port_mask = malloc(sizeof(uint32_t) *
1922           ((ctl_max_ports + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1923         softc->ctl_ports = malloc(sizeof(struct ctl_port *) * ctl_max_ports,
1924              M_DEVBUF, M_WAITOK | M_ZERO);
1925
1926         /*
1927          * In Copan's HA scheme, the "master" and "slave" roles are
1928          * figured out through the slot the controller is in.  Although it
1929          * is an active/active system, someone has to be in charge.
1930          */
1931         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1932             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1933             "HA head ID (0 - no HA)");
1934         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1935                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1936                 softc->is_single = 1;
1937                 softc->port_cnt = ctl_max_ports;
1938                 softc->port_min = 0;
1939         } else {
1940                 softc->port_cnt = ctl_max_ports / NUM_HA_SHELVES;
1941                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1942         }
1943         softc->port_max = softc->port_min + softc->port_cnt;
1944         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1945         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1946
1947         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1948             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1949             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1950
1951         STAILQ_INIT(&softc->lun_list);
1952         STAILQ_INIT(&softc->fe_list);
1953         STAILQ_INIT(&softc->port_list);
1954         STAILQ_INIT(&softc->be_list);
1955         ctl_tpc_init(softc);
1956
1957         if (worker_threads <= 0)
1958                 worker_threads = max(1, mp_ncpus / 4);
1959         if (worker_threads > CTL_MAX_THREADS)
1960                 worker_threads = CTL_MAX_THREADS;
1961
1962         for (i = 0; i < worker_threads; i++) {
1963                 struct ctl_thread *thr = &softc->threads[i];
1964
1965                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1966                 thr->ctl_softc = softc;
1967                 STAILQ_INIT(&thr->incoming_queue);
1968                 STAILQ_INIT(&thr->rtr_queue);
1969                 STAILQ_INIT(&thr->done_queue);
1970                 STAILQ_INIT(&thr->isc_queue);
1971
1972                 error = kproc_kthread_add(ctl_work_thread, thr,
1973                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1974                 if (error != 0) {
1975                         printf("error creating CTL work thread!\n");
1976                         return (error);
1977                 }
1978         }
1979         error = kproc_kthread_add(ctl_thresh_thread, softc,
1980             &softc->ctl_proc, &softc->thresh_thread, 0, 0, "ctl", "thresh");
1981         if (error != 0) {
1982                 printf("error creating CTL threshold thread!\n");
1983                 return (error);
1984         }
1985
1986         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1987             OID_AUTO, "ha_role",
1988             CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE,
1989             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1990
1991         if (softc->is_single == 0) {
1992                 if (ctl_frontend_register(&ha_frontend) != 0)
1993                         softc->is_single = 1;
1994         }
1995         return (0);
1996 }
1997
1998 static int
1999 ctl_shutdown(void)
2000 {
2001         struct ctl_softc *softc = control_softc;
2002         int i;
2003
2004         if (softc->is_single == 0)
2005                 ctl_frontend_deregister(&ha_frontend);
2006
2007         destroy_dev(softc->dev);
2008
2009         /* Shutdown CTL threads. */
2010         softc->shutdown = 1;
2011         for (i = 0; i < worker_threads; i++) {
2012                 struct ctl_thread *thr = &softc->threads[i];
2013                 while (thr->thread != NULL) {
2014                         wakeup(thr);
2015                         if (thr->thread != NULL)
2016                                 pause("CTL thr shutdown", 1);
2017                 }
2018                 mtx_destroy(&thr->queue_lock);
2019         }
2020         while (softc->thresh_thread != NULL) {
2021                 wakeup(softc->thresh_thread);
2022                 if (softc->thresh_thread != NULL)
2023                         pause("CTL thr shutdown", 1);
2024         }
2025
2026         ctl_tpc_shutdown(softc);
2027         uma_zdestroy(softc->io_zone);
2028         mtx_destroy(&softc->ctl_lock);
2029
2030         free(softc->ctl_luns, M_DEVBUF);
2031         free(softc->ctl_lun_mask, M_DEVBUF);
2032         free(softc->ctl_port_mask, M_DEVBUF);
2033         free(softc->ctl_ports, M_DEVBUF);
2034
2035         sysctl_ctx_free(&softc->sysctl_ctx);
2036
2037         free(softc, M_DEVBUF);
2038         control_softc = NULL;
2039         return (0);
2040 }
2041
2042 static int
2043 ctl_module_event_handler(module_t mod, int what, void *arg)
2044 {
2045
2046         switch (what) {
2047         case MOD_LOAD:
2048                 return (ctl_init());
2049         case MOD_UNLOAD:
2050                 return (ctl_shutdown());
2051         default:
2052                 return (EOPNOTSUPP);
2053         }
2054 }
2055
2056 /*
2057  * XXX KDM should we do some access checks here?  Bump a reference count to
2058  * prevent a CTL module from being unloaded while someone has it open?
2059  */
2060 static int
2061 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
2062 {
2063         return (0);
2064 }
2065
2066 static int
2067 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2068 {
2069         return (0);
2070 }
2071
2072 /*
2073  * Remove an initiator by port number and initiator ID.
2074  * Returns 0 for success, -1 for failure.
2075  */
2076 int
2077 ctl_remove_initiator(struct ctl_port *port, int iid)
2078 {
2079         struct ctl_softc *softc = port->ctl_softc;
2080         int last;
2081
2082         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2083
2084         if (iid > CTL_MAX_INIT_PER_PORT) {
2085                 printf("%s: initiator ID %u > maximun %u!\n",
2086                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2087                 return (-1);
2088         }
2089
2090         mtx_lock(&softc->ctl_lock);
2091         last = (--port->wwpn_iid[iid].in_use == 0);
2092         port->wwpn_iid[iid].last_use = time_uptime;
2093         mtx_unlock(&softc->ctl_lock);
2094         if (last)
2095                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
2096         ctl_isc_announce_iid(port, iid);
2097
2098         return (0);
2099 }
2100
2101 /*
2102  * Add an initiator to the initiator map.
2103  * Returns iid for success, < 0 for failure.
2104  */
2105 int
2106 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2107 {
2108         struct ctl_softc *softc = port->ctl_softc;
2109         time_t best_time;
2110         int i, best;
2111
2112         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2113
2114         if (iid >= CTL_MAX_INIT_PER_PORT) {
2115                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2116                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2117                 free(name, M_CTL);
2118                 return (-1);
2119         }
2120
2121         mtx_lock(&softc->ctl_lock);
2122
2123         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2124                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2125                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2126                                 iid = i;
2127                                 break;
2128                         }
2129                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2130                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2131                                 iid = i;
2132                                 break;
2133                         }
2134                 }
2135         }
2136
2137         if (iid < 0) {
2138                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2139                         if (port->wwpn_iid[i].in_use == 0 &&
2140                             port->wwpn_iid[i].wwpn == 0 &&
2141                             port->wwpn_iid[i].name == NULL) {
2142                                 iid = i;
2143                                 break;
2144                         }
2145                 }
2146         }
2147
2148         if (iid < 0) {
2149                 best = -1;
2150                 best_time = INT32_MAX;
2151                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2152                         if (port->wwpn_iid[i].in_use == 0) {
2153                                 if (port->wwpn_iid[i].last_use < best_time) {
2154                                         best = i;
2155                                         best_time = port->wwpn_iid[i].last_use;
2156                                 }
2157                         }
2158                 }
2159                 iid = best;
2160         }
2161
2162         if (iid < 0) {
2163                 mtx_unlock(&softc->ctl_lock);
2164                 free(name, M_CTL);
2165                 return (-2);
2166         }
2167
2168         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2169                 /*
2170                  * This is not an error yet.
2171                  */
2172                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2173 #if 0
2174                         printf("%s: port %d iid %u WWPN %#jx arrived"
2175                             " again\n", __func__, port->targ_port,
2176                             iid, (uintmax_t)wwpn);
2177 #endif
2178                         goto take;
2179                 }
2180                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2181                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2182 #if 0
2183                         printf("%s: port %d iid %u name '%s' arrived"
2184                             " again\n", __func__, port->targ_port,
2185                             iid, name);
2186 #endif
2187                         goto take;
2188                 }
2189
2190                 /*
2191                  * This is an error, but what do we do about it?  The
2192                  * driver is telling us we have a new WWPN for this
2193                  * initiator ID, so we pretty much need to use it.
2194                  */
2195                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2196                     " but WWPN %#jx '%s' is still at that address\n",
2197                     __func__, port->targ_port, iid, wwpn, name,
2198                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2199                     port->wwpn_iid[iid].name);
2200         }
2201 take:
2202         free(port->wwpn_iid[iid].name, M_CTL);
2203         port->wwpn_iid[iid].name = name;
2204         port->wwpn_iid[iid].wwpn = wwpn;
2205         port->wwpn_iid[iid].in_use++;
2206         mtx_unlock(&softc->ctl_lock);
2207         ctl_isc_announce_iid(port, iid);
2208
2209         return (iid);
2210 }
2211
2212 static int
2213 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2214 {
2215         int len;
2216
2217         switch (port->port_type) {
2218         case CTL_PORT_FC:
2219         {
2220                 struct scsi_transportid_fcp *id =
2221                     (struct scsi_transportid_fcp *)buf;
2222                 if (port->wwpn_iid[iid].wwpn == 0)
2223                         return (0);
2224                 memset(id, 0, sizeof(*id));
2225                 id->format_protocol = SCSI_PROTO_FC;
2226                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2227                 return (sizeof(*id));
2228         }
2229         case CTL_PORT_ISCSI:
2230         {
2231                 struct scsi_transportid_iscsi_port *id =
2232                     (struct scsi_transportid_iscsi_port *)buf;
2233                 if (port->wwpn_iid[iid].name == NULL)
2234                         return (0);
2235                 memset(id, 0, 256);
2236                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2237                     SCSI_PROTO_ISCSI;
2238                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2239                 len = roundup2(min(len, 252), 4);
2240                 scsi_ulto2b(len, id->additional_length);
2241                 return (sizeof(*id) + len);
2242         }
2243         case CTL_PORT_SAS:
2244         {
2245                 struct scsi_transportid_sas *id =
2246                     (struct scsi_transportid_sas *)buf;
2247                 if (port->wwpn_iid[iid].wwpn == 0)
2248                         return (0);
2249                 memset(id, 0, sizeof(*id));
2250                 id->format_protocol = SCSI_PROTO_SAS;
2251                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2252                 return (sizeof(*id));
2253         }
2254         default:
2255         {
2256                 struct scsi_transportid_spi *id =
2257                     (struct scsi_transportid_spi *)buf;
2258                 memset(id, 0, sizeof(*id));
2259                 id->format_protocol = SCSI_PROTO_SPI;
2260                 scsi_ulto2b(iid, id->scsi_addr);
2261                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2262                 return (sizeof(*id));
2263         }
2264         }
2265 }
2266
2267 /*
2268  * Serialize a command that went down the "wrong" side, and so was sent to
2269  * this controller for execution.  The logic is a little different than the
2270  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2271  * sent back to the other side, but in the success case, we execute the
2272  * command on this side (XFER mode) or tell the other side to execute it
2273  * (SER_ONLY mode).
2274  */
2275 static void
2276 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2277 {
2278         struct ctl_softc *softc = CTL_SOFTC(ctsio);
2279         struct ctl_port *port = CTL_PORT(ctsio);
2280         union ctl_ha_msg msg_info;
2281         struct ctl_lun *lun;
2282         const struct ctl_cmd_entry *entry;
2283         union ctl_io *bio;
2284         uint32_t targ_lun;
2285
2286         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2287
2288         /* Make sure that we know about this port. */
2289         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2290                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2291                                          /*retry_count*/ 1);
2292                 goto badjuju;
2293         }
2294
2295         /* Make sure that we know about this LUN. */
2296         mtx_lock(&softc->ctl_lock);
2297         if (targ_lun >= ctl_max_luns ||
2298             (lun = softc->ctl_luns[targ_lun]) == NULL) {
2299                 mtx_unlock(&softc->ctl_lock);
2300
2301                 /*
2302                  * The other node would not send this request to us unless
2303                  * received announce that we are primary node for this LUN.
2304                  * If this LUN does not exist now, it is probably result of
2305                  * a race, so respond to initiator in the most opaque way.
2306                  */
2307                 ctl_set_busy(ctsio);
2308                 goto badjuju;
2309         }
2310         mtx_lock(&lun->lun_lock);
2311         mtx_unlock(&softc->ctl_lock);
2312
2313         /*
2314          * If the LUN is invalid, pretend that it doesn't exist.
2315          * It will go away as soon as all pending I/Os completed.
2316          */
2317         if (lun->flags & CTL_LUN_DISABLED) {
2318                 mtx_unlock(&lun->lun_lock);
2319                 ctl_set_busy(ctsio);
2320                 goto badjuju;
2321         }
2322
2323         entry = ctl_get_cmd_entry(ctsio, NULL);
2324         ctsio->seridx = entry->seridx;
2325         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2326                 mtx_unlock(&lun->lun_lock);
2327                 goto badjuju;
2328         }
2329
2330         CTL_LUN(ctsio) = lun;
2331         CTL_BACKEND_LUN(ctsio) = lun->be_lun;
2332
2333         /*
2334          * Every I/O goes into the OOA queue for a
2335          * particular LUN, and stays there until completion.
2336          */
2337 #ifdef CTL_TIME_IO
2338         if (LIST_EMPTY(&lun->ooa_queue))
2339                 lun->idle_time += getsbinuptime() - lun->last_busy;
2340 #endif
2341         LIST_INSERT_HEAD(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2342
2343         bio = (union ctl_io *)LIST_NEXT(&ctsio->io_hdr, ooa_links);
2344         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio, &bio)) {
2345         case CTL_ACTION_PASS:
2346         case CTL_ACTION_SKIP:
2347                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2348                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2349                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2350                         mtx_unlock(&lun->lun_lock);
2351                 } else {
2352                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2353                         mtx_unlock(&lun->lun_lock);
2354
2355                         /* send msg back to other side */
2356                         msg_info.hdr.original_sc = ctsio->io_hdr.remote_io;
2357                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2358                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2359                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2360                             sizeof(msg_info.hdr), M_WAITOK);
2361                 }
2362                 break;
2363         case CTL_ACTION_BLOCK:
2364                 ctsio->io_hdr.blocker = bio;
2365                 TAILQ_INSERT_TAIL(&bio->io_hdr.blocked_queue, &ctsio->io_hdr,
2366                                   blocked_links);
2367                 mtx_unlock(&lun->lun_lock);
2368                 break;
2369         case CTL_ACTION_OVERLAP:
2370                 LIST_REMOVE(&ctsio->io_hdr, ooa_links);
2371                 mtx_unlock(&lun->lun_lock);
2372                 ctl_set_overlapped_cmd(ctsio);
2373                 goto badjuju;
2374         case CTL_ACTION_OVERLAP_TAG:
2375                 LIST_REMOVE(&ctsio->io_hdr, ooa_links);
2376                 mtx_unlock(&lun->lun_lock);
2377                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
2378 badjuju:
2379                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2380                 msg_info.hdr.original_sc = ctsio->io_hdr.remote_io;
2381                 msg_info.hdr.serializing_sc = NULL;
2382                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2383                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2384                     sizeof(msg_info.scsi), M_WAITOK);
2385                 ctl_free_io((union ctl_io *)ctsio);
2386                 break;
2387         default:
2388                 __assert_unreachable();
2389         }
2390 }
2391
2392 /*
2393  * Returns 0 for success, errno for failure.
2394  */
2395 static void
2396 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2397                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2398 {
2399         struct ctl_io_hdr *ioh;
2400
2401         mtx_lock(&lun->lun_lock);
2402         ioh = LIST_FIRST(&lun->ooa_queue);
2403         if (ioh == NULL) {
2404                 mtx_unlock(&lun->lun_lock);
2405                 return;
2406         }
2407         while (LIST_NEXT(ioh, ooa_links) != NULL)
2408                 ioh = LIST_NEXT(ioh, ooa_links);
2409         for ( ; ioh; ioh = LIST_PREV(ioh, &lun->ooa_queue, ctl_io_hdr, ooa_links)) {
2410                 union ctl_io *io = (union ctl_io *)ioh;
2411                 struct ctl_ooa_entry *entry;
2412
2413                 /*
2414                  * If we've got more than we can fit, just count the
2415                  * remaining entries.
2416                  */
2417                 if (*cur_fill_num >= ooa_hdr->alloc_num) {
2418                         (*cur_fill_num)++;
2419                         continue;
2420                 }
2421
2422                 entry = &kern_entries[*cur_fill_num];
2423
2424                 entry->tag_num = io->scsiio.tag_num;
2425                 entry->tag_type = io->scsiio.tag_type;
2426                 entry->lun_num = lun->lun;
2427 #ifdef CTL_TIME_IO
2428                 entry->start_bt = io->io_hdr.start_bt;
2429 #endif
2430                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2431                 entry->cdb_len = io->scsiio.cdb_len;
2432                 if (io->io_hdr.blocker != NULL)
2433                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2434
2435                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2436                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2437
2438                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2439                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2440
2441                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2442                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2443
2444                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2445                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2446
2447                 if (io->io_hdr.flags & CTL_FLAG_STATUS_QUEUED)
2448                         entry->cmd_flags |= CTL_OOACMD_FLAG_STATUS_QUEUED;
2449
2450                 if (io->io_hdr.flags & CTL_FLAG_STATUS_SENT)
2451                         entry->cmd_flags |= CTL_OOACMD_FLAG_STATUS_SENT;
2452                 (*cur_fill_num)++;
2453         }
2454         mtx_unlock(&lun->lun_lock);
2455 }
2456
2457 /*
2458  * Escape characters that are illegal or not recommended in XML.
2459  */
2460 int
2461 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2462 {
2463         char *end = str + size;
2464         int retval;
2465
2466         retval = 0;
2467
2468         for (; *str && str < end; str++) {
2469                 switch (*str) {
2470                 case '&':
2471                         retval = sbuf_cat(sb, "&amp;");
2472                         break;
2473                 case '>':
2474                         retval = sbuf_cat(sb, "&gt;");
2475                         break;
2476                 case '<':
2477                         retval = sbuf_cat(sb, "&lt;");
2478                         break;
2479                 default:
2480                         retval = sbuf_putc(sb, *str);
2481                         break;
2482                 }
2483
2484                 if (retval != 0)
2485                         break;
2486         }
2487
2488         return (retval);
2489 }
2490
2491 static void
2492 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2493 {
2494         struct scsi_vpd_id_descriptor *desc;
2495         int i;
2496
2497         if (id == NULL || id->len < 4)
2498                 return;
2499         desc = (struct scsi_vpd_id_descriptor *)id->data;
2500         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2501         case SVPD_ID_TYPE_T10:
2502                 sbuf_cat(sb, "t10.");
2503                 break;
2504         case SVPD_ID_TYPE_EUI64:
2505                 sbuf_cat(sb, "eui.");
2506                 break;
2507         case SVPD_ID_TYPE_NAA:
2508                 sbuf_cat(sb, "naa.");
2509                 break;
2510         case SVPD_ID_TYPE_SCSI_NAME:
2511                 break;
2512         }
2513         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2514         case SVPD_ID_CODESET_BINARY:
2515                 for (i = 0; i < desc->length; i++)
2516                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2517                 break;
2518         case SVPD_ID_CODESET_ASCII:
2519                 sbuf_printf(sb, "%.*s", (int)desc->length,
2520                     (char *)desc->identifier);
2521                 break;
2522         case SVPD_ID_CODESET_UTF8:
2523                 sbuf_cat(sb, (char *)desc->identifier);
2524                 break;
2525         }
2526 }
2527
2528 static int
2529 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2530           struct thread *td)
2531 {
2532         struct ctl_softc *softc = dev->si_drv1;
2533         struct ctl_port *port;
2534         struct ctl_lun *lun;
2535         int retval;
2536
2537         retval = 0;
2538
2539         switch (cmd) {
2540         case CTL_IO:
2541                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2542                 break;
2543         case CTL_ENABLE_PORT:
2544         case CTL_DISABLE_PORT:
2545         case CTL_SET_PORT_WWNS: {
2546                 struct ctl_port *port;
2547                 struct ctl_port_entry *entry;
2548
2549                 entry = (struct ctl_port_entry *)addr;
2550
2551                 mtx_lock(&softc->ctl_lock);
2552                 STAILQ_FOREACH(port, &softc->port_list, links) {
2553                         int action, done;
2554
2555                         if (port->targ_port < softc->port_min ||
2556                             port->targ_port >= softc->port_max)
2557                                 continue;
2558
2559                         action = 0;
2560                         done = 0;
2561                         if ((entry->port_type == CTL_PORT_NONE)
2562                          && (entry->targ_port == port->targ_port)) {
2563                                 /*
2564                                  * If the user only wants to enable or
2565                                  * disable or set WWNs on a specific port,
2566                                  * do the operation and we're done.
2567                                  */
2568                                 action = 1;
2569                                 done = 1;
2570                         } else if (entry->port_type & port->port_type) {
2571                                 /*
2572                                  * Compare the user's type mask with the
2573                                  * particular frontend type to see if we
2574                                  * have a match.
2575                                  */
2576                                 action = 1;
2577                                 done = 0;
2578
2579                                 /*
2580                                  * Make sure the user isn't trying to set
2581                                  * WWNs on multiple ports at the same time.
2582                                  */
2583                                 if (cmd == CTL_SET_PORT_WWNS) {
2584                                         printf("%s: Can't set WWNs on "
2585                                                "multiple ports\n", __func__);
2586                                         retval = EINVAL;
2587                                         break;
2588                                 }
2589                         }
2590                         if (action == 0)
2591                                 continue;
2592
2593                         /*
2594                          * XXX KDM we have to drop the lock here, because
2595                          * the online/offline operations can potentially
2596                          * block.  We need to reference count the frontends
2597                          * so they can't go away,
2598                          */
2599                         if (cmd == CTL_ENABLE_PORT) {
2600                                 mtx_unlock(&softc->ctl_lock);
2601                                 ctl_port_online(port);
2602                                 mtx_lock(&softc->ctl_lock);
2603                         } else if (cmd == CTL_DISABLE_PORT) {
2604                                 mtx_unlock(&softc->ctl_lock);
2605                                 ctl_port_offline(port);
2606                                 mtx_lock(&softc->ctl_lock);
2607                         } else if (cmd == CTL_SET_PORT_WWNS) {
2608                                 ctl_port_set_wwns(port,
2609                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2610                                     1 : 0, entry->wwnn,
2611                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2612                                     1 : 0, entry->wwpn);
2613                         }
2614                         if (done != 0)
2615                                 break;
2616                 }
2617                 mtx_unlock(&softc->ctl_lock);
2618                 break;
2619         }
2620         case CTL_GET_OOA: {
2621                 struct ctl_ooa *ooa_hdr;
2622                 struct ctl_ooa_entry *entries;
2623                 uint32_t cur_fill_num;
2624
2625                 ooa_hdr = (struct ctl_ooa *)addr;
2626
2627                 if ((ooa_hdr->alloc_len == 0)
2628                  || (ooa_hdr->alloc_num == 0)) {
2629                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2630                                "must be non-zero\n", __func__,
2631                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2632                         retval = EINVAL;
2633                         break;
2634                 }
2635
2636                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2637                     sizeof(struct ctl_ooa_entry))) {
2638                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2639                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2640                                __func__, ooa_hdr->alloc_len,
2641                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2642                         retval = EINVAL;
2643                         break;
2644                 }
2645
2646                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2647                 if (entries == NULL) {
2648                         printf("%s: could not allocate %d bytes for OOA "
2649                                "dump\n", __func__, ooa_hdr->alloc_len);
2650                         retval = ENOMEM;
2651                         break;
2652                 }
2653
2654                 mtx_lock(&softc->ctl_lock);
2655                 if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 &&
2656                     (ooa_hdr->lun_num >= ctl_max_luns ||
2657                      softc->ctl_luns[ooa_hdr->lun_num] == NULL)) {
2658                         mtx_unlock(&softc->ctl_lock);
2659                         free(entries, M_CTL);
2660                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2661                                __func__, (uintmax_t)ooa_hdr->lun_num);
2662                         retval = EINVAL;
2663                         break;
2664                 }
2665
2666                 cur_fill_num = 0;
2667
2668                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2669                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2670                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2671                                     ooa_hdr, entries);
2672                         }
2673                 } else {
2674                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2675                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2676                             entries);
2677                 }
2678                 mtx_unlock(&softc->ctl_lock);
2679
2680                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2681                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2682                         sizeof(struct ctl_ooa_entry);
2683                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2684                 if (retval != 0) {
2685                         printf("%s: error copying out %d bytes for OOA dump\n",
2686                                __func__, ooa_hdr->fill_len);
2687                 }
2688
2689                 getbinuptime(&ooa_hdr->cur_bt);
2690
2691                 if (cur_fill_num > ooa_hdr->alloc_num) {
2692                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2693                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2694                 } else {
2695                         ooa_hdr->dropped_num = 0;
2696                         ooa_hdr->status = CTL_OOA_OK;
2697                 }
2698
2699                 free(entries, M_CTL);
2700                 break;
2701         }
2702         case CTL_DELAY_IO: {
2703                 struct ctl_io_delay_info *delay_info;
2704
2705                 delay_info = (struct ctl_io_delay_info *)addr;
2706
2707 #ifdef CTL_IO_DELAY
2708                 mtx_lock(&softc->ctl_lock);
2709                 if (delay_info->lun_id >= ctl_max_luns ||
2710                     (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) {
2711                         mtx_unlock(&softc->ctl_lock);
2712                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2713                         break;
2714                 }
2715                 mtx_lock(&lun->lun_lock);
2716                 mtx_unlock(&softc->ctl_lock);
2717                 delay_info->status = CTL_DELAY_STATUS_OK;
2718                 switch (delay_info->delay_type) {
2719                 case CTL_DELAY_TYPE_CONT:
2720                 case CTL_DELAY_TYPE_ONESHOT:
2721                         break;
2722                 default:
2723                         delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE;
2724                         break;
2725                 }
2726                 switch (delay_info->delay_loc) {
2727                 case CTL_DELAY_LOC_DATAMOVE:
2728                         lun->delay_info.datamove_type = delay_info->delay_type;
2729                         lun->delay_info.datamove_delay = delay_info->delay_secs;
2730                         break;
2731                 case CTL_DELAY_LOC_DONE:
2732                         lun->delay_info.done_type = delay_info->delay_type;
2733                         lun->delay_info.done_delay = delay_info->delay_secs;
2734                         break;
2735                 default:
2736                         delay_info->status = CTL_DELAY_STATUS_INVALID_LOC;
2737                         break;
2738                 }
2739                 mtx_unlock(&lun->lun_lock);
2740 #else
2741                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2742 #endif /* CTL_IO_DELAY */
2743                 break;
2744         }
2745         case CTL_ERROR_INJECT: {
2746                 struct ctl_error_desc *err_desc, *new_err_desc;
2747
2748                 err_desc = (struct ctl_error_desc *)addr;
2749
2750                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2751                                       M_WAITOK | M_ZERO);
2752                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2753
2754                 mtx_lock(&softc->ctl_lock);
2755                 if (err_desc->lun_id >= ctl_max_luns ||
2756                     (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) {
2757                         mtx_unlock(&softc->ctl_lock);
2758                         free(new_err_desc, M_CTL);
2759                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2760                                __func__, (uintmax_t)err_desc->lun_id);
2761                         retval = EINVAL;
2762                         break;
2763                 }
2764                 mtx_lock(&lun->lun_lock);
2765                 mtx_unlock(&softc->ctl_lock);
2766
2767                 /*
2768                  * We could do some checking here to verify the validity
2769                  * of the request, but given the complexity of error
2770                  * injection requests, the checking logic would be fairly
2771                  * complex.
2772                  *
2773                  * For now, if the request is invalid, it just won't get
2774                  * executed and might get deleted.
2775                  */
2776                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2777
2778                 /*
2779                  * XXX KDM check to make sure the serial number is unique,
2780                  * in case we somehow manage to wrap.  That shouldn't
2781                  * happen for a very long time, but it's the right thing to
2782                  * do.
2783                  */
2784                 new_err_desc->serial = lun->error_serial;
2785                 err_desc->serial = lun->error_serial;
2786                 lun->error_serial++;
2787
2788                 mtx_unlock(&lun->lun_lock);
2789                 break;
2790         }
2791         case CTL_ERROR_INJECT_DELETE: {
2792                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2793                 int delete_done;
2794
2795                 delete_desc = (struct ctl_error_desc *)addr;
2796                 delete_done = 0;
2797
2798                 mtx_lock(&softc->ctl_lock);
2799                 if (delete_desc->lun_id >= ctl_max_luns ||
2800                     (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) {
2801                         mtx_unlock(&softc->ctl_lock);
2802                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2803                                __func__, (uintmax_t)delete_desc->lun_id);
2804                         retval = EINVAL;
2805                         break;
2806                 }
2807                 mtx_lock(&lun->lun_lock);
2808                 mtx_unlock(&softc->ctl_lock);
2809                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2810                         if (desc->serial != delete_desc->serial)
2811                                 continue;
2812
2813                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2814                                       links);
2815                         free(desc, M_CTL);
2816                         delete_done = 1;
2817                 }
2818                 mtx_unlock(&lun->lun_lock);
2819                 if (delete_done == 0) {
2820                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2821                                "error serial %ju on LUN %u\n", __func__,
2822                                delete_desc->serial, delete_desc->lun_id);
2823                         retval = EINVAL;
2824                         break;
2825                 }
2826                 break;
2827         }
2828         case CTL_DUMP_STRUCTS: {
2829                 int j, k;
2830                 struct ctl_port *port;
2831                 struct ctl_frontend *fe;
2832
2833                 mtx_lock(&softc->ctl_lock);
2834                 printf("CTL Persistent Reservation information start:\n");
2835                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2836                         mtx_lock(&lun->lun_lock);
2837                         if ((lun->flags & CTL_LUN_DISABLED) != 0) {
2838                                 mtx_unlock(&lun->lun_lock);
2839                                 continue;
2840                         }
2841
2842                         for (j = 0; j < ctl_max_ports; j++) {
2843                                 if (lun->pr_keys[j] == NULL)
2844                                         continue;
2845                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2846                                         if (lun->pr_keys[j][k] == 0)
2847                                                 continue;
2848                                         printf("  LUN %ju port %d iid %d key "
2849                                                "%#jx\n", lun->lun, j, k,
2850                                                (uintmax_t)lun->pr_keys[j][k]);
2851                                 }
2852                         }
2853                         mtx_unlock(&lun->lun_lock);
2854                 }
2855                 printf("CTL Persistent Reservation information end\n");
2856                 printf("CTL Ports:\n");
2857                 STAILQ_FOREACH(port, &softc->port_list, links) {
2858                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2859                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2860                                port->frontend->name, port->port_type,
2861                                port->physical_port, port->virtual_port,
2862                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2863                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2864                                 if (port->wwpn_iid[j].in_use == 0 &&
2865                                     port->wwpn_iid[j].wwpn == 0 &&
2866                                     port->wwpn_iid[j].name == NULL)
2867                                         continue;
2868
2869                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2870                                     j, port->wwpn_iid[j].in_use,
2871                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2872                                     port->wwpn_iid[j].name);
2873                         }
2874                 }
2875                 printf("CTL Port information end\n");
2876                 mtx_unlock(&softc->ctl_lock);
2877                 /*
2878                  * XXX KDM calling this without a lock.  We'd likely want
2879                  * to drop the lock before calling the frontend's dump
2880                  * routine anyway.
2881                  */
2882                 printf("CTL Frontends:\n");
2883                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2884                         printf("  Frontend '%s'\n", fe->name);
2885                         if (fe->fe_dump != NULL)
2886                                 fe->fe_dump();
2887                 }
2888                 printf("CTL Frontend information end\n");
2889                 break;
2890         }
2891         case CTL_LUN_REQ: {
2892                 struct ctl_lun_req *lun_req;
2893                 struct ctl_backend_driver *backend;
2894                 void *packed;
2895                 nvlist_t *tmp_args_nvl;
2896                 size_t packed_len;
2897
2898                 lun_req = (struct ctl_lun_req *)addr;
2899                 tmp_args_nvl = lun_req->args_nvl;
2900
2901                 backend = ctl_backend_find(lun_req->backend);
2902                 if (backend == NULL) {
2903                         lun_req->status = CTL_LUN_ERROR;
2904                         snprintf(lun_req->error_str,
2905                                  sizeof(lun_req->error_str),
2906                                  "Backend \"%s\" not found.",
2907                                  lun_req->backend);
2908                         break;
2909                 }
2910
2911                 if (lun_req->args != NULL) {
2912                         if (lun_req->args_len > CTL_MAX_ARGS_LEN) {
2913                                 lun_req->status = CTL_LUN_ERROR;
2914                                 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2915                                     "Too big args.");
2916                                 break;
2917                         }
2918                         packed = malloc(lun_req->args_len, M_CTL, M_WAITOK);
2919                         if (copyin(lun_req->args, packed, lun_req->args_len) != 0) {
2920                                 free(packed, M_CTL);
2921                                 lun_req->status = CTL_LUN_ERROR;
2922                                 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2923                                     "Cannot copyin args.");
2924                                 break;
2925                         }
2926                         lun_req->args_nvl = nvlist_unpack(packed,
2927                             lun_req->args_len, 0);
2928                         free(packed, M_CTL);
2929
2930                         if (lun_req->args_nvl == NULL) {
2931                                 lun_req->status = CTL_LUN_ERROR;
2932                                 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2933                                     "Cannot unpack args nvlist.");
2934                                 break;
2935                         }
2936                 } else
2937                         lun_req->args_nvl = nvlist_create(0);
2938
2939                 lun_req->result_nvl = NULL;
2940                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2941                 nvlist_destroy(lun_req->args_nvl);
2942                 lun_req->args_nvl = tmp_args_nvl;
2943
2944                 if (lun_req->result_nvl != NULL) {
2945                         if (lun_req->result != NULL) {
2946                                 packed = nvlist_pack(lun_req->result_nvl,
2947                                     &packed_len);
2948                                 if (packed == NULL) {
2949                                         lun_req->status = CTL_LUN_ERROR;
2950                                         snprintf(lun_req->error_str,
2951                                             sizeof(lun_req->error_str),
2952                                             "Cannot pack result nvlist.");
2953                                         break;
2954                                 }
2955
2956                                 if (packed_len > lun_req->result_len) {
2957                                         lun_req->status = CTL_LUN_ERROR;
2958                                         snprintf(lun_req->error_str,
2959                                             sizeof(lun_req->error_str),
2960                                             "Result nvlist too large.");
2961                                         free(packed, M_NVLIST);
2962                                         break;
2963                                 }
2964
2965                                 if (copyout(packed, lun_req->result, packed_len)) {
2966                                         lun_req->status = CTL_LUN_ERROR;
2967                                         snprintf(lun_req->error_str,
2968                                             sizeof(lun_req->error_str),
2969                                             "Cannot copyout() the result.");
2970                                         free(packed, M_NVLIST);
2971                                         break;
2972                                 }
2973
2974                                 lun_req->result_len = packed_len;
2975                                 free(packed, M_NVLIST);
2976                         }
2977
2978                         nvlist_destroy(lun_req->result_nvl);
2979                 }
2980                 break;
2981         }
2982         case CTL_LUN_LIST: {
2983                 struct sbuf *sb;
2984                 struct ctl_lun_list *list;
2985                 const char *name, *value;
2986                 void *cookie;
2987                 int type;
2988
2989                 list = (struct ctl_lun_list *)addr;
2990
2991                 /*
2992                  * Allocate a fixed length sbuf here, based on the length
2993                  * of the user's buffer.  We could allocate an auto-extending
2994                  * buffer, and then tell the user how much larger our
2995                  * amount of data is than his buffer, but that presents
2996                  * some problems:
2997                  *
2998                  * 1.  The sbuf(9) routines use a blocking malloc, and so
2999                  *     we can't hold a lock while calling them with an
3000                  *     auto-extending buffer.
3001                  *
3002                  * 2.  There is not currently a LUN reference counting
3003                  *     mechanism, outside of outstanding transactions on
3004                  *     the LUN's OOA queue.  So a LUN could go away on us
3005                  *     while we're getting the LUN number, backend-specific
3006                  *     information, etc.  Thus, given the way things
3007                  *     currently work, we need to hold the CTL lock while
3008                  *     grabbing LUN information.
3009                  *
3010                  * So, from the user's standpoint, the best thing to do is
3011                  * allocate what he thinks is a reasonable buffer length,
3012                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3013                  * double the buffer length and try again.  (And repeat
3014                  * that until he succeeds.)
3015                  */
3016                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3017                 if (sb == NULL) {
3018                         list->status = CTL_LUN_LIST_ERROR;
3019                         snprintf(list->error_str, sizeof(list->error_str),
3020                                  "Unable to allocate %d bytes for LUN list",
3021                                  list->alloc_len);
3022                         break;
3023                 }
3024
3025                 sbuf_cat(sb, "<ctllunlist>\n");
3026
3027                 mtx_lock(&softc->ctl_lock);
3028                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3029                         mtx_lock(&lun->lun_lock);
3030                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3031                                              (uintmax_t)lun->lun);
3032
3033                         /*
3034                          * Bail out as soon as we see that we've overfilled
3035                          * the buffer.
3036                          */
3037                         if (retval != 0)
3038                                 break;
3039
3040                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3041                                              "</backend_type>\n",
3042                                              (lun->backend == NULL) ?  "none" :
3043                                              lun->backend->name);
3044
3045                         if (retval != 0)
3046                                 break;
3047
3048                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3049                                              lun->be_lun->lun_type);
3050
3051                         if (retval != 0)
3052                                 break;
3053
3054                         if (lun->backend == NULL) {
3055                                 retval = sbuf_cat(sb, "</lun>\n");
3056                                 if (retval != 0)
3057                                         break;
3058                                 continue;
3059                         }
3060
3061                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3062                                              (lun->be_lun->maxlba > 0) ?
3063                                              lun->be_lun->maxlba + 1 : 0);
3064
3065                         if (retval != 0)
3066                                 break;
3067
3068                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3069                                              lun->be_lun->blocksize);
3070
3071                         if (retval != 0)
3072                                 break;
3073
3074                         retval = sbuf_cat(sb, "\t<serial_number>");
3075
3076                         if (retval != 0)
3077                                 break;
3078
3079                         retval = ctl_sbuf_printf_esc(sb,
3080                             lun->be_lun->serial_num,
3081                             sizeof(lun->be_lun->serial_num));
3082
3083                         if (retval != 0)
3084                                 break;
3085
3086                         retval = sbuf_cat(sb, "</serial_number>\n");
3087
3088                         if (retval != 0)
3089                                 break;
3090
3091                         retval = sbuf_cat(sb, "\t<device_id>");
3092
3093                         if (retval != 0)
3094                                 break;
3095
3096                         retval = ctl_sbuf_printf_esc(sb,
3097                             lun->be_lun->device_id,
3098                             sizeof(lun->be_lun->device_id));
3099
3100                         if (retval != 0)
3101                                 break;
3102
3103                         retval = sbuf_cat(sb, "</device_id>\n");
3104
3105                         if (retval != 0)
3106                                 break;
3107
3108                         if (lun->backend->lun_info != NULL) {
3109                                 retval = lun->backend->lun_info(lun->be_lun, sb);
3110                                 if (retval != 0)
3111                                         break;
3112                         }
3113
3114                         cookie = NULL;
3115                         while ((name = nvlist_next(lun->be_lun->options, &type,
3116                             &cookie)) != NULL) {
3117                                 sbuf_printf(sb, "\t<%s>", name);
3118
3119                                 if (type == NV_TYPE_STRING) {
3120                                         value = dnvlist_get_string(
3121                                             lun->be_lun->options, name, NULL);
3122                                         if (value != NULL)
3123                                                 sbuf_cat(sb, value);
3124                                 }
3125
3126                                 sbuf_printf(sb, "</%s>\n", name);
3127                         }
3128
3129                         retval = sbuf_cat(sb, "</lun>\n");
3130
3131                         if (retval != 0)
3132                                 break;
3133                         mtx_unlock(&lun->lun_lock);
3134                 }
3135                 if (lun != NULL)
3136                         mtx_unlock(&lun->lun_lock);
3137                 mtx_unlock(&softc->ctl_lock);
3138
3139                 if ((retval != 0)
3140                  || ((retval = sbuf_cat(sb, "</ctllunlist>\n")) != 0)) {
3141                         retval = 0;
3142                         sbuf_delete(sb);
3143                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3144                         snprintf(list->error_str, sizeof(list->error_str),
3145                                  "Out of space, %d bytes is too small",
3146                                  list->alloc_len);
3147                         break;
3148                 }
3149
3150                 sbuf_finish(sb);
3151
3152                 retval = copyout(sbuf_data(sb), list->lun_xml,
3153                                  sbuf_len(sb) + 1);
3154
3155                 list->fill_len = sbuf_len(sb) + 1;
3156                 list->status = CTL_LUN_LIST_OK;
3157                 sbuf_delete(sb);
3158                 break;
3159         }
3160         case CTL_ISCSI: {
3161                 struct ctl_iscsi *ci;
3162                 struct ctl_frontend *fe;
3163
3164                 ci = (struct ctl_iscsi *)addr;
3165
3166                 fe = ctl_frontend_find("iscsi");
3167                 if (fe == NULL) {
3168                         ci->status = CTL_ISCSI_ERROR;
3169                         snprintf(ci->error_str, sizeof(ci->error_str),
3170                             "Frontend \"iscsi\" not found.");
3171                         break;
3172                 }
3173
3174                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3175                 break;
3176         }
3177         case CTL_PORT_REQ: {
3178                 struct ctl_req *req;
3179                 struct ctl_frontend *fe;
3180                 void *packed;
3181                 nvlist_t *tmp_args_nvl;
3182                 size_t packed_len;
3183
3184                 req = (struct ctl_req *)addr;
3185                 tmp_args_nvl = req->args_nvl;
3186
3187                 fe = ctl_frontend_find(req->driver);
3188                 if (fe == NULL) {
3189                         req->status = CTL_LUN_ERROR;
3190                         snprintf(req->error_str, sizeof(req->error_str),
3191                             "Frontend \"%s\" not found.", req->driver);
3192                         break;
3193                 }
3194
3195                 if (req->args != NULL) {
3196                         if (req->args_len > CTL_MAX_ARGS_LEN) {
3197                                 req->status = CTL_LUN_ERROR;
3198                                 snprintf(req->error_str, sizeof(req->error_str),
3199                                     "Too big args.");
3200                                 break;
3201                         }
3202                         packed = malloc(req->args_len, M_CTL, M_WAITOK);
3203                         if (copyin(req->args, packed, req->args_len) != 0) {
3204                                 free(packed, M_CTL);
3205                                 req->status = CTL_LUN_ERROR;
3206                                 snprintf(req->error_str, sizeof(req->error_str),
3207                                     "Cannot copyin args.");
3208                                 break;
3209                         }
3210                         req->args_nvl = nvlist_unpack(packed,
3211                             req->args_len, 0);
3212                         free(packed, M_CTL);
3213
3214                         if (req->args_nvl == NULL) {
3215                                 req->status = CTL_LUN_ERROR;
3216                                 snprintf(req->error_str, sizeof(req->error_str),
3217                                     "Cannot unpack args nvlist.");
3218                                 break;
3219                         }
3220                 } else
3221                         req->args_nvl = nvlist_create(0);
3222
3223                 req->result_nvl = NULL;
3224                 if (fe->ioctl)
3225                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3226                 else
3227                         retval = ENODEV;
3228
3229                 nvlist_destroy(req->args_nvl);
3230                 req->args_nvl = tmp_args_nvl;
3231
3232                 if (req->result_nvl != NULL) {
3233                         if (req->result != NULL) {
3234                                 packed = nvlist_pack(req->result_nvl,
3235                                     &packed_len);
3236                                 if (packed == NULL) {
3237                                         req->status = CTL_LUN_ERROR;
3238                                         snprintf(req->error_str,
3239                                             sizeof(req->error_str),
3240                                             "Cannot pack result nvlist.");
3241                                         break;
3242                                 }
3243
3244                                 if (packed_len > req->result_len) {
3245                                         req->status = CTL_LUN_ERROR;
3246                                         snprintf(req->error_str,
3247                                             sizeof(req->error_str),
3248                                             "Result nvlist too large.");
3249                                         free(packed, M_NVLIST);
3250                                         break;
3251                                 }
3252
3253                                 if (copyout(packed, req->result, packed_len)) {
3254                                         req->status = CTL_LUN_ERROR;
3255                                         snprintf(req->error_str,
3256                                             sizeof(req->error_str),
3257                                             "Cannot copyout() the result.");
3258                                         free(packed, M_NVLIST);
3259                                         break;
3260                                 }
3261
3262                                 req->result_len = packed_len;
3263                                 free(packed, M_NVLIST);
3264                         }
3265
3266                         nvlist_destroy(req->result_nvl);
3267                 }
3268                 break;
3269         }
3270         case CTL_PORT_LIST: {
3271                 struct sbuf *sb;
3272                 struct ctl_port *port;
3273                 struct ctl_lun_list *list;
3274                 const char *name, *value;
3275                 void *cookie;
3276                 int j, type;
3277                 uint32_t plun;
3278
3279                 list = (struct ctl_lun_list *)addr;
3280
3281                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3282                 if (sb == NULL) {
3283                         list->status = CTL_LUN_LIST_ERROR;
3284                         snprintf(list->error_str, sizeof(list->error_str),
3285                                  "Unable to allocate %d bytes for LUN list",
3286                                  list->alloc_len);
3287                         break;
3288                 }
3289
3290                 sbuf_cat(sb, "<ctlportlist>\n");
3291
3292                 mtx_lock(&softc->ctl_lock);
3293                 STAILQ_FOREACH(port, &softc->port_list, links) {
3294                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3295                                              (uintmax_t)port->targ_port);
3296
3297                         /*
3298                          * Bail out as soon as we see that we've overfilled
3299                          * the buffer.
3300                          */
3301                         if (retval != 0)
3302                                 break;
3303
3304                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3305                             "</frontend_type>\n", port->frontend->name);
3306                         if (retval != 0)
3307                                 break;
3308
3309                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3310                                              port->port_type);
3311                         if (retval != 0)
3312                                 break;
3313
3314                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3315                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3316                         if (retval != 0)
3317                                 break;
3318
3319                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3320                             port->port_name);
3321                         if (retval != 0)
3322                                 break;
3323
3324                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3325                             port->physical_port);
3326                         if (retval != 0)
3327                                 break;
3328
3329                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3330                             port->virtual_port);
3331                         if (retval != 0)
3332                                 break;
3333
3334                         if (port->target_devid != NULL) {
3335                                 sbuf_cat(sb, "\t<target>");
3336                                 ctl_id_sbuf(port->target_devid, sb);
3337                                 sbuf_cat(sb, "</target>\n");
3338                         }
3339
3340                         if (port->port_devid != NULL) {
3341                                 sbuf_cat(sb, "\t<port>");
3342                                 ctl_id_sbuf(port->port_devid, sb);
3343                                 sbuf_cat(sb, "</port>\n");
3344                         }
3345
3346                         if (port->port_info != NULL) {
3347                                 retval = port->port_info(port->onoff_arg, sb);
3348                                 if (retval != 0)
3349                                         break;
3350                         }
3351
3352                         cookie = NULL;
3353                         while ((name = nvlist_next(port->options, &type,
3354                             &cookie)) != NULL) {
3355                                 sbuf_printf(sb, "\t<%s>", name);
3356
3357                                 if (type == NV_TYPE_STRING) {
3358                                         value = dnvlist_get_string(port->options,
3359                                             name, NULL);
3360                                         if (value != NULL)
3361                                                 sbuf_printf(sb, "%s", value);
3362                                 }
3363
3364                                 sbuf_printf(sb, "</%s>\n", name);
3365                         }
3366
3367                         if (port->lun_map != NULL) {
3368                                 sbuf_cat(sb, "\t<lun_map>on</lun_map>\n");
3369                                 for (j = 0; j < port->lun_map_size; j++) {
3370                                         plun = ctl_lun_map_from_port(port, j);
3371                                         if (plun == UINT32_MAX)
3372                                                 continue;
3373                                         sbuf_printf(sb,
3374                                             "\t<lun id=\"%u\">%u</lun>\n",
3375                                             j, plun);
3376                                 }
3377                         }
3378
3379                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3380                                 if (port->wwpn_iid[j].in_use == 0 ||
3381                                     (port->wwpn_iid[j].wwpn == 0 &&
3382                                      port->wwpn_iid[j].name == NULL))
3383                                         continue;
3384
3385                                 if (port->wwpn_iid[j].name != NULL)
3386                                         retval = sbuf_printf(sb,
3387                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3388                                             j, port->wwpn_iid[j].name);
3389                                 else
3390                                         retval = sbuf_printf(sb,
3391                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3392                                             j, port->wwpn_iid[j].wwpn);
3393                                 if (retval != 0)
3394                                         break;
3395                         }
3396                         if (retval != 0)
3397                                 break;
3398
3399                         retval = sbuf_cat(sb, "</targ_port>\n");
3400                         if (retval != 0)
3401                                 break;
3402                 }
3403                 mtx_unlock(&softc->ctl_lock);
3404
3405                 if ((retval != 0)
3406                  || ((retval = sbuf_cat(sb, "</ctlportlist>\n")) != 0)) {
3407                         retval = 0;
3408                         sbuf_delete(sb);
3409                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3410                         snprintf(list->error_str, sizeof(list->error_str),
3411                                  "Out of space, %d bytes is too small",
3412                                  list->alloc_len);
3413                         break;
3414                 }
3415
3416                 sbuf_finish(sb);
3417
3418                 retval = copyout(sbuf_data(sb), list->lun_xml,
3419                                  sbuf_len(sb) + 1);
3420
3421                 list->fill_len = sbuf_len(sb) + 1;
3422                 list->status = CTL_LUN_LIST_OK;
3423                 sbuf_delete(sb);
3424                 break;
3425         }
3426         case CTL_LUN_MAP: {
3427                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3428                 struct ctl_port *port;
3429
3430                 mtx_lock(&softc->ctl_lock);
3431                 if (lm->port < softc->port_min ||
3432                     lm->port >= softc->port_max ||
3433                     (port = softc->ctl_ports[lm->port]) == NULL) {
3434                         mtx_unlock(&softc->ctl_lock);
3435                         return (ENXIO);
3436                 }
3437                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3438                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3439                                 if (ctl_lun_map_to_port(port, lun->lun) ==
3440                                     UINT32_MAX)
3441                                         continue;
3442                                 mtx_lock(&lun->lun_lock);
3443                                 ctl_est_ua_port(lun, lm->port, -1,
3444                                     CTL_UA_LUN_CHANGE);
3445                                 mtx_unlock(&lun->lun_lock);
3446                         }
3447                 }
3448                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3449                 if (lm->plun != UINT32_MAX) {
3450                         if (lm->lun == UINT32_MAX)
3451                                 retval = ctl_lun_map_unset(port, lm->plun);
3452                         else if (lm->lun < ctl_max_luns &&
3453                             softc->ctl_luns[lm->lun] != NULL)
3454                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3455                         else
3456                                 return (ENXIO);
3457                 } else {
3458                         if (lm->lun == UINT32_MAX)
3459                                 retval = ctl_lun_map_deinit(port);
3460                         else
3461                                 retval = ctl_lun_map_init(port);
3462                 }
3463                 if (port->status & CTL_PORT_STATUS_ONLINE)
3464                         ctl_isc_announce_port(port);
3465                 break;
3466         }
3467         case CTL_GET_LUN_STATS: {
3468                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3469                 int i;
3470
3471                 /*
3472                  * XXX KDM no locking here.  If the LUN list changes,
3473                  * things can blow up.
3474                  */
3475                 i = 0;
3476                 stats->status = CTL_SS_OK;
3477                 stats->fill_len = 0;
3478                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3479                         if (lun->lun < stats->first_item)
3480                                 continue;
3481                         if (stats->fill_len + sizeof(lun->stats) >
3482                             stats->alloc_len) {
3483                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3484                                 break;
3485                         }
3486                         retval = copyout(&lun->stats, &stats->stats[i++],
3487                                          sizeof(lun->stats));
3488                         if (retval != 0)
3489                                 break;
3490                         stats->fill_len += sizeof(lun->stats);
3491                 }
3492                 stats->num_items = softc->num_luns;
3493                 stats->flags = CTL_STATS_FLAG_NONE;
3494 #ifdef CTL_TIME_IO
3495                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3496 #endif
3497                 getnanouptime(&stats->timestamp);
3498                 break;
3499         }
3500         case CTL_GET_PORT_STATS: {
3501                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3502                 int i;
3503
3504                 /*
3505                  * XXX KDM no locking here.  If the LUN list changes,
3506                  * things can blow up.
3507                  */
3508                 i = 0;
3509                 stats->status = CTL_SS_OK;
3510                 stats->fill_len = 0;
3511                 STAILQ_FOREACH(port, &softc->port_list, links) {
3512                         if (port->targ_port < stats->first_item)
3513                                 continue;
3514                         if (stats->fill_len + sizeof(port->stats) >
3515                             stats->alloc_len) {
3516                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3517                                 break;
3518                         }
3519                         retval = copyout(&port->stats, &stats->stats[i++],
3520                                          sizeof(port->stats));
3521                         if (retval != 0)
3522                                 break;
3523                         stats->fill_len += sizeof(port->stats);
3524                 }
3525                 stats->num_items = softc->num_ports;
3526                 stats->flags = CTL_STATS_FLAG_NONE;
3527 #ifdef CTL_TIME_IO
3528                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3529 #endif
3530                 getnanouptime(&stats->timestamp);
3531                 break;
3532         }
3533         default: {
3534                 /* XXX KDM should we fix this? */
3535 #if 0
3536                 struct ctl_backend_driver *backend;
3537                 unsigned int type;
3538                 int found;
3539
3540                 found = 0;
3541
3542                 /*
3543                  * We encode the backend type as the ioctl type for backend
3544                  * ioctls.  So parse it out here, and then search for a
3545                  * backend of this type.
3546                  */
3547                 type = _IOC_TYPE(cmd);
3548
3549                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3550                         if (backend->type == type) {
3551                                 found = 1;
3552                                 break;
3553                         }
3554                 }
3555                 if (found == 0) {
3556                         printf("ctl: unknown ioctl command %#lx or backend "
3557                                "%d\n", cmd, type);
3558                         retval = EINVAL;
3559                         break;
3560                 }
3561                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3562 #endif
3563                 retval = ENOTTY;
3564                 break;
3565         }
3566         }
3567         return (retval);
3568 }
3569
3570 uint32_t
3571 ctl_get_initindex(struct ctl_nexus *nexus)
3572 {
3573         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3574 }
3575
3576 int
3577 ctl_lun_map_init(struct ctl_port *port)
3578 {
3579         struct ctl_softc *softc = port->ctl_softc;
3580         struct ctl_lun *lun;
3581         int size = ctl_lun_map_size;
3582         uint32_t i;
3583
3584         if (port->lun_map == NULL || port->lun_map_size < size) {
3585                 port->lun_map_size = 0;
3586                 free(port->lun_map, M_CTL);
3587                 port->lun_map = malloc(size * sizeof(uint32_t),
3588                     M_CTL, M_NOWAIT);
3589         }
3590         if (port->lun_map == NULL)
3591                 return (ENOMEM);
3592         for (i = 0; i < size; i++)
3593                 port->lun_map[i] = UINT32_MAX;
3594         port->lun_map_size = size;
3595         if (port->status & CTL_PORT_STATUS_ONLINE) {
3596                 if (port->lun_disable != NULL) {
3597                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3598                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3599                 }
3600                 ctl_isc_announce_port(port);
3601         }
3602         return (0);
3603 }
3604
3605 int
3606 ctl_lun_map_deinit(struct ctl_port *port)
3607 {
3608         struct ctl_softc *softc = port->ctl_softc;
3609         struct ctl_lun *lun;
3610
3611         if (port->lun_map == NULL)
3612                 return (0);
3613         port->lun_map_size = 0;
3614         free(port->lun_map, M_CTL);
3615         port->lun_map = NULL;
3616         if (port->status & CTL_PORT_STATUS_ONLINE) {
3617                 if (port->lun_enable != NULL) {
3618                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3619                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3620                 }
3621                 ctl_isc_announce_port(port);
3622         }
3623         return (0);
3624 }
3625
3626 int
3627 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3628 {
3629         int status;
3630         uint32_t old;
3631
3632         if (port->lun_map == NULL) {
3633                 status = ctl_lun_map_init(port);
3634                 if (status != 0)
3635                         return (status);
3636         }
3637         if (plun >= port->lun_map_size)
3638                 return (EINVAL);
3639         old = port->lun_map[plun];
3640         port->lun_map[plun] = glun;
3641         if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) {
3642                 if (port->lun_enable != NULL)
3643                         port->lun_enable(port->targ_lun_arg, plun);
3644                 ctl_isc_announce_port(port);
3645         }
3646         return (0);
3647 }
3648
3649 int
3650 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3651 {
3652         uint32_t old;
3653
3654         if (port->lun_map == NULL || plun >= port->lun_map_size)
3655                 return (0);
3656         old = port->lun_map[plun];
3657         port->lun_map[plun] = UINT32_MAX;
3658         if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) {
3659                 if (port->lun_disable != NULL)
3660                         port->lun_disable(port->targ_lun_arg, plun);
3661                 ctl_isc_announce_port(port);
3662         }
3663         return (0);
3664 }
3665
3666 uint32_t
3667 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3668 {
3669
3670         if (port == NULL)
3671                 return (UINT32_MAX);
3672         if (port->lun_map == NULL)
3673                 return (lun_id);
3674         if (lun_id > port->lun_map_size)
3675                 return (UINT32_MAX);
3676         return (port->lun_map[lun_id]);
3677 }
3678
3679 uint32_t
3680 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3681 {
3682         uint32_t i;
3683
3684         if (port == NULL)
3685                 return (UINT32_MAX);
3686         if (port->lun_map == NULL)
3687                 return (lun_id);
3688         for (i = 0; i < port->lun_map_size; i++) {
3689                 if (port->lun_map[i] == lun_id)
3690                         return (i);
3691         }
3692         return (UINT32_MAX);
3693 }
3694
3695 uint32_t
3696 ctl_decode_lun(uint64_t encoded)
3697 {
3698         uint8_t lun[8];
3699         uint32_t result = 0xffffffff;
3700
3701         be64enc(lun, encoded);
3702         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3703         case RPL_LUNDATA_ATYP_PERIPH:
3704                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3705                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3706                         result = lun[1];
3707                 break;
3708         case RPL_LUNDATA_ATYP_FLAT:
3709                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3710                     lun[6] == 0 && lun[7] == 0)
3711                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3712                 break;
3713         case RPL_LUNDATA_ATYP_EXTLUN:
3714                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3715                 case 0x02:
3716                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3717                         case 0x00:
3718                                 result = lun[1];
3719                                 break;
3720                         case 0x10:
3721                                 result = (lun[1] << 16) + (lun[2] << 8) +
3722                                     lun[3];
3723                                 break;
3724                         case 0x20:
3725                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3726                                         result = (lun[2] << 24) +
3727                                             (lun[3] << 16) + (lun[4] << 8) +
3728                                             lun[5];
3729                                 break;
3730                         }
3731                         break;
3732                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3733                         result = 0xffffffff;
3734                         break;
3735                 }
3736                 break;
3737         }
3738         return (result);
3739 }
3740
3741 uint64_t
3742 ctl_encode_lun(uint32_t decoded)
3743 {
3744         uint64_t l = decoded;
3745
3746         if (l <= 0xff)
3747                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3748         if (l <= 0x3fff)
3749                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3750         if (l <= 0xffffff)
3751                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3752                     (l << 32));
3753         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3754 }
3755
3756 int
3757 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3758 {
3759         int i;
3760
3761         for (i = first; i < last; i++) {
3762                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3763                         return (i);
3764         }
3765         return (-1);
3766 }
3767
3768 int
3769 ctl_set_mask(uint32_t *mask, uint32_t bit)
3770 {
3771         uint32_t chunk, piece;
3772
3773         chunk = bit >> 5;
3774         piece = bit % (sizeof(uint32_t) * 8);
3775
3776         if ((mask[chunk] & (1 << piece)) != 0)
3777                 return (-1);
3778         else
3779                 mask[chunk] |= (1 << piece);
3780
3781         return (0);
3782 }
3783
3784 int
3785 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3786 {
3787         uint32_t chunk, piece;
3788
3789         chunk = bit >> 5;
3790         piece = bit % (sizeof(uint32_t) * 8);
3791
3792         if ((mask[chunk] & (1 << piece)) == 0)
3793                 return (-1);
3794         else
3795                 mask[chunk] &= ~(1 << piece);
3796
3797         return (0);
3798 }
3799
3800 int
3801 ctl_is_set(uint32_t *mask, uint32_t bit)
3802 {
3803         uint32_t chunk, piece;
3804
3805         chunk = bit >> 5;
3806         piece = bit % (sizeof(uint32_t) * 8);
3807
3808         if ((mask[chunk] & (1 << piece)) == 0)
3809                 return (0);
3810         else
3811                 return (1);
3812 }
3813
3814 static uint64_t
3815 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3816 {
3817         uint64_t *t;
3818
3819         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3820         if (t == NULL)
3821                 return (0);
3822         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3823 }
3824
3825 static void
3826 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3827 {
3828         uint64_t *t;
3829
3830         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3831         if (t == NULL)
3832                 return;
3833         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3834 }
3835
3836 static void
3837 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3838 {
3839         uint64_t *p;
3840         u_int i;
3841
3842         i = residx/CTL_MAX_INIT_PER_PORT;
3843         if (lun->pr_keys[i] != NULL)
3844                 return;
3845         mtx_unlock(&lun->lun_lock);
3846         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3847             M_WAITOK | M_ZERO);
3848         mtx_lock(&lun->lun_lock);
3849         if (lun->pr_keys[i] == NULL)
3850                 lun->pr_keys[i] = p;
3851         else
3852                 free(p, M_CTL);
3853 }
3854
3855 static void
3856 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3857 {
3858         uint64_t *t;
3859
3860         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3861         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3862         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3863 }
3864
3865 /*
3866  * ctl_softc, pool_name, total_ctl_io are passed in.
3867  * npool is passed out.
3868  */
3869 int
3870 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3871                 uint32_t total_ctl_io, void **npool)
3872 {
3873         struct ctl_io_pool *pool;
3874
3875         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3876                                             M_NOWAIT | M_ZERO);
3877         if (pool == NULL)
3878                 return (ENOMEM);
3879
3880         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3881         pool->ctl_softc = ctl_softc;
3882 #ifdef IO_POOLS
3883         pool->zone = uma_zsecond_create(pool->name, NULL,
3884             NULL, NULL, NULL, ctl_softc->io_zone);
3885         /* uma_prealloc(pool->zone, total_ctl_io); */
3886 #else
3887         pool->zone = ctl_softc->io_zone;
3888 #endif
3889
3890         *npool = pool;
3891         return (0);
3892 }
3893
3894 void
3895 ctl_pool_free(struct ctl_io_pool *pool)
3896 {
3897
3898         if (pool == NULL)
3899                 return;
3900
3901 #ifdef IO_POOLS
3902         uma_zdestroy(pool->zone);
3903 #endif
3904         free(pool, M_CTL);
3905 }
3906
3907 union ctl_io *
3908 ctl_alloc_io(void *pool_ref)
3909 {
3910         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3911         union ctl_io *io;
3912
3913         io = uma_zalloc(pool->zone, M_WAITOK);
3914         if (io != NULL) {
3915                 io->io_hdr.pool = pool_ref;
3916                 CTL_SOFTC(io) = pool->ctl_softc;
3917                 TAILQ_INIT(&io->io_hdr.blocked_queue);
3918         }
3919         return (io);
3920 }
3921
3922 union ctl_io *
3923 ctl_alloc_io_nowait(void *pool_ref)
3924 {
3925         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3926         union ctl_io *io;
3927
3928         io = uma_zalloc(pool->zone, M_NOWAIT);
3929         if (io != NULL) {
3930                 io->io_hdr.pool = pool_ref;
3931                 CTL_SOFTC(io) = pool->ctl_softc;
3932                 TAILQ_INIT(&io->io_hdr.blocked_queue);
3933         }
3934         return (io);
3935 }
3936
3937 void
3938 ctl_free_io(union ctl_io *io)
3939 {
3940         struct ctl_io_pool *pool;
3941
3942         if (io == NULL)
3943                 return;
3944
3945         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3946         uma_zfree(pool->zone, io);
3947 }
3948
3949 void
3950 ctl_zero_io(union ctl_io *io)
3951 {
3952         struct ctl_io_pool *pool;
3953
3954         if (io == NULL)
3955                 return;
3956
3957         /*
3958          * May need to preserve linked list pointers at some point too.
3959          */
3960         pool = io->io_hdr.pool;
3961         memset(io, 0, sizeof(*io));
3962         io->io_hdr.pool = pool;
3963         CTL_SOFTC(io) = pool->ctl_softc;
3964         TAILQ_INIT(&io->io_hdr.blocked_queue);
3965 }
3966
3967 int
3968 ctl_expand_number(const char *buf, uint64_t *num)
3969 {
3970         char *endptr;
3971         uint64_t number;
3972         unsigned shift;
3973
3974         number = strtoq(buf, &endptr, 0);
3975
3976         switch (tolower((unsigned char)*endptr)) {
3977         case 'e':
3978                 shift = 60;
3979                 break;
3980         case 'p':
3981                 shift = 50;
3982                 break;
3983         case 't':
3984                 shift = 40;
3985                 break;
3986         case 'g':
3987                 shift = 30;
3988                 break;
3989         case 'm':
3990                 shift = 20;
3991                 break;
3992         case 'k':
3993                 shift = 10;
3994                 break;
3995         case 'b':
3996         case '\0': /* No unit. */
3997                 *num = number;
3998                 return (0);
3999         default:
4000                 /* Unrecognized unit. */
4001                 return (-1);
4002         }
4003
4004         if ((number << shift) >> shift != number) {
4005                 /* Overflow */
4006                 return (-1);
4007         }
4008         *num = number << shift;
4009         return (0);
4010 }
4011
4012 /*
4013  * This routine could be used in the future to load default and/or saved
4014  * mode page parameters for a particuar lun.
4015  */
4016 static int
4017 ctl_init_page_index(struct ctl_lun *lun)
4018 {
4019         int i, page_code;
4020         struct ctl_page_index *page_index;
4021         const char *value;
4022         uint64_t ival;
4023
4024         memcpy(&lun->mode_pages.index, page_index_template,
4025                sizeof(page_index_template));
4026
4027         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4028                 page_index = &lun->mode_pages.index[i];
4029                 if (lun->be_lun->lun_type == T_DIRECT &&
4030                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4031                         continue;
4032                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4033                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4034                         continue;
4035                 if (lun->be_lun->lun_type == T_CDROM &&
4036                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4037                         continue;
4038
4039                 page_code = page_index->page_code & SMPH_PC_MASK;
4040                 switch (page_code) {
4041                 case SMS_RW_ERROR_RECOVERY_PAGE: {
4042                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4043                             ("subpage %#x for page %#x is incorrect!",
4044                             page_index->subpage, page_code));
4045                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4046                                &rw_er_page_default,
4047                                sizeof(rw_er_page_default));
4048                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4049                                &rw_er_page_changeable,
4050                                sizeof(rw_er_page_changeable));
4051                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4052                                &rw_er_page_default,
4053                                sizeof(rw_er_page_default));
4054                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4055                                &rw_er_page_default,
4056                                sizeof(rw_er_page_default));
4057                         page_index->page_data =
4058                                 (uint8_t *)lun->mode_pages.rw_er_page;
4059                         break;
4060                 }
4061                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4062                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4063                             ("subpage %#x for page %#x is incorrect!",
4064                             page_index->subpage, page_code));
4065                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4066                                &verify_er_page_default,
4067                                sizeof(verify_er_page_default));
4068                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4069                                &verify_er_page_changeable,
4070                                sizeof(verify_er_page_changeable));
4071                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4072                                &verify_er_page_default,
4073                                sizeof(verify_er_page_default));
4074                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4075                                &verify_er_page_default,
4076                                sizeof(verify_er_page_default));
4077                         page_index->page_data =
4078                                 (uint8_t *)lun->mode_pages.verify_er_page;
4079                         break;
4080                 }
4081                 case SMS_CACHING_PAGE: {
4082                         struct scsi_caching_page *caching_page;
4083
4084                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4085                             ("subpage %#x for page %#x is incorrect!",
4086                             page_index->subpage, page_code));
4087                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4088                                &caching_page_default,
4089                                sizeof(caching_page_default));
4090                         memcpy(&lun->mode_pages.caching_page[
4091                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4092                                sizeof(caching_page_changeable));
4093                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4094                                &caching_page_default,
4095                                sizeof(caching_page_default));
4096                         caching_page = &lun->mode_pages.caching_page[
4097                             CTL_PAGE_SAVED];
4098                         value = dnvlist_get_string(lun->be_lun->options,
4099                             "writecache", NULL);
4100                         if (value != NULL && strcmp(value, "off") == 0)
4101                                 caching_page->flags1 &= ~SCP_WCE;
4102                         value = dnvlist_get_string(lun->be_lun->options,
4103                             "readcache", NULL);
4104                         if (value != NULL && strcmp(value, "off") == 0)
4105                                 caching_page->flags1 |= SCP_RCD;
4106                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4107                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4108                                sizeof(caching_page_default));
4109                         page_index->page_data =
4110                                 (uint8_t *)lun->mode_pages.caching_page;
4111                         break;
4112                 }
4113                 case SMS_CONTROL_MODE_PAGE: {
4114                         switch (page_index->subpage) {
4115                         case SMS_SUBPAGE_PAGE_0: {
4116                                 struct scsi_control_page *control_page;
4117
4118                                 memcpy(&lun->mode_pages.control_page[
4119                                     CTL_PAGE_DEFAULT],
4120                                        &control_page_default,
4121                                        sizeof(control_page_default));
4122                                 memcpy(&lun->mode_pages.control_page[
4123                                     CTL_PAGE_CHANGEABLE],
4124                                        &control_page_changeable,
4125                                        sizeof(control_page_changeable));
4126                                 memcpy(&lun->mode_pages.control_page[
4127                                     CTL_PAGE_SAVED],
4128                                        &control_page_default,
4129                                        sizeof(control_page_default));
4130                                 control_page = &lun->mode_pages.control_page[
4131                                     CTL_PAGE_SAVED];
4132                                 value = dnvlist_get_string(lun->be_lun->options,
4133                                     "reordering", NULL);
4134                                 if (value != NULL &&
4135                                     strcmp(value, "unrestricted") == 0) {
4136                                         control_page->queue_flags &=
4137                                             ~SCP_QUEUE_ALG_MASK;
4138                                         control_page->queue_flags |=
4139                                             SCP_QUEUE_ALG_UNRESTRICTED;
4140                                 }
4141                                 memcpy(&lun->mode_pages.control_page[
4142                                     CTL_PAGE_CURRENT],
4143                                        &lun->mode_pages.control_page[
4144                                     CTL_PAGE_SAVED],
4145                                        sizeof(control_page_default));
4146                                 page_index->page_data =
4147                                     (uint8_t *)lun->mode_pages.control_page;
4148                                 break;
4149                         }
4150                         case 0x01:
4151                                 memcpy(&lun->mode_pages.control_ext_page[
4152                                     CTL_PAGE_DEFAULT],
4153                                        &control_ext_page_default,
4154                                        sizeof(control_ext_page_default));
4155                                 memcpy(&lun->mode_pages.control_ext_page[
4156                                     CTL_PAGE_CHANGEABLE],
4157                                        &control_ext_page_changeable,
4158                                        sizeof(control_ext_page_changeable));
4159                                 memcpy(&lun->mode_pages.control_ext_page[
4160                                     CTL_PAGE_SAVED],
4161                                        &control_ext_page_default,
4162                                        sizeof(control_ext_page_default));
4163                                 memcpy(&lun->mode_pages.control_ext_page[
4164                                     CTL_PAGE_CURRENT],
4165                                        &lun->mode_pages.control_ext_page[
4166                                     CTL_PAGE_SAVED],
4167                                        sizeof(control_ext_page_default));
4168                                 page_index->page_data =
4169                                     (uint8_t *)lun->mode_pages.control_ext_page;
4170                                 break;
4171                         default:
4172                                 panic("subpage %#x for page %#x is incorrect!",
4173                                       page_index->subpage, page_code);
4174                         }
4175                         break;
4176                 }
4177                 case SMS_INFO_EXCEPTIONS_PAGE: {
4178                         switch (page_index->subpage) {
4179                         case SMS_SUBPAGE_PAGE_0:
4180                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4181                                        &ie_page_default,
4182                                        sizeof(ie_page_default));
4183                                 memcpy(&lun->mode_pages.ie_page[
4184                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4185                                        sizeof(ie_page_changeable));
4186                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4187                                        &ie_page_default,
4188                                        sizeof(ie_page_default));
4189                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4190                                        &ie_page_default,
4191                                        sizeof(ie_page_default));
4192                                 page_index->page_data =
4193                                         (uint8_t *)lun->mode_pages.ie_page;
4194                                 break;
4195                         case 0x02: {
4196                                 struct ctl_logical_block_provisioning_page *page;
4197
4198                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4199                                        &lbp_page_default,
4200                                        sizeof(lbp_page_default));
4201                                 memcpy(&lun->mode_pages.lbp_page[
4202                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4203                                        sizeof(lbp_page_changeable));
4204                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4205                                        &lbp_page_default,
4206                                        sizeof(lbp_page_default));
4207                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4208                                 value = dnvlist_get_string(lun->be_lun->options,
4209                                     "avail-threshold", NULL);
4210                                 if (value != NULL &&
4211                                     ctl_expand_number(value, &ival) == 0) {
4212                                         page->descr[0].flags |= SLBPPD_ENABLED |
4213                                             SLBPPD_ARMING_DEC;
4214                                         if (lun->be_lun->blocksize)
4215                                                 ival /= lun->be_lun->blocksize;
4216                                         else
4217                                                 ival /= 512;
4218                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4219                                             page->descr[0].count);
4220                                 }
4221                                 value = dnvlist_get_string(lun->be_lun->options,
4222                                     "used-threshold", NULL);
4223                                 if (value != NULL &&
4224                                     ctl_expand_number(value, &ival) == 0) {
4225                                         page->descr[1].flags |= SLBPPD_ENABLED |
4226                                             SLBPPD_ARMING_INC;
4227                                         if (lun->be_lun->blocksize)
4228                                                 ival /= lun->be_lun->blocksize;
4229                                         else
4230                                                 ival /= 512;
4231                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4232                                             page->descr[1].count);
4233                                 }
4234                                 value = dnvlist_get_string(lun->be_lun->options,
4235                                     "pool-avail-threshold", NULL);
4236                                 if (value != NULL &&
4237                                     ctl_expand_number(value, &ival) == 0) {
4238                                         page->descr[2].flags |= SLBPPD_ENABLED |
4239                                             SLBPPD_ARMING_DEC;
4240                                         if (lun->be_lun->blocksize)
4241                                                 ival /= lun->be_lun->blocksize;
4242                                         else
4243                                                 ival /= 512;
4244                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4245                                             page->descr[2].count);
4246                                 }
4247                                 value = dnvlist_get_string(lun->be_lun->options,
4248                                     "pool-used-threshold", NULL);
4249                                 if (value != NULL &&
4250                                     ctl_expand_number(value, &ival) == 0) {
4251                                         page->descr[3].flags |= SLBPPD_ENABLED |
4252                                             SLBPPD_ARMING_INC;
4253                                         if (lun->be_lun->blocksize)
4254                                                 ival /= lun->be_lun->blocksize;
4255                                         else
4256                                                 ival /= 512;
4257                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4258                                             page->descr[3].count);
4259                                 }
4260                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4261                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4262                                        sizeof(lbp_page_default));
4263                                 page_index->page_data =
4264                                         (uint8_t *)lun->mode_pages.lbp_page;
4265                                 break;
4266                         }
4267                         default:
4268                                 panic("subpage %#x for page %#x is incorrect!",
4269                                       page_index->subpage, page_code);
4270                         }
4271                         break;
4272                 }
4273                 case SMS_CDDVD_CAPS_PAGE:{
4274                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4275                             ("subpage %#x for page %#x is incorrect!",
4276                             page_index->subpage, page_code));
4277                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4278                                &cddvd_page_default,
4279                                sizeof(cddvd_page_default));
4280                         memcpy(&lun->mode_pages.cddvd_page[
4281                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4282                                sizeof(cddvd_page_changeable));
4283                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4284                                &cddvd_page_default,
4285                                sizeof(cddvd_page_default));
4286                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4287                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4288                                sizeof(cddvd_page_default));
4289                         page_index->page_data =
4290                                 (uint8_t *)lun->mode_pages.cddvd_page;
4291                         break;
4292                 }
4293                 default:
4294                         panic("invalid page code value %#x", page_code);
4295                 }
4296         }
4297
4298         return (CTL_RETVAL_COMPLETE);
4299 }
4300
4301 static int
4302 ctl_init_log_page_index(struct ctl_lun *lun)
4303 {
4304         struct ctl_page_index *page_index;
4305         int i, j, k, prev;
4306
4307         memcpy(&lun->log_pages.index, log_page_index_template,
4308                sizeof(log_page_index_template));
4309
4310         prev = -1;
4311         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4312                 page_index = &lun->log_pages.index[i];
4313                 if (lun->be_lun->lun_type == T_DIRECT &&
4314                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4315                         continue;
4316                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4317                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4318                         continue;
4319                 if (lun->be_lun->lun_type == T_CDROM &&
4320                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4321                         continue;
4322
4323                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4324                     lun->backend->lun_attr == NULL)
4325                         continue;
4326
4327                 if (page_index->page_code != prev) {
4328                         lun->log_pages.pages_page[j] = page_index->page_code;
4329                         prev = page_index->page_code;
4330                         j++;
4331                 }
4332                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4333                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4334                 k++;
4335         }
4336         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4337         lun->log_pages.index[0].page_len = j;
4338         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4339         lun->log_pages.index[1].page_len = k * 2;
4340         lun->log_pages.index[2].page_data = (uint8_t *)&lun->log_pages.temp_page;
4341         lun->log_pages.index[2].page_len = sizeof(lun->log_pages.temp_page);
4342         lun->log_pages.index[3].page_data = &lun->log_pages.lbp_page[0];
4343         lun->log_pages.index[3].page_len = 12*CTL_NUM_LBP_PARAMS;
4344         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.stat_page;
4345         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.stat_page);
4346         lun->log_pages.index[5].page_data = (uint8_t *)&lun->log_pages.ie_page;
4347         lun->log_pages.index[5].page_len = sizeof(lun->log_pages.ie_page);
4348
4349         return (CTL_RETVAL_COMPLETE);
4350 }
4351
4352 static int
4353 hex2bin(const char *str, uint8_t *buf, int buf_size)
4354 {
4355         int i;
4356         u_char c;
4357
4358         memset(buf, 0, buf_size);
4359         while (isspace(str[0]))
4360                 str++;
4361         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4362                 str += 2;
4363         buf_size *= 2;
4364         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4365                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4366                         str++;
4367                 c = str[i];
4368                 if (isdigit(c))
4369                         c -= '0';
4370                 else if (isalpha(c))
4371                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4372                 else
4373                         break;
4374                 if (c >= 16)
4375                         break;
4376                 if ((i & 1) == 0)
4377                         buf[i / 2] |= (c << 4);
4378                 else
4379                         buf[i / 2] |= c;
4380         }
4381         return ((i + 1) / 2);
4382 }
4383
4384 /*
4385  * Add LUN.
4386  *
4387  * Returns 0 for success, non-zero (errno) for failure.
4388  */
4389 int
4390 ctl_add_lun(struct ctl_be_lun *be_lun)
4391 {
4392         struct ctl_softc *ctl_softc = control_softc;
4393         struct ctl_lun *nlun, *lun;
4394         struct scsi_vpd_id_descriptor *desc;
4395         struct scsi_vpd_id_t10 *t10id;
4396         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4397         int lun_number;
4398         int devidlen, idlen1, idlen2 = 0, len;
4399
4400         /*
4401          * We support only Direct Access, CD-ROM or Processor LUN types.
4402          */
4403         switch (be_lun->lun_type) {
4404         case T_DIRECT:
4405         case T_PROCESSOR:
4406         case T_CDROM:
4407                 break;
4408         case T_SEQUENTIAL:
4409         case T_CHANGER:
4410         default:
4411                 return (EINVAL);
4412         }
4413         lun = malloc(sizeof(*lun), M_CTL, M_WAITOK | M_ZERO);
4414
4415         lun->pending_sense = malloc(sizeof(struct scsi_sense_data *) *
4416             ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO);
4417         lun->pending_ua = malloc(sizeof(ctl_ua_type *) * ctl_max_ports,
4418             M_DEVBUF, M_WAITOK | M_ZERO);
4419         lun->pr_keys = malloc(sizeof(uint64_t *) * ctl_max_ports,
4420             M_DEVBUF, M_WAITOK | M_ZERO);
4421
4422         /* Generate LUN ID. */
4423         devidlen = max(CTL_DEVID_MIN_LEN,
4424             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4425         idlen1 = sizeof(*t10id) + devidlen;
4426         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4427         scsiname = dnvlist_get_string(be_lun->options, "scsiname", NULL);
4428         if (scsiname != NULL) {
4429                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4430                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4431         }
4432         eui = dnvlist_get_string(be_lun->options, "eui", NULL);
4433         if (eui != NULL) {
4434                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4435         }
4436         naa = dnvlist_get_string(be_lun->options, "naa", NULL);
4437         if (naa != NULL) {
4438                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4439         }
4440         uuid = dnvlist_get_string(be_lun->options, "uuid", NULL);
4441         if (uuid != NULL) {
4442                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4443         }
4444         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4445             M_CTL, M_WAITOK | M_ZERO);
4446         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4447         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4448         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4449         desc->length = idlen1;
4450         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4451         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4452         if ((vendor = dnvlist_get_string(be_lun->options, "vendor", NULL)) == NULL) {
4453                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4454         } else {
4455                 strncpy(t10id->vendor, vendor,
4456                     min(sizeof(t10id->vendor), strlen(vendor)));
4457         }
4458         strncpy((char *)t10id->vendor_spec_id,
4459             (char *)be_lun->device_id, devidlen);
4460         if (scsiname != NULL) {
4461                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4462                     desc->length);
4463                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4464                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4465                     SVPD_ID_TYPE_SCSI_NAME;
4466                 desc->length = idlen2;
4467                 strlcpy(desc->identifier, scsiname, idlen2);
4468         }
4469         if (eui != NULL) {
4470                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4471                     desc->length);
4472                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4473                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4474                     SVPD_ID_TYPE_EUI64;
4475                 desc->length = hex2bin(eui, desc->identifier, 16);
4476                 desc->length = desc->length > 12 ? 16 :
4477                     (desc->length > 8 ? 12 : 8);
4478                 len -= 16 - desc->length;
4479         }
4480         if (naa != NULL) {
4481                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4482                     desc->length);
4483                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4484                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4485                     SVPD_ID_TYPE_NAA;
4486                 desc->length = hex2bin(naa, desc->identifier, 16);
4487                 desc->length = desc->length > 8 ? 16 : 8;
4488                 len -= 16 - desc->length;
4489         }
4490         if (uuid != NULL) {
4491                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4492                     desc->length);
4493                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4494                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4495                     SVPD_ID_TYPE_UUID;
4496                 desc->identifier[0] = 0x10;
4497                 hex2bin(uuid, &desc->identifier[2], 16);
4498                 desc->length = 18;
4499         }
4500         lun->lun_devid->len = len;
4501
4502         mtx_lock(&ctl_softc->ctl_lock);
4503         /*
4504          * See if the caller requested a particular LUN number.  If so, see
4505          * if it is available.  Otherwise, allocate the first available LUN.
4506          */
4507         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4508                 if ((be_lun->req_lun_id > (ctl_max_luns - 1))
4509                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4510                         mtx_unlock(&ctl_softc->ctl_lock);
4511                         if (be_lun->req_lun_id > (ctl_max_luns - 1)) {
4512                                 printf("ctl: requested LUN ID %d is higher "
4513                                        "than ctl_max_luns - 1 (%d)\n",
4514                                        be_lun->req_lun_id, ctl_max_luns - 1);
4515                         } else {
4516                                 /*
4517                                  * XXX KDM return an error, or just assign
4518                                  * another LUN ID in this case??
4519                                  */
4520                                 printf("ctl: requested LUN ID %d is already "
4521                                        "in use\n", be_lun->req_lun_id);
4522                         }
4523 fail:
4524                         free(lun->lun_devid, M_CTL);
4525                         free(lun, M_CTL);
4526                         return (ENOSPC);
4527                 }
4528                 lun_number = be_lun->req_lun_id;
4529         } else {
4530                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, ctl_max_luns);
4531                 if (lun_number == -1) {
4532                         mtx_unlock(&ctl_softc->ctl_lock);
4533                         printf("ctl: can't allocate LUN, out of LUNs\n");
4534                         goto fail;
4535                 }
4536         }
4537         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4538         mtx_unlock(&ctl_softc->ctl_lock);
4539
4540         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4541         lun->lun = lun_number;
4542         lun->be_lun = be_lun;
4543         /*
4544          * The processor LUN is always enabled.  Disk LUNs come on line
4545          * disabled, and must be enabled by the backend.
4546          */
4547         lun->flags |= CTL_LUN_DISABLED;
4548         lun->backend = be_lun->be;
4549         be_lun->ctl_lun = lun;
4550         be_lun->lun_id = lun_number;
4551         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4552                 lun->flags |= CTL_LUN_EJECTED;
4553         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4554                 lun->flags |= CTL_LUN_NO_MEDIA;
4555         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4556                 lun->flags |= CTL_LUN_STOPPED;
4557
4558         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4559                 lun->flags |= CTL_LUN_PRIMARY_SC;
4560
4561         value = dnvlist_get_string(be_lun->options, "removable", NULL);
4562         if (value != NULL) {
4563                 if (strcmp(value, "on") == 0)
4564                         lun->flags |= CTL_LUN_REMOVABLE;
4565         } else if (be_lun->lun_type == T_CDROM)
4566                 lun->flags |= CTL_LUN_REMOVABLE;
4567
4568         lun->ctl_softc = ctl_softc;
4569 #ifdef CTL_TIME_IO
4570         lun->last_busy = getsbinuptime();
4571 #endif
4572         LIST_INIT(&lun->ooa_queue);
4573         STAILQ_INIT(&lun->error_list);
4574         lun->ie_reported = 1;
4575         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4576         ctl_tpc_lun_init(lun);
4577         if (lun->flags & CTL_LUN_REMOVABLE) {
4578                 lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4,
4579                     M_CTL, M_WAITOK);
4580         }
4581
4582         /*
4583          * Initialize the mode and log page index.
4584          */
4585         ctl_init_page_index(lun);
4586         ctl_init_log_page_index(lun);
4587
4588         /* Setup statistics gathering */
4589         lun->stats.item = lun_number;
4590
4591         /*
4592          * Now, before we insert this lun on the lun list, set the lun
4593          * inventory changed UA for all other luns.
4594          */
4595         mtx_lock(&ctl_softc->ctl_lock);
4596         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4597                 mtx_lock(&nlun->lun_lock);
4598                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4599                 mtx_unlock(&nlun->lun_lock);
4600         }
4601         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4602         ctl_softc->ctl_luns[lun_number] = lun;
4603         ctl_softc->num_luns++;
4604         mtx_unlock(&ctl_softc->ctl_lock);
4605
4606         /*
4607          * We successfully added the LUN, attempt to enable it.
4608          */
4609         if (ctl_enable_lun(lun) != 0) {
4610                 printf("%s: ctl_enable_lun() failed!\n", __func__);
4611                 mtx_lock(&ctl_softc->ctl_lock);
4612                 STAILQ_REMOVE(&ctl_softc->lun_list, lun, ctl_lun, links);
4613                 ctl_clear_mask(ctl_softc->ctl_lun_mask, lun_number);
4614                 ctl_softc->ctl_luns[lun_number] = NULL;
4615                 ctl_softc->num_luns--;
4616                 mtx_unlock(&ctl_softc->ctl_lock);
4617                 free(lun->lun_devid, M_CTL);
4618                 free(lun, M_CTL);
4619                 return (EIO);
4620         }
4621
4622         return (0);
4623 }
4624
4625 /*
4626  * Free LUN that has no active requests.
4627  */
4628 static int
4629 ctl_free_lun(struct ctl_lun *lun)
4630 {
4631         struct ctl_softc *softc = lun->ctl_softc;
4632         struct ctl_lun *nlun;
4633         int i;
4634
4635         KASSERT(LIST_EMPTY(&lun->ooa_queue),
4636             ("Freeing a LUN %p with outstanding I/O!\n", lun));
4637
4638         mtx_lock(&softc->ctl_lock);
4639         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4640         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4641         softc->ctl_luns[lun->lun] = NULL;
4642         softc->num_luns--;
4643         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4644                 mtx_lock(&nlun->lun_lock);
4645                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4646                 mtx_unlock(&nlun->lun_lock);
4647         }
4648         mtx_unlock(&softc->ctl_lock);
4649
4650         /*
4651          * Tell the backend to free resources, if this LUN has a backend.
4652          */
4653         lun->be_lun->lun_shutdown(lun->be_lun);
4654
4655         lun->ie_reportcnt = UINT32_MAX;
4656         callout_drain(&lun->ie_callout);
4657         ctl_tpc_lun_shutdown(lun);
4658         mtx_destroy(&lun->lun_lock);
4659         free(lun->lun_devid, M_CTL);
4660         for (i = 0; i < ctl_max_ports; i++)
4661                 free(lun->pending_ua[i], M_CTL);
4662         free(lun->pending_ua, M_DEVBUF);
4663         for (i = 0; i < ctl_max_ports; i++)
4664                 free(lun->pr_keys[i], M_CTL);
4665         free(lun->pr_keys, M_DEVBUF);
4666         free(lun->write_buffer, M_CTL);
4667         free(lun->prevent, M_CTL);
4668         free(lun, M_CTL);
4669
4670         return (0);
4671 }
4672
4673 static int
4674 ctl_enable_lun(struct ctl_lun *lun)
4675 {
4676         struct ctl_softc *softc;
4677         struct ctl_port *port, *nport;
4678         int retval;
4679
4680         softc = lun->ctl_softc;
4681
4682         mtx_lock(&softc->ctl_lock);
4683         mtx_lock(&lun->lun_lock);
4684         KASSERT((lun->flags & CTL_LUN_DISABLED) != 0,
4685             ("%s: LUN not disabled", __func__));
4686         lun->flags &= ~CTL_LUN_DISABLED;
4687         mtx_unlock(&lun->lun_lock);
4688
4689         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4690                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4691                     port->lun_map != NULL || port->lun_enable == NULL)
4692                         continue;
4693
4694                 /*
4695                  * Drop the lock while we call the FETD's enable routine.
4696                  * This can lead to a callback into CTL (at least in the
4697                  * case of the internal initiator frontend.
4698                  */
4699                 mtx_unlock(&softc->ctl_lock);
4700                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4701                 mtx_lock(&softc->ctl_lock);
4702                 if (retval != 0) {
4703                         printf("%s: FETD %s port %d returned error "
4704                                "%d for lun_enable on lun %jd\n",
4705                                __func__, port->port_name, port->targ_port,
4706                                retval, (intmax_t)lun->lun);
4707                 }
4708         }
4709
4710         mtx_unlock(&softc->ctl_lock);
4711         ctl_isc_announce_lun(lun);
4712
4713         return (0);
4714 }
4715
4716 static int
4717 ctl_disable_lun(struct ctl_lun *lun)
4718 {
4719         struct ctl_softc *softc;
4720         struct ctl_port *port;
4721         int retval;
4722
4723         softc = lun->ctl_softc;
4724
4725         mtx_lock(&softc->ctl_lock);
4726         mtx_lock(&lun->lun_lock);
4727         KASSERT((lun->flags & CTL_LUN_DISABLED) == 0,
4728             ("%s: LUN not enabled", __func__));
4729         lun->flags |= CTL_LUN_DISABLED;
4730         mtx_unlock(&lun->lun_lock);
4731
4732         STAILQ_FOREACH(port, &softc->port_list, links) {
4733                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4734                     port->lun_map != NULL || port->lun_disable == NULL)
4735                         continue;
4736
4737                 /*
4738                  * Drop the lock before we call the frontend's disable
4739                  * routine, to avoid lock order reversals.
4740                  *
4741                  * XXX KDM what happens if the frontend list changes while
4742                  * we're traversing it?  It's unlikely, but should be handled.
4743                  */
4744                 mtx_unlock(&softc->ctl_lock);
4745                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4746                 mtx_lock(&softc->ctl_lock);
4747                 if (retval != 0) {
4748                         printf("%s: FETD %s port %d returned error "
4749                                "%d for lun_disable on lun %jd\n",
4750                                __func__, port->port_name, port->targ_port,
4751                                retval, (intmax_t)lun->lun);
4752                 }
4753         }
4754
4755         mtx_unlock(&softc->ctl_lock);
4756         ctl_isc_announce_lun(lun);
4757
4758         return (0);
4759 }
4760
4761 int
4762 ctl_start_lun(struct ctl_be_lun *be_lun)
4763 {
4764         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4765
4766         mtx_lock(&lun->lun_lock);
4767         lun->flags &= ~CTL_LUN_STOPPED;
4768         mtx_unlock(&lun->lun_lock);
4769         return (0);
4770 }
4771
4772 int
4773 ctl_stop_lun(struct ctl_be_lun *be_lun)
4774 {
4775         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4776
4777         mtx_lock(&lun->lun_lock);
4778         lun->flags |= CTL_LUN_STOPPED;
4779         mtx_unlock(&lun->lun_lock);
4780         return (0);
4781 }
4782
4783 int
4784 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4785 {
4786         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4787
4788         mtx_lock(&lun->lun_lock);
4789         lun->flags |= CTL_LUN_NO_MEDIA;
4790         mtx_unlock(&lun->lun_lock);
4791         return (0);
4792 }
4793
4794 int
4795 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4796 {
4797         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4798         union ctl_ha_msg msg;
4799
4800         mtx_lock(&lun->lun_lock);
4801         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4802         if (lun->flags & CTL_LUN_REMOVABLE)
4803                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4804         mtx_unlock(&lun->lun_lock);
4805         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4806             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4807                 bzero(&msg.ua, sizeof(msg.ua));
4808                 msg.hdr.msg_type = CTL_MSG_UA;
4809                 msg.hdr.nexus.initid = -1;
4810                 msg.hdr.nexus.targ_port = -1;
4811                 msg.hdr.nexus.targ_lun = lun->lun;
4812                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4813                 msg.ua.ua_all = 1;
4814                 msg.ua.ua_set = 1;
4815                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4816                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4817                     M_WAITOK);
4818         }
4819         return (0);
4820 }
4821
4822 int
4823 ctl_lun_ejected(struct ctl_be_lun *be_lun)
4824 {
4825         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4826
4827         mtx_lock(&lun->lun_lock);
4828         lun->flags |= CTL_LUN_EJECTED;
4829         mtx_unlock(&lun->lun_lock);
4830         return (0);
4831 }
4832
4833 int
4834 ctl_lun_primary(struct ctl_be_lun *be_lun)
4835 {
4836         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4837
4838         mtx_lock(&lun->lun_lock);
4839         lun->flags |= CTL_LUN_PRIMARY_SC;
4840         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4841         mtx_unlock(&lun->lun_lock);
4842         ctl_isc_announce_lun(lun);
4843         return (0);
4844 }
4845
4846 int
4847 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4848 {
4849         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4850
4851         mtx_lock(&lun->lun_lock);
4852         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4853         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4854         mtx_unlock(&lun->lun_lock);
4855         ctl_isc_announce_lun(lun);
4856         return (0);
4857 }
4858
4859 /*
4860  * Remove LUN.  If there are active requests, wait for completion.
4861  *
4862  * Returns 0 for success, non-zero (errno) for failure.
4863  * Completion is reported to backed via the lun_shutdown() method.
4864  */
4865 int
4866 ctl_remove_lun(struct ctl_be_lun *be_lun)
4867 {
4868         struct ctl_lun *lun;
4869
4870         lun = (struct ctl_lun *)be_lun->ctl_lun;
4871
4872         ctl_disable_lun(lun);
4873
4874         mtx_lock(&lun->lun_lock);
4875         lun->flags |= CTL_LUN_INVALID;
4876
4877         /*
4878          * If there is nothing in the OOA queue, go ahead and free the LUN.
4879          * If we have something in the OOA queue, we'll free it when the
4880          * last I/O completes.
4881          */
4882         if (LIST_EMPTY(&lun->ooa_queue)) {
4883                 mtx_unlock(&lun->lun_lock);
4884                 ctl_free_lun(lun);
4885         } else
4886                 mtx_unlock(&lun->lun_lock);
4887
4888         return (0);
4889 }
4890
4891 void
4892 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4893 {
4894         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4895         union ctl_ha_msg msg;
4896
4897         mtx_lock(&lun->lun_lock);
4898         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
4899         mtx_unlock(&lun->lun_lock);
4900         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4901                 /* Send msg to other side. */
4902                 bzero(&msg.ua, sizeof(msg.ua));
4903                 msg.hdr.msg_type = CTL_MSG_UA;
4904                 msg.hdr.nexus.initid = -1;
4905                 msg.hdr.nexus.targ_port = -1;
4906                 msg.hdr.nexus.targ_lun = lun->lun;
4907                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4908                 msg.ua.ua_all = 1;
4909                 msg.ua.ua_set = 1;
4910                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
4911                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4912                     M_WAITOK);
4913         }
4914 }
4915
4916 /*
4917  * Backend "memory move is complete" callback for requests that never
4918  * make it down to say RAIDCore's configuration code.
4919  */
4920 int
4921 ctl_config_move_done(union ctl_io *io, bool samethr)
4922 {
4923         int retval;
4924
4925         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4926         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4927             ("%s: unexpected I/O type %x", __func__, io->io_hdr.io_type));
4928
4929         if (ctl_debug & CTL_DEBUG_CDB_DATA)
4930                 ctl_data_print(io);
4931         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
4932             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4933              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
4934             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4935                 /*
4936                  * XXX KDM just assuming a single pointer here, and not a
4937                  * S/G list.  If we start using S/G lists for config data,
4938                  * we'll need to know how to clean them up here as well.
4939                  */
4940                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4941                         free(io->scsiio.kern_data_ptr, M_CTL);
4942                 ctl_done(io);
4943                 retval = CTL_RETVAL_COMPLETE;
4944         } else {
4945                 /*
4946                  * XXX KDM now we need to continue data movement.  Some
4947                  * options:
4948                  * - call ctl_scsiio() again?  We don't do this for data
4949                  *   writes, because for those at least we know ahead of
4950                  *   time where the write will go and how long it is.  For
4951                  *   config writes, though, that information is largely
4952                  *   contained within the write itself, thus we need to
4953                  *   parse out the data again.
4954                  *
4955                  * - Call some other function once the data is in?
4956                  */
4957
4958                 /*
4959                  * XXX KDM call ctl_scsiio() again for now, and check flag
4960                  * bits to see whether we're allocated or not.
4961                  */
4962                 retval = ctl_scsiio(&io->scsiio);
4963         }
4964         return (retval);
4965 }
4966
4967 /*
4968  * This gets called by a backend driver when it is done with a
4969  * data_submit method.
4970  */
4971 void
4972 ctl_data_submit_done(union ctl_io *io)
4973 {
4974         /*
4975          * If the IO_CONT flag is set, we need to call the supplied
4976          * function to continue processing the I/O, instead of completing
4977          * the I/O just yet.
4978          *
4979          * If there is an error, though, we don't want to keep processing.
4980          * Instead, just send status back to the initiator.
4981          */
4982         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4983             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4984             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4985              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4986                 io->scsiio.io_cont(io);
4987                 return;
4988         }
4989         ctl_done(io);
4990 }
4991
4992 /*
4993  * This gets called by a backend driver when it is done with a
4994  * configuration write.
4995  */
4996 void
4997 ctl_config_write_done(union ctl_io *io)
4998 {
4999         uint8_t *buf;
5000
5001         /*
5002          * If the IO_CONT flag is set, we need to call the supplied
5003          * function to continue processing the I/O, instead of completing
5004          * the I/O just yet.
5005          *
5006          * If there is an error, though, we don't want to keep processing.
5007          * Instead, just send status back to the initiator.
5008          */
5009         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5010             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5011             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5012              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5013                 io->scsiio.io_cont(io);
5014                 return;
5015         }
5016         /*
5017          * Since a configuration write can be done for commands that actually
5018          * have data allocated, like write buffer, and commands that have
5019          * no data, like start/stop unit, we need to check here.
5020          */
5021         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5022                 buf = io->scsiio.kern_data_ptr;
5023         else
5024                 buf = NULL;
5025         ctl_done(io);
5026         if (buf)
5027                 free(buf, M_CTL);
5028 }
5029
5030 void
5031 ctl_config_read_done(union ctl_io *io)
5032 {
5033         uint8_t *buf;
5034
5035         /*
5036          * If there is some error -- we are done, skip data transfer.
5037          */
5038         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5039             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5040              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5041                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5042                         buf = io->scsiio.kern_data_ptr;
5043                 else
5044                         buf = NULL;
5045                 ctl_done(io);
5046                 if (buf)
5047                         free(buf, M_CTL);
5048                 return;
5049         }
5050
5051         /*
5052          * If the IO_CONT flag is set, we need to call the supplied
5053          * function to continue processing the I/O, instead of completing
5054          * the I/O just yet.
5055          */
5056         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5057                 io->scsiio.io_cont(io);
5058                 return;
5059         }
5060
5061         ctl_datamove(io);
5062 }
5063
5064 /*
5065  * SCSI release command.
5066  */
5067 int
5068 ctl_scsi_release(struct ctl_scsiio *ctsio)
5069 {
5070         struct ctl_lun *lun = CTL_LUN(ctsio);
5071         uint32_t residx;
5072
5073         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5074
5075         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5076
5077         /*
5078          * XXX KDM right now, we only support LUN reservation.  We don't
5079          * support 3rd party reservations, or extent reservations, which
5080          * might actually need the parameter list.  If we've gotten this
5081          * far, we've got a LUN reservation.  Anything else got kicked out
5082          * above.  So, according to SPC, ignore the length.
5083          */
5084
5085         mtx_lock(&lun->lun_lock);
5086
5087         /*
5088          * According to SPC, it is not an error for an intiator to attempt
5089          * to release a reservation on a LUN that isn't reserved, or that
5090          * is reserved by another initiator.  The reservation can only be
5091          * released, though, by the initiator who made it or by one of
5092          * several reset type events.
5093          */
5094         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5095                         lun->flags &= ~CTL_LUN_RESERVED;
5096
5097         mtx_unlock(&lun->lun_lock);
5098
5099         ctl_set_success(ctsio);
5100         ctl_done((union ctl_io *)ctsio);
5101         return (CTL_RETVAL_COMPLETE);
5102 }
5103
5104 int
5105 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5106 {
5107         struct ctl_lun *lun = CTL_LUN(ctsio);
5108         uint32_t residx;
5109
5110         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5111
5112         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5113
5114         /*
5115          * XXX KDM right now, we only support LUN reservation.  We don't
5116          * support 3rd party reservations, or extent reservations, which
5117          * might actually need the parameter list.  If we've gotten this
5118          * far, we've got a LUN reservation.  Anything else got kicked out
5119          * above.  So, according to SPC, ignore the length.
5120          */
5121
5122         mtx_lock(&lun->lun_lock);
5123         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5124                 ctl_set_reservation_conflict(ctsio);
5125                 goto bailout;
5126         }
5127
5128         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5129         if (lun->flags & CTL_LUN_PR_RESERVED) {
5130                 ctl_set_success(ctsio);
5131                 goto bailout;
5132         }
5133
5134         lun->flags |= CTL_LUN_RESERVED;
5135         lun->res_idx = residx;
5136         ctl_set_success(ctsio);
5137
5138 bailout:
5139         mtx_unlock(&lun->lun_lock);
5140         ctl_done((union ctl_io *)ctsio);
5141         return (CTL_RETVAL_COMPLETE);
5142 }
5143
5144 int
5145 ctl_start_stop(struct ctl_scsiio *ctsio)
5146 {
5147         struct ctl_lun *lun = CTL_LUN(ctsio);
5148         struct scsi_start_stop_unit *cdb;
5149         int retval;
5150
5151         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5152
5153         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5154
5155         if ((cdb->how & SSS_PC_MASK) == 0) {
5156                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5157                     (cdb->how & SSS_START) == 0) {
5158                         uint32_t residx;
5159
5160                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5161                         if (ctl_get_prkey(lun, residx) == 0 ||
5162                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5163                                 ctl_set_reservation_conflict(ctsio);
5164                                 ctl_done((union ctl_io *)ctsio);
5165                                 return (CTL_RETVAL_COMPLETE);
5166                         }
5167                 }
5168
5169                 if ((cdb->how & SSS_LOEJ) &&
5170                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5171                         ctl_set_invalid_field(ctsio,
5172                                               /*sks_valid*/ 1,
5173                                               /*command*/ 1,
5174                                               /*field*/ 4,
5175                                               /*bit_valid*/ 1,
5176                                               /*bit*/ 1);
5177                         ctl_done((union ctl_io *)ctsio);
5178                         return (CTL_RETVAL_COMPLETE);
5179                 }
5180
5181                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5182                     lun->prevent_count > 0) {
5183                         /* "Medium removal prevented" */
5184                         ctl_set_sense(ctsio, /*current_error*/ 1,
5185                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5186                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5187                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5188                         ctl_done((union ctl_io *)ctsio);
5189                         return (CTL_RETVAL_COMPLETE);
5190                 }
5191         }
5192
5193         retval = lun->backend->config_write((union ctl_io *)ctsio);
5194         return (retval);
5195 }
5196
5197 int
5198 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5199 {
5200         struct ctl_lun *lun = CTL_LUN(ctsio);
5201         struct scsi_prevent *cdb;
5202         int retval;
5203         uint32_t initidx;
5204
5205         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5206
5207         cdb = (struct scsi_prevent *)ctsio->cdb;
5208
5209         if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) {
5210                 ctl_set_invalid_opcode(ctsio);
5211                 ctl_done((union ctl_io *)ctsio);
5212                 return (CTL_RETVAL_COMPLETE);
5213         }
5214
5215         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5216         mtx_lock(&lun->lun_lock);
5217         if ((cdb->how & PR_PREVENT) &&
5218             ctl_is_set(lun->prevent, initidx) == 0) {
5219                 ctl_set_mask(lun->prevent, initidx);
5220                 lun->prevent_count++;
5221         } else if ((cdb->how & PR_PREVENT) == 0 &&
5222             ctl_is_set(lun->prevent, initidx)) {
5223                 ctl_clear_mask(lun->prevent, initidx);
5224                 lun->prevent_count--;
5225         }
5226         mtx_unlock(&lun->lun_lock);
5227         retval = lun->backend->config_write((union ctl_io *)ctsio);
5228         return (retval);
5229 }
5230
5231 /*
5232  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5233  * we don't really do anything with the LBA and length fields if the user
5234  * passes them in.  Instead we'll just flush out the cache for the entire
5235  * LUN.
5236  */
5237 int
5238 ctl_sync_cache(struct ctl_scsiio *ctsio)
5239 {
5240         struct ctl_lun *lun = CTL_LUN(ctsio);
5241         struct ctl_lba_len_flags *lbalen;
5242         uint64_t starting_lba;
5243         uint32_t block_count;
5244         int retval;
5245         uint8_t byte2;
5246
5247         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5248
5249         retval = 0;
5250
5251         switch (ctsio->cdb[0]) {
5252         case SYNCHRONIZE_CACHE: {
5253                 struct scsi_sync_cache *cdb;
5254                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5255
5256                 starting_lba = scsi_4btoul(cdb->begin_lba);
5257                 block_count = scsi_2btoul(cdb->lb_count);
5258                 byte2 = cdb->byte2;
5259                 break;
5260         }
5261         case SYNCHRONIZE_CACHE_16: {
5262                 struct scsi_sync_cache_16 *cdb;
5263                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5264
5265                 starting_lba = scsi_8btou64(cdb->begin_lba);
5266                 block_count = scsi_4btoul(cdb->lb_count);
5267                 byte2 = cdb->byte2;
5268                 break;
5269         }
5270         default:
5271                 ctl_set_invalid_opcode(ctsio);
5272                 ctl_done((union ctl_io *)ctsio);
5273                 goto bailout;
5274                 break; /* NOTREACHED */
5275         }
5276
5277         /*
5278          * We check the LBA and length, but don't do anything with them.
5279          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5280          * get flushed.  This check will just help satisfy anyone who wants
5281          * to see an error for an out of range LBA.
5282          */
5283         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5284                 ctl_set_lba_out_of_range(ctsio,
5285                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5286                 ctl_done((union ctl_io *)ctsio);
5287                 goto bailout;
5288         }
5289
5290         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5291         lbalen->lba = starting_lba;
5292         lbalen->len = block_count;
5293         lbalen->flags = byte2;
5294         retval = lun->backend->config_write((union ctl_io *)ctsio);
5295
5296 bailout:
5297         return (retval);
5298 }
5299
5300 int
5301 ctl_format(struct ctl_scsiio *ctsio)
5302 {
5303         struct scsi_format *cdb;
5304         int length, defect_list_len;
5305
5306         CTL_DEBUG_PRINT(("ctl_format\n"));
5307
5308         cdb = (struct scsi_format *)ctsio->cdb;
5309
5310         length = 0;
5311         if (cdb->byte2 & SF_FMTDATA) {
5312                 if (cdb->byte2 & SF_LONGLIST)
5313                         length = sizeof(struct scsi_format_header_long);
5314                 else
5315                         length = sizeof(struct scsi_format_header_short);
5316         }
5317
5318         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5319          && (length > 0)) {
5320                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5321                 ctsio->kern_data_len = length;
5322                 ctsio->kern_total_len = length;
5323                 ctsio->kern_rel_offset = 0;
5324                 ctsio->kern_sg_entries = 0;
5325                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5326                 ctsio->be_move_done = ctl_config_move_done;
5327                 ctl_datamove((union ctl_io *)ctsio);
5328
5329                 return (CTL_RETVAL_COMPLETE);
5330         }
5331
5332         defect_list_len = 0;
5333
5334         if (cdb->byte2 & SF_FMTDATA) {
5335                 if (cdb->byte2 & SF_LONGLIST) {
5336                         struct scsi_format_header_long *header;
5337
5338                         header = (struct scsi_format_header_long *)
5339                                 ctsio->kern_data_ptr;
5340
5341                         defect_list_len = scsi_4btoul(header->defect_list_len);
5342                         if (defect_list_len != 0) {
5343                                 ctl_set_invalid_field(ctsio,
5344                                                       /*sks_valid*/ 1,
5345                                                       /*command*/ 0,
5346                                                       /*field*/ 2,
5347                                                       /*bit_valid*/ 0,
5348                                                       /*bit*/ 0);
5349                                 goto bailout;
5350                         }
5351                 } else {
5352                         struct scsi_format_header_short *header;
5353
5354                         header = (struct scsi_format_header_short *)
5355                                 ctsio->kern_data_ptr;
5356
5357                         defect_list_len = scsi_2btoul(header->defect_list_len);
5358                         if (defect_list_len != 0) {
5359                                 ctl_set_invalid_field(ctsio,
5360                                                       /*sks_valid*/ 1,
5361                                                       /*command*/ 0,
5362                                                       /*field*/ 2,
5363                                                       /*bit_valid*/ 0,
5364                                                       /*bit*/ 0);
5365                                 goto bailout;
5366                         }
5367                 }
5368         }
5369
5370         ctl_set_success(ctsio);
5371 bailout:
5372
5373         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5374                 free(ctsio->kern_data_ptr, M_CTL);
5375                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5376         }
5377
5378         ctl_done((union ctl_io *)ctsio);
5379         return (CTL_RETVAL_COMPLETE);
5380 }
5381
5382 int
5383 ctl_read_buffer(struct ctl_scsiio *ctsio)
5384 {
5385         struct ctl_lun *lun = CTL_LUN(ctsio);
5386         uint64_t buffer_offset;
5387         uint32_t len;
5388         uint8_t byte2;
5389         static uint8_t descr[4];
5390         static uint8_t echo_descr[4] = { 0 };
5391
5392         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5393
5394         switch (ctsio->cdb[0]) {
5395         case READ_BUFFER: {
5396                 struct scsi_read_buffer *cdb;
5397
5398                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5399                 buffer_offset = scsi_3btoul(cdb->offset);
5400                 len = scsi_3btoul(cdb->length);
5401                 byte2 = cdb->byte2;
5402                 break;
5403         }
5404         case READ_BUFFER_16: {
5405                 struct scsi_read_buffer_16 *cdb;
5406
5407                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5408                 buffer_offset = scsi_8btou64(cdb->offset);
5409                 len = scsi_4btoul(cdb->length);
5410                 byte2 = cdb->byte2;
5411                 break;
5412         }
5413         default: /* This shouldn't happen. */
5414                 ctl_set_invalid_opcode(ctsio);
5415                 ctl_done((union ctl_io *)ctsio);
5416                 return (CTL_RETVAL_COMPLETE);
5417         }
5418
5419         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5420             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5421                 ctl_set_invalid_field(ctsio,
5422                                       /*sks_valid*/ 1,
5423                                       /*command*/ 1,
5424                                       /*field*/ 6,
5425                                       /*bit_valid*/ 0,
5426                                       /*bit*/ 0);
5427                 ctl_done((union ctl_io *)ctsio);
5428                 return (CTL_RETVAL_COMPLETE);
5429         }
5430
5431         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5432                 descr[0] = 0;
5433                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5434                 ctsio->kern_data_ptr = descr;
5435                 len = min(len, sizeof(descr));
5436         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5437                 ctsio->kern_data_ptr = echo_descr;
5438                 len = min(len, sizeof(echo_descr));
5439         } else {
5440                 if (lun->write_buffer == NULL) {
5441                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5442                             M_CTL, M_WAITOK);
5443                 }
5444                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5445         }
5446         ctsio->kern_data_len = len;
5447         ctsio->kern_total_len = len;
5448         ctsio->kern_rel_offset = 0;
5449         ctsio->kern_sg_entries = 0;
5450         ctl_set_success(ctsio);
5451         ctsio->be_move_done = ctl_config_move_done;
5452         ctl_datamove((union ctl_io *)ctsio);
5453         return (CTL_RETVAL_COMPLETE);
5454 }
5455
5456 int
5457 ctl_write_buffer(struct ctl_scsiio *ctsio)
5458 {
5459         struct ctl_lun *lun = CTL_LUN(ctsio);
5460         struct scsi_write_buffer *cdb;
5461         int buffer_offset, len;
5462
5463         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5464
5465         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5466
5467         len = scsi_3btoul(cdb->length);
5468         buffer_offset = scsi_3btoul(cdb->offset);
5469
5470         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5471                 ctl_set_invalid_field(ctsio,
5472                                       /*sks_valid*/ 1,
5473                                       /*command*/ 1,
5474                                       /*field*/ 6,
5475                                       /*bit_valid*/ 0,
5476                                       /*bit*/ 0);
5477                 ctl_done((union ctl_io *)ctsio);
5478                 return (CTL_RETVAL_COMPLETE);
5479         }
5480
5481         /*
5482          * If we've got a kernel request that hasn't been malloced yet,
5483          * malloc it and tell the caller the data buffer is here.
5484          */
5485         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5486                 if (lun->write_buffer == NULL) {
5487                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5488                             M_CTL, M_WAITOK);
5489                 }
5490                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5491                 ctsio->kern_data_len = len;
5492                 ctsio->kern_total_len = len;
5493                 ctsio->kern_rel_offset = 0;
5494                 ctsio->kern_sg_entries = 0;
5495                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5496                 ctsio->be_move_done = ctl_config_move_done;
5497                 ctl_datamove((union ctl_io *)ctsio);
5498
5499                 return (CTL_RETVAL_COMPLETE);
5500         }
5501
5502         ctl_set_success(ctsio);
5503         ctl_done((union ctl_io *)ctsio);
5504         return (CTL_RETVAL_COMPLETE);
5505 }
5506
5507 static int
5508 ctl_write_same_cont(union ctl_io *io)
5509 {
5510         struct ctl_lun *lun = CTL_LUN(io);
5511         struct ctl_scsiio *ctsio;
5512         struct ctl_lba_len_flags *lbalen;
5513         int retval;
5514
5515         ctsio = &io->scsiio;
5516         ctsio->io_hdr.status = CTL_STATUS_NONE;
5517         lbalen = (struct ctl_lba_len_flags *)
5518             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5519         lbalen->lba += lbalen->len;
5520         if ((lun->be_lun->maxlba + 1) - lbalen->lba <= UINT32_MAX) {
5521                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
5522                 lbalen->len = (lun->be_lun->maxlba + 1) - lbalen->lba;
5523         }
5524
5525         CTL_DEBUG_PRINT(("ctl_write_same_cont: calling config_write()\n"));
5526         retval = lun->backend->config_write((union ctl_io *)ctsio);
5527         return (retval);
5528 }
5529
5530 int
5531 ctl_write_same(struct ctl_scsiio *ctsio)
5532 {
5533         struct ctl_lun *lun = CTL_LUN(ctsio);
5534         struct ctl_lba_len_flags *lbalen;
5535         const char *val;
5536         uint64_t lba, ival;
5537         uint32_t num_blocks;
5538         int len, retval;
5539         uint8_t byte2;
5540
5541         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5542
5543         switch (ctsio->cdb[0]) {
5544         case WRITE_SAME_10: {
5545                 struct scsi_write_same_10 *cdb;
5546
5547                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5548
5549                 lba = scsi_4btoul(cdb->addr);
5550                 num_blocks = scsi_2btoul(cdb->length);
5551                 byte2 = cdb->byte2;
5552                 break;
5553         }
5554         case WRITE_SAME_16: {
5555                 struct scsi_write_same_16 *cdb;
5556
5557                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5558
5559                 lba = scsi_8btou64(cdb->addr);
5560                 num_blocks = scsi_4btoul(cdb->length);
5561                 byte2 = cdb->byte2;
5562                 break;
5563         }
5564         default:
5565                 /*
5566                  * We got a command we don't support.  This shouldn't
5567                  * happen, commands should be filtered out above us.
5568                  */
5569                 ctl_set_invalid_opcode(ctsio);
5570                 ctl_done((union ctl_io *)ctsio);
5571
5572                 return (CTL_RETVAL_COMPLETE);
5573                 break; /* NOTREACHED */
5574         }
5575
5576         /* ANCHOR flag can be used only together with UNMAP */
5577         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5578                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5579                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5580                 ctl_done((union ctl_io *)ctsio);
5581                 return (CTL_RETVAL_COMPLETE);
5582         }
5583
5584         /*
5585          * The first check is to make sure we're in bounds, the second
5586          * check is to catch wrap-around problems.  If the lba + num blocks
5587          * is less than the lba, then we've wrapped around and the block
5588          * range is invalid anyway.
5589          */
5590         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5591          || ((lba + num_blocks) < lba)) {
5592                 ctl_set_lba_out_of_range(ctsio,
5593                     MAX(lba, lun->be_lun->maxlba + 1));
5594                 ctl_done((union ctl_io *)ctsio);
5595                 return (CTL_RETVAL_COMPLETE);
5596         }
5597
5598         /* Zero number of blocks means "to the last logical block" */
5599         if (num_blocks == 0) {
5600                 ival = UINT64_MAX;
5601                 val = dnvlist_get_string(lun->be_lun->options,
5602                     "write_same_max_lba", NULL);
5603                 if (val != NULL)
5604                         ctl_expand_number(val, &ival);
5605                 if ((lun->be_lun->maxlba + 1) - lba > ival) {
5606                         ctl_set_invalid_field(ctsio,
5607                             /*sks_valid*/ 1, /*command*/ 1,
5608                             /*field*/ ctsio->cdb[0] == WRITE_SAME_10 ? 7 : 10,
5609                             /*bit_valid*/ 0, /*bit*/ 0);
5610                         ctl_done((union ctl_io *)ctsio);
5611                         return (CTL_RETVAL_COMPLETE);
5612                 }
5613                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5614                         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
5615                         ctsio->io_cont = ctl_write_same_cont;
5616                         num_blocks = 1 << 31;
5617                 } else
5618                         num_blocks = (lun->be_lun->maxlba + 1) - lba;
5619         }
5620
5621         len = lun->be_lun->blocksize;
5622
5623         /*
5624          * If we've got a kernel request that hasn't been malloced yet,
5625          * malloc it and tell the caller the data buffer is here.
5626          */
5627         if ((byte2 & SWS_NDOB) == 0 &&
5628             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5629                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5630                 ctsio->kern_data_len = len;
5631                 ctsio->kern_total_len = len;
5632                 ctsio->kern_rel_offset = 0;
5633                 ctsio->kern_sg_entries = 0;
5634                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5635                 ctsio->be_move_done = ctl_config_move_done;
5636                 ctl_datamove((union ctl_io *)ctsio);
5637
5638                 return (CTL_RETVAL_COMPLETE);
5639         }
5640
5641         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5642         lbalen->lba = lba;
5643         lbalen->len = num_blocks;
5644         lbalen->flags = byte2;
5645         retval = lun->backend->config_write((union ctl_io *)ctsio);
5646
5647         return (retval);
5648 }
5649
5650 int
5651 ctl_unmap(struct ctl_scsiio *ctsio)
5652 {
5653         struct ctl_lun *lun = CTL_LUN(ctsio);
5654         struct scsi_unmap *cdb;
5655         struct ctl_ptr_len_flags *ptrlen;
5656         struct scsi_unmap_header *hdr;
5657         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5658         uint64_t lba;
5659         uint32_t num_blocks;
5660         int len, retval;
5661         uint8_t byte2;
5662
5663         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5664
5665         cdb = (struct scsi_unmap *)ctsio->cdb;
5666         len = scsi_2btoul(cdb->length);
5667         byte2 = cdb->byte2;
5668
5669         /*
5670          * If we've got a kernel request that hasn't been malloced yet,
5671          * malloc it and tell the caller the data buffer is here.
5672          */
5673         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5674                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5675                 ctsio->kern_data_len = len;
5676                 ctsio->kern_total_len = len;
5677                 ctsio->kern_rel_offset = 0;
5678                 ctsio->kern_sg_entries = 0;
5679                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5680                 ctsio->be_move_done = ctl_config_move_done;
5681                 ctl_datamove((union ctl_io *)ctsio);
5682
5683                 return (CTL_RETVAL_COMPLETE);
5684         }
5685
5686         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5687         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5688         if (len < sizeof (*hdr) ||
5689             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5690             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5691             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5692                 ctl_set_invalid_field(ctsio,
5693                                       /*sks_valid*/ 0,
5694                                       /*command*/ 0,
5695                                       /*field*/ 0,
5696                                       /*bit_valid*/ 0,
5697                                       /*bit*/ 0);
5698                 goto done;
5699         }
5700         len = scsi_2btoul(hdr->desc_length);
5701         buf = (struct scsi_unmap_desc *)(hdr + 1);
5702         end = buf + len / sizeof(*buf);
5703
5704         endnz = buf;
5705         for (range = buf; range < end; range++) {
5706                 lba = scsi_8btou64(range->lba);
5707                 num_blocks = scsi_4btoul(range->length);
5708                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5709                  || ((lba + num_blocks) < lba)) {
5710                         ctl_set_lba_out_of_range(ctsio,
5711                             MAX(lba, lun->be_lun->maxlba + 1));
5712                         ctl_done((union ctl_io *)ctsio);
5713                         return (CTL_RETVAL_COMPLETE);
5714                 }
5715                 if (num_blocks != 0)
5716                         endnz = range + 1;
5717         }
5718
5719         /*
5720          * Block backend can not handle zero last range.
5721          * Filter it out and return if there is nothing left.
5722          */
5723         len = (uint8_t *)endnz - (uint8_t *)buf;
5724         if (len == 0) {
5725                 ctl_set_success(ctsio);
5726                 goto done;
5727         }
5728
5729         mtx_lock(&lun->lun_lock);
5730         ptrlen = (struct ctl_ptr_len_flags *)
5731             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5732         ptrlen->ptr = (void *)buf;
5733         ptrlen->len = len;
5734         ptrlen->flags = byte2;
5735         ctl_try_unblock_others(lun, (union ctl_io *)ctsio, FALSE);
5736         mtx_unlock(&lun->lun_lock);
5737
5738         retval = lun->backend->config_write((union ctl_io *)ctsio);
5739         return (retval);
5740
5741 done:
5742         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5743                 free(ctsio->kern_data_ptr, M_CTL);
5744                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5745         }
5746         ctl_done((union ctl_io *)ctsio);
5747         return (CTL_RETVAL_COMPLETE);
5748 }
5749
5750 int
5751 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5752                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5753 {
5754         struct ctl_lun *lun = CTL_LUN(ctsio);
5755         uint8_t *current_cp;
5756         int set_ua;
5757         uint32_t initidx;
5758
5759         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5760         set_ua = 0;
5761
5762         current_cp = (page_index->page_data + (page_index->page_len *
5763             CTL_PAGE_CURRENT));
5764
5765         mtx_lock(&lun->lun_lock);
5766         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5767                 memcpy(current_cp, page_ptr, page_index->page_len);
5768                 set_ua = 1;
5769         }
5770         if (set_ua != 0)
5771                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5772         mtx_unlock(&lun->lun_lock);
5773         if (set_ua) {
5774                 ctl_isc_announce_mode(lun,
5775                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5776                     page_index->page_code, page_index->subpage);
5777         }
5778         return (CTL_RETVAL_COMPLETE);
5779 }
5780
5781 static void
5782 ctl_ie_timer(void *arg)
5783 {
5784         struct ctl_lun *lun = arg;
5785         uint64_t t;
5786
5787         if (lun->ie_asc == 0)
5788                 return;
5789
5790         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5791                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5792         else
5793                 lun->ie_reported = 0;
5794
5795         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5796                 lun->ie_reportcnt++;
5797                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5798                 if (t == 0 || t == UINT32_MAX)
5799                         t = 3000;  /* 5 min */
5800                 callout_schedule_sbt(&lun->ie_callout, SBT_1S / 10 * t,
5801                     SBT_1S / 10, 0);
5802         }
5803 }
5804
5805 int
5806 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5807                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5808 {
5809         struct ctl_lun *lun = CTL_LUN(ctsio);
5810         struct scsi_info_exceptions_page *pg;
5811         uint64_t t;
5812
5813         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5814
5815         pg = (struct scsi_info_exceptions_page *)page_ptr;
5816         mtx_lock(&lun->lun_lock);
5817         if (pg->info_flags & SIEP_FLAGS_TEST) {
5818                 lun->ie_asc = 0x5d;
5819                 lun->ie_ascq = 0xff;
5820                 if (pg->mrie == SIEP_MRIE_UA) {
5821                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
5822                         lun->ie_reported = 1;
5823                 } else {
5824                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5825                         lun->ie_reported = -1;
5826                 }
5827                 lun->ie_reportcnt = 1;
5828                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
5829                         lun->ie_reportcnt++;
5830                         t = scsi_4btoul(pg->interval_timer);
5831                         if (t == 0 || t == UINT32_MAX)
5832                                 t = 3000;  /* 5 min */
5833                         callout_reset_sbt(&lun->ie_callout, SBT_1S / 10 * t,
5834                             SBT_1S / 10, ctl_ie_timer, lun, 0);
5835                 }
5836         } else {
5837                 lun->ie_asc = 0;
5838                 lun->ie_ascq = 0;
5839                 lun->ie_reported = 1;
5840                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5841                 lun->ie_reportcnt = UINT32_MAX;
5842                 callout_stop(&lun->ie_callout);
5843         }
5844         mtx_unlock(&lun->lun_lock);
5845         return (CTL_RETVAL_COMPLETE);
5846 }
5847
5848 static int
5849 ctl_do_mode_select(union ctl_io *io)
5850 {
5851         struct ctl_lun *lun = CTL_LUN(io);
5852         struct scsi_mode_page_header *page_header;
5853         struct ctl_page_index *page_index;
5854         struct ctl_scsiio *ctsio;
5855         int page_len, page_len_offset, page_len_size;
5856         union ctl_modepage_info *modepage_info;
5857         uint16_t *len_left, *len_used;
5858         int retval, i;
5859
5860         ctsio = &io->scsiio;
5861         page_index = NULL;
5862         page_len = 0;
5863
5864         modepage_info = (union ctl_modepage_info *)
5865                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5866         len_left = &modepage_info->header.len_left;
5867         len_used = &modepage_info->header.len_used;
5868
5869 do_next_page:
5870
5871         page_header = (struct scsi_mode_page_header *)
5872                 (ctsio->kern_data_ptr + *len_used);
5873
5874         if (*len_left == 0) {
5875                 free(ctsio->kern_data_ptr, M_CTL);
5876                 ctl_set_success(ctsio);
5877                 ctl_done((union ctl_io *)ctsio);
5878                 return (CTL_RETVAL_COMPLETE);
5879         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
5880                 free(ctsio->kern_data_ptr, M_CTL);
5881                 ctl_set_param_len_error(ctsio);
5882                 ctl_done((union ctl_io *)ctsio);
5883                 return (CTL_RETVAL_COMPLETE);
5884
5885         } else if ((page_header->page_code & SMPH_SPF)
5886                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
5887                 free(ctsio->kern_data_ptr, M_CTL);
5888                 ctl_set_param_len_error(ctsio);
5889                 ctl_done((union ctl_io *)ctsio);
5890                 return (CTL_RETVAL_COMPLETE);
5891         }
5892
5893         /*
5894          * XXX KDM should we do something with the block descriptor?
5895          */
5896         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5897                 page_index = &lun->mode_pages.index[i];
5898                 if (lun->be_lun->lun_type == T_DIRECT &&
5899                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
5900                         continue;
5901                 if (lun->be_lun->lun_type == T_PROCESSOR &&
5902                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
5903                         continue;
5904                 if (lun->be_lun->lun_type == T_CDROM &&
5905                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
5906                         continue;
5907
5908                 if ((page_index->page_code & SMPH_PC_MASK) !=
5909                     (page_header->page_code & SMPH_PC_MASK))
5910                         continue;
5911
5912                 /*
5913                  * If neither page has a subpage code, then we've got a
5914                  * match.
5915                  */
5916                 if (((page_index->page_code & SMPH_SPF) == 0)
5917                  && ((page_header->page_code & SMPH_SPF) == 0)) {
5918                         page_len = page_header->page_length;
5919                         break;
5920                 }
5921
5922                 /*
5923                  * If both pages have subpages, then the subpage numbers
5924                  * have to match.
5925                  */
5926                 if ((page_index->page_code & SMPH_SPF)
5927                   && (page_header->page_code & SMPH_SPF)) {
5928                         struct scsi_mode_page_header_sp *sph;
5929
5930                         sph = (struct scsi_mode_page_header_sp *)page_header;
5931                         if (page_index->subpage == sph->subpage) {
5932                                 page_len = scsi_2btoul(sph->page_length);
5933                                 break;
5934                         }
5935                 }
5936         }
5937
5938         /*
5939          * If we couldn't find the page, or if we don't have a mode select
5940          * handler for it, send back an error to the user.
5941          */
5942         if ((i >= CTL_NUM_MODE_PAGES)
5943          || (page_index->select_handler == NULL)) {
5944                 ctl_set_invalid_field(ctsio,
5945                                       /*sks_valid*/ 1,
5946                                       /*command*/ 0,
5947                                       /*field*/ *len_used,
5948                                       /*bit_valid*/ 0,
5949                                       /*bit*/ 0);
5950                 free(ctsio->kern_data_ptr, M_CTL);
5951                 ctl_done((union ctl_io *)ctsio);
5952                 return (CTL_RETVAL_COMPLETE);
5953         }
5954
5955         if (page_index->page_code & SMPH_SPF) {
5956                 page_len_offset = 2;
5957                 page_len_size = 2;
5958         } else {
5959                 page_len_size = 1;
5960                 page_len_offset = 1;
5961         }
5962
5963         /*
5964          * If the length the initiator gives us isn't the one we specify in
5965          * the mode page header, or if they didn't specify enough data in
5966          * the CDB to avoid truncating this page, kick out the request.
5967          */
5968         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
5969                 ctl_set_invalid_field(ctsio,
5970                                       /*sks_valid*/ 1,
5971                                       /*command*/ 0,
5972                                       /*field*/ *len_used + page_len_offset,
5973                                       /*bit_valid*/ 0,
5974                                       /*bit*/ 0);
5975                 free(ctsio->kern_data_ptr, M_CTL);
5976                 ctl_done((union ctl_io *)ctsio);
5977                 return (CTL_RETVAL_COMPLETE);
5978         }
5979         if (*len_left < page_index->page_len) {
5980                 free(ctsio->kern_data_ptr, M_CTL);
5981                 ctl_set_param_len_error(ctsio);
5982                 ctl_done((union ctl_io *)ctsio);
5983                 return (CTL_RETVAL_COMPLETE);
5984         }
5985
5986         /*
5987          * Run through the mode page, checking to make sure that the bits
5988          * the user changed are actually legal for him to change.
5989          */
5990         for (i = 0; i < page_index->page_len; i++) {
5991                 uint8_t *user_byte, *change_mask, *current_byte;
5992                 int bad_bit;
5993                 int j;
5994
5995                 user_byte = (uint8_t *)page_header + i;
5996                 change_mask = page_index->page_data +
5997                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
5998                 current_byte = page_index->page_data +
5999                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6000
6001                 /*
6002                  * Check to see whether the user set any bits in this byte
6003                  * that he is not allowed to set.
6004                  */
6005                 if ((*user_byte & ~(*change_mask)) ==
6006                     (*current_byte & ~(*change_mask)))
6007                         continue;
6008
6009                 /*
6010                  * Go through bit by bit to determine which one is illegal.
6011                  */
6012                 bad_bit = 0;
6013                 for (j = 7; j >= 0; j--) {
6014                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6015                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6016                                 bad_bit = i;
6017                                 break;
6018                         }
6019                 }
6020                 ctl_set_invalid_field(ctsio,
6021                                       /*sks_valid*/ 1,
6022                                       /*command*/ 0,
6023                                       /*field*/ *len_used + i,
6024                                       /*bit_valid*/ 1,
6025                                       /*bit*/ bad_bit);
6026                 free(ctsio->kern_data_ptr, M_CTL);
6027                 ctl_done((union ctl_io *)ctsio);
6028                 return (CTL_RETVAL_COMPLETE);
6029         }
6030
6031         /*
6032          * Decrement these before we call the page handler, since we may
6033          * end up getting called back one way or another before the handler
6034          * returns to this context.
6035          */
6036         *len_left -= page_index->page_len;
6037         *len_used += page_index->page_len;
6038
6039         retval = page_index->select_handler(ctsio, page_index,
6040                                             (uint8_t *)page_header);
6041
6042         /*
6043          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6044          * wait until this queued command completes to finish processing
6045          * the mode page.  If it returns anything other than
6046          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6047          * already set the sense information, freed the data pointer, and
6048          * completed the io for us.
6049          */
6050         if (retval != CTL_RETVAL_COMPLETE)
6051                 goto bailout_no_done;
6052
6053         /*
6054          * If the initiator sent us more than one page, parse the next one.
6055          */
6056         if (*len_left > 0)
6057                 goto do_next_page;
6058
6059         ctl_set_success(ctsio);
6060         free(ctsio->kern_data_ptr, M_CTL);
6061         ctl_done((union ctl_io *)ctsio);
6062
6063 bailout_no_done:
6064
6065         return (CTL_RETVAL_COMPLETE);
6066
6067 }
6068
6069 int
6070 ctl_mode_select(struct ctl_scsiio *ctsio)
6071 {
6072         struct ctl_lun *lun = CTL_LUN(ctsio);
6073         union ctl_modepage_info *modepage_info;
6074         int bd_len, i, header_size, param_len, rtd;
6075         uint32_t initidx;
6076
6077         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6078         switch (ctsio->cdb[0]) {
6079         case MODE_SELECT_6: {
6080                 struct scsi_mode_select_6 *cdb;
6081
6082                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6083
6084                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6085                 param_len = cdb->length;
6086                 header_size = sizeof(struct scsi_mode_header_6);
6087                 break;
6088         }
6089         case MODE_SELECT_10: {
6090                 struct scsi_mode_select_10 *cdb;
6091
6092                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6093
6094                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6095                 param_len = scsi_2btoul(cdb->length);
6096                 header_size = sizeof(struct scsi_mode_header_10);
6097                 break;
6098         }
6099         default:
6100                 ctl_set_invalid_opcode(ctsio);
6101                 ctl_done((union ctl_io *)ctsio);
6102                 return (CTL_RETVAL_COMPLETE);
6103         }
6104
6105         if (rtd) {
6106                 if (param_len != 0) {
6107                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 0,
6108                             /*command*/ 1, /*field*/ 0,
6109                             /*bit_valid*/ 0, /*bit*/ 0);
6110                         ctl_done((union ctl_io *)ctsio);
6111                         return (CTL_RETVAL_COMPLETE);
6112                 }
6113
6114                 /* Revert to defaults. */
6115                 ctl_init_page_index(lun);
6116                 mtx_lock(&lun->lun_lock);
6117                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6118                 mtx_unlock(&lun->lun_lock);
6119                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6120                         ctl_isc_announce_mode(lun, -1,
6121                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
6122                             lun->mode_pages.index[i].subpage);
6123                 }
6124                 ctl_set_success(ctsio);
6125                 ctl_done((union ctl_io *)ctsio);
6126                 return (CTL_RETVAL_COMPLETE);
6127         }
6128
6129         /*
6130          * From SPC-3:
6131          * "A parameter list length of zero indicates that the Data-Out Buffer
6132          * shall be empty. This condition shall not be considered as an error."
6133          */
6134         if (param_len == 0) {
6135                 ctl_set_success(ctsio);
6136                 ctl_done((union ctl_io *)ctsio);
6137                 return (CTL_RETVAL_COMPLETE);
6138         }
6139
6140         /*
6141          * Since we'll hit this the first time through, prior to
6142          * allocation, we don't need to free a data buffer here.
6143          */
6144         if (param_len < header_size) {
6145                 ctl_set_param_len_error(ctsio);
6146                 ctl_done((union ctl_io *)ctsio);
6147                 return (CTL_RETVAL_COMPLETE);
6148         }
6149
6150         /*
6151          * Allocate the data buffer and grab the user's data.  In theory,
6152          * we shouldn't have to sanity check the parameter list length here
6153          * because the maximum size is 64K.  We should be able to malloc
6154          * that much without too many problems.
6155          */
6156         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6157                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6158                 ctsio->kern_data_len = param_len;
6159                 ctsio->kern_total_len = param_len;
6160                 ctsio->kern_rel_offset = 0;
6161                 ctsio->kern_sg_entries = 0;
6162                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6163                 ctsio->be_move_done = ctl_config_move_done;
6164                 ctl_datamove((union ctl_io *)ctsio);
6165
6166                 return (CTL_RETVAL_COMPLETE);
6167         }
6168
6169         switch (ctsio->cdb[0]) {
6170         case MODE_SELECT_6: {
6171                 struct scsi_mode_header_6 *mh6;
6172
6173                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6174                 bd_len = mh6->blk_desc_len;
6175                 break;
6176         }
6177         case MODE_SELECT_10: {
6178                 struct scsi_mode_header_10 *mh10;
6179
6180                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6181                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6182                 break;
6183         }
6184         default:
6185                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6186         }
6187
6188         if (param_len < (header_size + bd_len)) {
6189                 free(ctsio->kern_data_ptr, M_CTL);
6190                 ctl_set_param_len_error(ctsio);
6191                 ctl_done((union ctl_io *)ctsio);
6192                 return (CTL_RETVAL_COMPLETE);
6193         }
6194
6195         /*
6196          * Set the IO_CONT flag, so that if this I/O gets passed to
6197          * ctl_config_write_done(), it'll get passed back to
6198          * ctl_do_mode_select() for further processing, or completion if
6199          * we're all done.
6200          */
6201         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6202         ctsio->io_cont = ctl_do_mode_select;
6203
6204         modepage_info = (union ctl_modepage_info *)
6205                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6206         memset(modepage_info, 0, sizeof(*modepage_info));
6207         modepage_info->header.len_left = param_len - header_size - bd_len;
6208         modepage_info->header.len_used = header_size + bd_len;
6209
6210         return (ctl_do_mode_select((union ctl_io *)ctsio));
6211 }
6212
6213 int
6214 ctl_mode_sense(struct ctl_scsiio *ctsio)
6215 {
6216         struct ctl_lun *lun = CTL_LUN(ctsio);
6217         int pc, page_code, llba, subpage;
6218         int alloc_len, page_len, header_len, bd_len, total_len;
6219         void *block_desc;
6220         struct ctl_page_index *page_index;
6221
6222         llba = 0;
6223
6224         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6225
6226         switch (ctsio->cdb[0]) {
6227         case MODE_SENSE_6: {
6228                 struct scsi_mode_sense_6 *cdb;
6229
6230                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6231
6232                 header_len = sizeof(struct scsi_mode_hdr_6);
6233                 if (cdb->byte2 & SMS_DBD)
6234                         bd_len = 0;
6235                 else
6236                         bd_len = sizeof(struct scsi_mode_block_descr);
6237                 header_len += bd_len;
6238
6239                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6240                 page_code = cdb->page & SMS_PAGE_CODE;
6241                 subpage = cdb->subpage;
6242                 alloc_len = cdb->length;
6243                 break;
6244         }
6245         case MODE_SENSE_10: {
6246                 struct scsi_mode_sense_10 *cdb;
6247
6248                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6249
6250                 header_len = sizeof(struct scsi_mode_hdr_10);
6251                 if (cdb->byte2 & SMS_DBD) {
6252                         bd_len = 0;
6253                 } else if (lun->be_lun->lun_type == T_DIRECT) {
6254                         if (cdb->byte2 & SMS10_LLBAA) {
6255                                 llba = 1;
6256                                 bd_len = sizeof(struct scsi_mode_block_descr_dlong);
6257                         } else
6258                                 bd_len = sizeof(struct scsi_mode_block_descr_dshort);
6259                 } else
6260                         bd_len = sizeof(struct scsi_mode_block_descr);
6261                 header_len += bd_len;
6262
6263                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6264                 page_code = cdb->page & SMS_PAGE_CODE;
6265                 subpage = cdb->subpage;
6266                 alloc_len = scsi_2btoul(cdb->length);
6267                 break;
6268         }
6269         default:
6270                 ctl_set_invalid_opcode(ctsio);
6271                 ctl_done((union ctl_io *)ctsio);
6272                 return (CTL_RETVAL_COMPLETE);
6273                 break; /* NOTREACHED */
6274         }
6275
6276         /*
6277          * We have to make a first pass through to calculate the size of
6278          * the pages that match the user's query.  Then we allocate enough
6279          * memory to hold it, and actually copy the data into the buffer.
6280          */
6281         switch (page_code) {
6282         case SMS_ALL_PAGES_PAGE: {
6283                 u_int i;
6284
6285                 page_len = 0;
6286
6287                 /*
6288                  * At the moment, values other than 0 and 0xff here are
6289                  * reserved according to SPC-3.
6290                  */
6291                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6292                  && (subpage != SMS_SUBPAGE_ALL)) {
6293                         ctl_set_invalid_field(ctsio,
6294                                               /*sks_valid*/ 1,
6295                                               /*command*/ 1,
6296                                               /*field*/ 3,
6297                                               /*bit_valid*/ 0,
6298                                               /*bit*/ 0);
6299                         ctl_done((union ctl_io *)ctsio);
6300                         return (CTL_RETVAL_COMPLETE);
6301                 }
6302
6303                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6304                         page_index = &lun->mode_pages.index[i];
6305
6306                         /* Make sure the page is supported for this dev type */
6307                         if (lun->be_lun->lun_type == T_DIRECT &&
6308                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6309                                 continue;
6310                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6311                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6312                                 continue;
6313                         if (lun->be_lun->lun_type == T_CDROM &&
6314                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6315                                 continue;
6316
6317                         /*
6318                          * We don't use this subpage if the user didn't
6319                          * request all subpages.
6320                          */
6321                         if ((page_index->subpage != 0)
6322                          && (subpage == SMS_SUBPAGE_PAGE_0))
6323                                 continue;
6324
6325                         page_len += page_index->page_len;
6326                 }
6327                 break;
6328         }
6329         default: {
6330                 u_int i;
6331
6332                 page_len = 0;
6333
6334                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6335                         page_index = &lun->mode_pages.index[i];
6336
6337                         /* Make sure the page is supported for this dev type */
6338                         if (lun->be_lun->lun_type == T_DIRECT &&
6339                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6340                                 continue;
6341                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6342                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6343                                 continue;
6344                         if (lun->be_lun->lun_type == T_CDROM &&
6345                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6346                                 continue;
6347
6348                         /* Look for the right page code */
6349                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6350                                 continue;
6351
6352                         /* Look for the right subpage or the subpage wildcard*/
6353                         if ((page_index->subpage != subpage)
6354                          && (subpage != SMS_SUBPAGE_ALL))
6355                                 continue;
6356
6357                         page_len += page_index->page_len;
6358                 }
6359
6360                 if (page_len == 0) {
6361                         ctl_set_invalid_field(ctsio,
6362                                               /*sks_valid*/ 1,
6363                                               /*command*/ 1,
6364                                               /*field*/ 2,
6365                                               /*bit_valid*/ 1,
6366                                               /*bit*/ 5);
6367                         ctl_done((union ctl_io *)ctsio);
6368                         return (CTL_RETVAL_COMPLETE);
6369                 }
6370                 break;
6371         }
6372         }
6373
6374         total_len = header_len + page_len;
6375
6376         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6377         ctsio->kern_sg_entries = 0;
6378         ctsio->kern_rel_offset = 0;
6379         ctsio->kern_data_len = min(total_len, alloc_len);
6380         ctsio->kern_total_len = ctsio->kern_data_len;
6381
6382         switch (ctsio->cdb[0]) {
6383         case MODE_SENSE_6: {
6384                 struct scsi_mode_hdr_6 *header;
6385
6386                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6387
6388                 header->datalen = MIN(total_len - 1, 254);
6389                 if (lun->be_lun->lun_type == T_DIRECT) {
6390                         header->dev_specific = 0x10; /* DPOFUA */
6391                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6392                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6393                                 header->dev_specific |= 0x80; /* WP */
6394                 }
6395                 header->block_descr_len = bd_len;
6396                 block_desc = &header[1];
6397                 break;
6398         }
6399         case MODE_SENSE_10: {
6400                 struct scsi_mode_hdr_10 *header;
6401                 int datalen;
6402
6403                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6404
6405                 datalen = MIN(total_len - 2, 65533);
6406                 scsi_ulto2b(datalen, header->datalen);
6407                 if (lun->be_lun->lun_type == T_DIRECT) {
6408                         header->dev_specific = 0x10; /* DPOFUA */
6409                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6410                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6411                                 header->dev_specific |= 0x80; /* WP */
6412                 }
6413                 if (llba)
6414                         header->flags |= SMH_LONGLBA;
6415                 scsi_ulto2b(bd_len, header->block_descr_len);
6416                 block_desc = &header[1];
6417                 break;
6418         }
6419         default:
6420                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6421         }
6422
6423         /*
6424          * If we've got a disk, use its blocksize in the block
6425          * descriptor.  Otherwise, just set it to 0.
6426          */
6427         if (bd_len > 0) {
6428                 if (lun->be_lun->lun_type == T_DIRECT) {
6429                         if (llba) {
6430                                 struct scsi_mode_block_descr_dlong *bd = block_desc;
6431                                 if (lun->be_lun->maxlba != 0)
6432                                         scsi_u64to8b(lun->be_lun->maxlba + 1,
6433                                             bd->num_blocks);
6434                                 scsi_ulto4b(lun->be_lun->blocksize,
6435                                     bd->block_len);
6436                         } else {
6437                                 struct scsi_mode_block_descr_dshort *bd = block_desc;
6438                                 if (lun->be_lun->maxlba != 0)
6439                                         scsi_ulto4b(MIN(lun->be_lun->maxlba+1,
6440                                             UINT32_MAX), bd->num_blocks);
6441                                 scsi_ulto3b(lun->be_lun->blocksize,
6442                                     bd->block_len);
6443                         }
6444                 } else {
6445                         struct scsi_mode_block_descr *bd = block_desc;
6446                         scsi_ulto3b(0, bd->block_len);
6447                 }
6448         }
6449
6450         switch (page_code) {
6451         case SMS_ALL_PAGES_PAGE: {
6452                 int i, data_used;
6453
6454                 data_used = header_len;
6455                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6456                         struct ctl_page_index *page_index;
6457
6458                         page_index = &lun->mode_pages.index[i];
6459                         if (lun->be_lun->lun_type == T_DIRECT &&
6460                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6461                                 continue;
6462                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6463                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6464                                 continue;
6465                         if (lun->be_lun->lun_type == T_CDROM &&
6466                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6467                                 continue;
6468
6469                         /*
6470                          * We don't use this subpage if the user didn't
6471                          * request all subpages.  We already checked (above)
6472                          * to make sure the user only specified a subpage
6473                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6474                          */
6475                         if ((page_index->subpage != 0)
6476                          && (subpage == SMS_SUBPAGE_PAGE_0))
6477                                 continue;
6478
6479                         /*
6480                          * Call the handler, if it exists, to update the
6481                          * page to the latest values.
6482                          */
6483                         if (page_index->sense_handler != NULL)
6484                                 page_index->sense_handler(ctsio, page_index,pc);
6485
6486                         memcpy(ctsio->kern_data_ptr + data_used,
6487                                page_index->page_data +
6488                                (page_index->page_len * pc),
6489                                page_index->page_len);
6490                         data_used += page_index->page_len;
6491                 }
6492                 break;
6493         }
6494         default: {
6495                 int i, data_used;
6496
6497                 data_used = header_len;
6498
6499                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6500                         struct ctl_page_index *page_index;
6501
6502                         page_index = &lun->mode_pages.index[i];
6503
6504                         /* Look for the right page code */
6505                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6506                                 continue;
6507
6508                         /* Look for the right subpage or the subpage wildcard*/
6509                         if ((page_index->subpage != subpage)
6510                          && (subpage != SMS_SUBPAGE_ALL))
6511                                 continue;
6512
6513                         /* Make sure the page is supported for this dev type */
6514                         if (lun->be_lun->lun_type == T_DIRECT &&
6515                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6516                                 continue;
6517                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6518                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6519                                 continue;
6520                         if (lun->be_lun->lun_type == T_CDROM &&
6521                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6522                                 continue;
6523
6524                         /*
6525                          * Call the handler, if it exists, to update the
6526                          * page to the latest values.
6527                          */
6528                         if (page_index->sense_handler != NULL)
6529                                 page_index->sense_handler(ctsio, page_index,pc);
6530
6531                         memcpy(ctsio->kern_data_ptr + data_used,
6532                                page_index->page_data +
6533                                (page_index->page_len * pc),
6534                                page_index->page_len);
6535                         data_used += page_index->page_len;
6536                 }
6537                 break;
6538         }
6539         }
6540
6541         ctl_set_success(ctsio);
6542         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6543         ctsio->be_move_done = ctl_config_move_done;
6544         ctl_datamove((union ctl_io *)ctsio);
6545         return (CTL_RETVAL_COMPLETE);
6546 }
6547
6548 int
6549 ctl_temp_log_sense_handler(struct ctl_scsiio *ctsio,
6550                                struct ctl_page_index *page_index,
6551                                int pc)
6552 {
6553         struct ctl_lun *lun = CTL_LUN(ctsio);
6554         struct scsi_log_temperature *data;
6555         const char *value;
6556
6557         data = (struct scsi_log_temperature *)page_index->page_data;
6558
6559         scsi_ulto2b(SLP_TEMPERATURE, data->hdr.param_code);
6560         data->hdr.param_control = SLP_LBIN;
6561         data->hdr.param_len = sizeof(struct scsi_log_temperature) -
6562             sizeof(struct scsi_log_param_header);
6563         if ((value = dnvlist_get_string(lun->be_lun->options, "temperature",
6564             NULL)) != NULL)
6565                 data->temperature = strtol(value, NULL, 0);
6566         else
6567                 data->temperature = 0xff;
6568         data++;
6569
6570         scsi_ulto2b(SLP_REFTEMPERATURE, data->hdr.param_code);
6571         data->hdr.param_control = SLP_LBIN;
6572         data->hdr.param_len = sizeof(struct scsi_log_temperature) -
6573             sizeof(struct scsi_log_param_header);
6574         if ((value = dnvlist_get_string(lun->be_lun->options, "reftemperature",
6575             NULL)) != NULL)
6576                 data->temperature = strtol(value, NULL, 0);
6577         else
6578                 data->temperature = 0xff;
6579         return (0);
6580 }
6581
6582 int
6583 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6584                                struct ctl_page_index *page_index,
6585                                int pc)
6586 {
6587         struct ctl_lun *lun = CTL_LUN(ctsio);
6588         struct scsi_log_param_header *phdr;
6589         uint8_t *data;
6590         uint64_t val;
6591
6592         data = page_index->page_data;
6593
6594         if (lun->backend->lun_attr != NULL &&
6595             (val = lun->backend->lun_attr(lun->be_lun, "blocksavail"))
6596              != UINT64_MAX) {
6597                 phdr = (struct scsi_log_param_header *)data;
6598                 scsi_ulto2b(0x0001, phdr->param_code);
6599                 phdr->param_control = SLP_LBIN | SLP_LP;
6600                 phdr->param_len = 8;
6601                 data = (uint8_t *)(phdr + 1);
6602                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6603                 data[4] = 0x02; /* per-pool */
6604                 data += phdr->param_len;
6605         }
6606
6607         if (lun->backend->lun_attr != NULL &&
6608             (val = lun->backend->lun_attr(lun->be_lun, "blocksused"))
6609              != UINT64_MAX) {
6610                 phdr = (struct scsi_log_param_header *)data;
6611                 scsi_ulto2b(0x0002, phdr->param_code);
6612                 phdr->param_control = SLP_LBIN | SLP_LP;
6613                 phdr->param_len = 8;
6614                 data = (uint8_t *)(phdr + 1);
6615                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6616                 data[4] = 0x01; /* per-LUN */
6617                 data += phdr->param_len;
6618         }
6619
6620         if (lun->backend->lun_attr != NULL &&
6621             (val = lun->backend->lun_attr(lun->be_lun, "poolblocksavail"))
6622              != UINT64_MAX) {
6623                 phdr = (struct scsi_log_param_header *)data;
6624                 scsi_ulto2b(0x00f1, phdr->param_code);
6625                 phdr->param_control = SLP_LBIN | SLP_LP;
6626                 phdr->param_len = 8;
6627                 data = (uint8_t *)(phdr + 1);
6628                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6629                 data[4] = 0x02; /* per-pool */
6630                 data += phdr->param_len;
6631         }
6632
6633         if (lun->backend->lun_attr != NULL &&
6634             (val = lun->backend->lun_attr(lun->be_lun, "poolblocksused"))
6635              != UINT64_MAX) {
6636                 phdr = (struct scsi_log_param_header *)data;
6637                 scsi_ulto2b(0x00f2, phdr->param_code);
6638                 phdr->param_control = SLP_LBIN | SLP_LP;
6639                 phdr->param_len = 8;
6640                 data = (uint8_t *)(phdr + 1);
6641                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6642                 data[4] = 0x02; /* per-pool */
6643                 data += phdr->param_len;
6644         }
6645
6646         page_index->page_len = data - page_index->page_data;
6647         return (0);
6648 }
6649
6650 int
6651 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6652                                struct ctl_page_index *page_index,
6653                                int pc)
6654 {
6655         struct ctl_lun *lun = CTL_LUN(ctsio);
6656         struct stat_page *data;
6657         struct bintime *t;
6658
6659         data = (struct stat_page *)page_index->page_data;
6660
6661         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6662         data->sap.hdr.param_control = SLP_LBIN;
6663         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6664             sizeof(struct scsi_log_param_header);
6665         scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
6666             data->sap.read_num);
6667         scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
6668             data->sap.write_num);
6669         if (lun->be_lun->blocksize > 0) {
6670                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
6671                     lun->be_lun->blocksize, data->sap.recvieved_lba);
6672                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
6673                     lun->be_lun->blocksize, data->sap.transmitted_lba);
6674         }
6675         t = &lun->stats.time[CTL_STATS_READ];
6676         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6677             data->sap.read_int);
6678         t = &lun->stats.time[CTL_STATS_WRITE];
6679         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6680             data->sap.write_int);
6681         scsi_u64to8b(0, data->sap.weighted_num);
6682         scsi_u64to8b(0, data->sap.weighted_int);
6683         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6684         data->it.hdr.param_control = SLP_LBIN;
6685         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6686             sizeof(struct scsi_log_param_header);
6687 #ifdef CTL_TIME_IO
6688         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6689 #endif
6690         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6691         data->it.hdr.param_control = SLP_LBIN;
6692         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6693             sizeof(struct scsi_log_param_header);
6694         scsi_ulto4b(3, data->ti.exponent);
6695         scsi_ulto4b(1, data->ti.integer);
6696         return (0);
6697 }
6698
6699 int
6700 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6701                                struct ctl_page_index *page_index,
6702                                int pc)
6703 {
6704         struct ctl_lun *lun = CTL_LUN(ctsio);
6705         struct scsi_log_informational_exceptions *data;
6706         const char *value;
6707
6708         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6709
6710         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6711         data->hdr.param_control = SLP_LBIN;
6712         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6713             sizeof(struct scsi_log_param_header);
6714         data->ie_asc = lun->ie_asc;
6715         data->ie_ascq = lun->ie_ascq;
6716         if ((value = dnvlist_get_string(lun->be_lun->options, "temperature",
6717             NULL)) != NULL)
6718                 data->temperature = strtol(value, NULL, 0);
6719         else
6720                 data->temperature = 0xff;
6721         return (0);
6722 }
6723
6724 int
6725 ctl_log_sense(struct ctl_scsiio *ctsio)
6726 {
6727         struct ctl_lun *lun = CTL_LUN(ctsio);
6728         int i, pc, page_code, subpage;
6729         int alloc_len, total_len;
6730         struct ctl_page_index *page_index;
6731         struct scsi_log_sense *cdb;
6732         struct scsi_log_header *header;
6733
6734         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6735
6736         cdb = (struct scsi_log_sense *)ctsio->cdb;
6737         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6738         page_code = cdb->page & SLS_PAGE_CODE;
6739         subpage = cdb->subpage;
6740         alloc_len = scsi_2btoul(cdb->length);
6741
6742         page_index = NULL;
6743         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6744                 page_index = &lun->log_pages.index[i];
6745
6746                 /* Look for the right page code */
6747                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6748                         continue;
6749
6750                 /* Look for the right subpage or the subpage wildcard*/
6751                 if (page_index->subpage != subpage)
6752                         continue;
6753
6754                 break;
6755         }
6756         if (i >= CTL_NUM_LOG_PAGES) {
6757                 ctl_set_invalid_field(ctsio,
6758                                       /*sks_valid*/ 1,
6759                                       /*command*/ 1,
6760                                       /*field*/ 2,
6761                                       /*bit_valid*/ 0,
6762                                       /*bit*/ 0);
6763                 ctl_done((union ctl_io *)ctsio);
6764                 return (CTL_RETVAL_COMPLETE);
6765         }
6766
6767         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6768
6769         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6770         ctsio->kern_sg_entries = 0;
6771         ctsio->kern_rel_offset = 0;
6772         ctsio->kern_data_len = min(total_len, alloc_len);
6773         ctsio->kern_total_len = ctsio->kern_data_len;
6774
6775         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6776         header->page = page_index->page_code;
6777         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6778                 header->page |= SL_DS;
6779         if (page_index->subpage) {
6780                 header->page |= SL_SPF;
6781                 header->subpage = page_index->subpage;
6782         }
6783         scsi_ulto2b(page_index->page_len, header->datalen);
6784
6785         /*
6786          * Call the handler, if it exists, to update the
6787          * page to the latest values.
6788          */
6789         if (page_index->sense_handler != NULL)
6790                 page_index->sense_handler(ctsio, page_index, pc);
6791
6792         memcpy(header + 1, page_index->page_data, page_index->page_len);
6793
6794         ctl_set_success(ctsio);
6795         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6796         ctsio->be_move_done = ctl_config_move_done;
6797         ctl_datamove((union ctl_io *)ctsio);
6798         return (CTL_RETVAL_COMPLETE);
6799 }
6800
6801 int
6802 ctl_read_capacity(struct ctl_scsiio *ctsio)
6803 {
6804         struct ctl_lun *lun = CTL_LUN(ctsio);
6805         struct scsi_read_capacity *cdb;
6806         struct scsi_read_capacity_data *data;
6807         uint32_t lba;
6808
6809         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6810
6811         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6812
6813         lba = scsi_4btoul(cdb->addr);
6814         if (((cdb->pmi & SRC_PMI) == 0)
6815          && (lba != 0)) {
6816                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6817                                       /*sks_valid*/ 1,
6818                                       /*command*/ 1,
6819                                       /*field*/ 2,
6820                                       /*bit_valid*/ 0,
6821                                       /*bit*/ 0);
6822                 ctl_done((union ctl_io *)ctsio);
6823                 return (CTL_RETVAL_COMPLETE);
6824         }
6825
6826         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6827         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6828         ctsio->kern_data_len = sizeof(*data);
6829         ctsio->kern_total_len = sizeof(*data);
6830         ctsio->kern_rel_offset = 0;
6831         ctsio->kern_sg_entries = 0;
6832
6833         /*
6834          * If the maximum LBA is greater than 0xfffffffe, the user must
6835          * issue a SERVICE ACTION IN (16) command, with the read capacity
6836          * serivce action set.
6837          */
6838         if (lun->be_lun->maxlba > 0xfffffffe)
6839                 scsi_ulto4b(0xffffffff, data->addr);
6840         else
6841                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6842
6843         /*
6844          * XXX KDM this may not be 512 bytes...
6845          */
6846         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6847
6848         ctl_set_success(ctsio);
6849         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6850         ctsio->be_move_done = ctl_config_move_done;
6851         ctl_datamove((union ctl_io *)ctsio);
6852         return (CTL_RETVAL_COMPLETE);
6853 }
6854
6855 int
6856 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6857 {
6858         struct ctl_lun *lun = CTL_LUN(ctsio);
6859         struct scsi_read_capacity_16 *cdb;
6860         struct scsi_read_capacity_data_long *data;
6861         uint64_t lba;
6862         uint32_t alloc_len;
6863
6864         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6865
6866         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6867
6868         alloc_len = scsi_4btoul(cdb->alloc_len);
6869         lba = scsi_8btou64(cdb->addr);
6870
6871         if ((cdb->reladr & SRC16_PMI)
6872          && (lba != 0)) {
6873                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6874                                       /*sks_valid*/ 1,
6875                                       /*command*/ 1,
6876                                       /*field*/ 2,
6877                                       /*bit_valid*/ 0,
6878                                       /*bit*/ 0);
6879                 ctl_done((union ctl_io *)ctsio);
6880                 return (CTL_RETVAL_COMPLETE);
6881         }
6882
6883         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6884         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6885         ctsio->kern_rel_offset = 0;
6886         ctsio->kern_sg_entries = 0;
6887         ctsio->kern_data_len = min(sizeof(*data), alloc_len);
6888         ctsio->kern_total_len = ctsio->kern_data_len;
6889
6890         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6891         /* XXX KDM this may not be 512 bytes... */
6892         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6893         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6894         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6895         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6896                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6897
6898         ctl_set_success(ctsio);
6899         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6900         ctsio->be_move_done = ctl_config_move_done;
6901         ctl_datamove((union ctl_io *)ctsio);
6902         return (CTL_RETVAL_COMPLETE);
6903 }
6904
6905 int
6906 ctl_get_lba_status(struct ctl_scsiio *ctsio)
6907 {
6908         struct ctl_lun *lun = CTL_LUN(ctsio);
6909         struct scsi_get_lba_status *cdb;
6910         struct scsi_get_lba_status_data *data;
6911         struct ctl_lba_len_flags *lbalen;
6912         uint64_t lba;
6913         uint32_t alloc_len, total_len;
6914         int retval;
6915
6916         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
6917
6918         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
6919         lba = scsi_8btou64(cdb->addr);
6920         alloc_len = scsi_4btoul(cdb->alloc_len);
6921
6922         if (lba > lun->be_lun->maxlba) {
6923                 ctl_set_lba_out_of_range(ctsio, lba);
6924                 ctl_done((union ctl_io *)ctsio);
6925                 return (CTL_RETVAL_COMPLETE);
6926         }
6927
6928         total_len = sizeof(*data) + sizeof(data->descr[0]);
6929         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6930         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
6931         ctsio->kern_rel_offset = 0;
6932         ctsio->kern_sg_entries = 0;
6933         ctsio->kern_data_len = min(total_len, alloc_len);
6934         ctsio->kern_total_len = ctsio->kern_data_len;
6935
6936         /* Fill dummy data in case backend can't tell anything. */
6937         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
6938         scsi_u64to8b(lba, data->descr[0].addr);
6939         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
6940             data->descr[0].length);
6941         data->descr[0].status = 0; /* Mapped or unknown. */
6942
6943         ctl_set_success(ctsio);
6944         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6945         ctsio->be_move_done = ctl_config_move_done;
6946
6947         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6948         lbalen->lba = lba;
6949         lbalen->len = total_len;
6950         lbalen->flags = 0;
6951         retval = lun->backend->config_read((union ctl_io *)ctsio);
6952         return (retval);
6953 }
6954
6955 int
6956 ctl_read_defect(struct ctl_scsiio *ctsio)
6957 {
6958         struct scsi_read_defect_data_10 *ccb10;
6959         struct scsi_read_defect_data_12 *ccb12;
6960         struct scsi_read_defect_data_hdr_10 *data10;
6961         struct scsi_read_defect_data_hdr_12 *data12;
6962         uint32_t alloc_len, data_len;
6963         uint8_t format;
6964
6965         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
6966
6967         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
6968                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
6969                 format = ccb10->format;
6970                 alloc_len = scsi_2btoul(ccb10->alloc_length);
6971                 data_len = sizeof(*data10);
6972         } else {
6973                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
6974                 format = ccb12->format;
6975                 alloc_len = scsi_4btoul(ccb12->alloc_length);
6976                 data_len = sizeof(*data12);
6977         }
6978         if (alloc_len == 0) {
6979                 ctl_set_success(ctsio);
6980                 ctl_done((union ctl_io *)ctsio);
6981                 return (CTL_RETVAL_COMPLETE);
6982         }
6983
6984         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
6985         ctsio->kern_rel_offset = 0;
6986         ctsio->kern_sg_entries = 0;
6987         ctsio->kern_data_len = min(data_len, alloc_len);
6988         ctsio->kern_total_len = ctsio->kern_data_len;
6989
6990         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
6991                 data10 = (struct scsi_read_defect_data_hdr_10 *)
6992                     ctsio->kern_data_ptr;
6993                 data10->format = format;
6994                 scsi_ulto2b(0, data10->length);
6995         } else {
6996                 data12 = (struct scsi_read_defect_data_hdr_12 *)
6997                     ctsio->kern_data_ptr;
6998                 data12->format = format;
6999                 scsi_ulto2b(0, data12->generation);
7000                 scsi_ulto4b(0, data12->length);
7001         }
7002
7003         ctl_set_success(ctsio);
7004         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7005         ctsio->be_move_done = ctl_config_move_done;
7006         ctl_datamove((union ctl_io *)ctsio);
7007         return (CTL_RETVAL_COMPLETE);
7008 }
7009
7010 int
7011 ctl_report_ident_info(struct ctl_scsiio *ctsio)
7012 {
7013         struct ctl_lun *lun = CTL_LUN(ctsio);
7014         struct scsi_report_ident_info *cdb;
7015         struct scsi_report_ident_info_data *rii_ptr;
7016         struct scsi_report_ident_info_descr *riid_ptr;
7017         const char *oii, *otii;
7018         int retval, alloc_len, total_len = 0, len = 0;
7019
7020         CTL_DEBUG_PRINT(("ctl_report_ident_info\n"));
7021
7022         cdb = (struct scsi_report_ident_info *)ctsio->cdb;
7023         retval = CTL_RETVAL_COMPLETE;
7024
7025         total_len = sizeof(struct scsi_report_ident_info_data);
7026         switch (cdb->type) {
7027         case RII_LUII:
7028                 oii = dnvlist_get_string(lun->be_lun->options,
7029                     "ident_info", NULL);
7030                 if (oii)
7031                         len = strlen(oii);      /* Approximately */
7032                 break;
7033         case RII_LUTII:
7034                 otii = dnvlist_get_string(lun->be_lun->options,
7035                     "text_ident_info", NULL);
7036                 if (otii)
7037                         len = strlen(otii) + 1; /* NULL-terminated */
7038                 break;
7039         case RII_IIS:
7040                 len = 2 * sizeof(struct scsi_report_ident_info_descr);
7041                 break;
7042         default:
7043                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7044                                       /*sks_valid*/ 1,
7045                                       /*command*/ 1,
7046                                       /*field*/ 11,
7047                                       /*bit_valid*/ 1,
7048                                       /*bit*/ 2);
7049                 ctl_done((union ctl_io *)ctsio);
7050                 return(retval);
7051         }
7052         total_len += len;
7053         alloc_len = scsi_4btoul(cdb->length);
7054
7055         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7056         ctsio->kern_sg_entries = 0;
7057         ctsio->kern_rel_offset = 0;
7058         ctsio->kern_data_len = min(total_len, alloc_len);
7059         ctsio->kern_total_len = ctsio->kern_data_len;
7060
7061         rii_ptr = (struct scsi_report_ident_info_data *)ctsio->kern_data_ptr;
7062         switch (cdb->type) {
7063         case RII_LUII:
7064                 if (oii) {
7065                         if (oii[0] == '0' && oii[1] == 'x')
7066                                 len = hex2bin(oii, (uint8_t *)(rii_ptr + 1), len);
7067                         else
7068                                 strncpy((uint8_t *)(rii_ptr + 1), oii, len);
7069                 }
7070                 break;
7071         case RII_LUTII:
7072                 if (otii)
7073                         strlcpy((uint8_t *)(rii_ptr + 1), otii, len);
7074                 break;
7075         case RII_IIS:
7076                 riid_ptr = (struct scsi_report_ident_info_descr *)(rii_ptr + 1);
7077                 riid_ptr->type = RII_LUII;
7078                 scsi_ulto2b(0xffff, riid_ptr->length);
7079                 riid_ptr++;
7080                 riid_ptr->type = RII_LUTII;
7081                 scsi_ulto2b(0xffff, riid_ptr->length);
7082         }
7083         scsi_ulto2b(len, rii_ptr->length);
7084
7085         ctl_set_success(ctsio);
7086         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7087         ctsio->be_move_done = ctl_config_move_done;
7088         ctl_datamove((union ctl_io *)ctsio);
7089         return(retval);
7090 }
7091
7092 int
7093 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7094 {
7095         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7096         struct ctl_lun *lun = CTL_LUN(ctsio);
7097         struct scsi_maintenance_in *cdb;
7098         int retval;
7099         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7100         int num_ha_groups, num_target_ports, shared_group;
7101         struct ctl_port *port;
7102         struct scsi_target_group_data *rtg_ptr;
7103         struct scsi_target_group_data_extended *rtg_ext_ptr;
7104         struct scsi_target_port_group_descriptor *tpg_desc;
7105
7106         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7107
7108         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7109         retval = CTL_RETVAL_COMPLETE;
7110
7111         switch (cdb->byte2 & STG_PDF_MASK) {
7112         case STG_PDF_LENGTH:
7113                 ext = 0;
7114                 break;
7115         case STG_PDF_EXTENDED:
7116                 ext = 1;
7117                 break;
7118         default:
7119                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7120                                       /*sks_valid*/ 1,
7121                                       /*command*/ 1,
7122                                       /*field*/ 2,
7123                                       /*bit_valid*/ 1,
7124                                       /*bit*/ 5);
7125                 ctl_done((union ctl_io *)ctsio);
7126                 return(retval);
7127         }
7128
7129         num_target_ports = 0;
7130         shared_group = (softc->is_single != 0);
7131         mtx_lock(&softc->ctl_lock);
7132         STAILQ_FOREACH(port, &softc->port_list, links) {
7133                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7134                         continue;
7135                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7136                         continue;
7137                 num_target_ports++;
7138                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7139                         shared_group = 1;
7140         }
7141         mtx_unlock(&softc->ctl_lock);
7142         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7143
7144         if (ext)
7145                 total_len = sizeof(struct scsi_target_group_data_extended);
7146         else
7147                 total_len = sizeof(struct scsi_target_group_data);
7148         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7149                 (shared_group + num_ha_groups) +
7150             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7151
7152         alloc_len = scsi_4btoul(cdb->length);
7153
7154         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7155         ctsio->kern_sg_entries = 0;
7156         ctsio->kern_rel_offset = 0;
7157         ctsio->kern_data_len = min(total_len, alloc_len);
7158         ctsio->kern_total_len = ctsio->kern_data_len;
7159
7160         if (ext) {
7161                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7162                     ctsio->kern_data_ptr;
7163                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7164                 rtg_ext_ptr->format_type = 0x10;
7165                 rtg_ext_ptr->implicit_transition_time = 0;
7166                 tpg_desc = &rtg_ext_ptr->groups[0];
7167         } else {
7168                 rtg_ptr = (struct scsi_target_group_data *)
7169                     ctsio->kern_data_ptr;
7170                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7171                 tpg_desc = &rtg_ptr->groups[0];
7172         }
7173
7174         mtx_lock(&softc->ctl_lock);
7175         pg = softc->port_min / softc->port_cnt;
7176         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7177                 /* Some shelf is known to be primary. */
7178                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7179                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7180                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7181                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7182                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7183                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7184                 else
7185                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7186                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7187                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7188                 } else {
7189                         ts = os;
7190                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7191                 }
7192         } else {
7193                 /* No known primary shelf. */
7194                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7195                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7196                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7197                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7198                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7199                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7200                 } else {
7201                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7202                 }
7203         }
7204         if (shared_group) {
7205                 tpg_desc->pref_state = ts;
7206                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7207                     TPG_U_SUP | TPG_T_SUP;
7208                 scsi_ulto2b(1, tpg_desc->target_port_group);
7209                 tpg_desc->status = TPG_IMPLICIT;
7210                 pc = 0;
7211                 STAILQ_FOREACH(port, &softc->port_list, links) {
7212                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7213                                 continue;
7214                         if (!softc->is_single &&
7215                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7216                                 continue;
7217                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7218                                 continue;
7219                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7220                             relative_target_port_identifier);
7221                         pc++;
7222                 }
7223                 tpg_desc->target_port_count = pc;
7224                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7225                     &tpg_desc->descriptors[pc];
7226         }
7227         for (g = 0; g < num_ha_groups; g++) {
7228                 tpg_desc->pref_state = (g == pg) ? ts : os;
7229                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7230                     TPG_U_SUP | TPG_T_SUP;
7231                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7232                 tpg_desc->status = TPG_IMPLICIT;
7233                 pc = 0;
7234                 STAILQ_FOREACH(port, &softc->port_list, links) {
7235                         if (port->targ_port < g * softc->port_cnt ||
7236                             port->targ_port >= (g + 1) * softc->port_cnt)
7237                                 continue;
7238                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7239                                 continue;
7240                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7241                                 continue;
7242                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7243                                 continue;
7244                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7245                             relative_target_port_identifier);
7246                         pc++;
7247                 }
7248                 tpg_desc->target_port_count = pc;
7249                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7250                     &tpg_desc->descriptors[pc];
7251         }
7252         mtx_unlock(&softc->ctl_lock);
7253
7254         ctl_set_success(ctsio);
7255         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7256         ctsio->be_move_done = ctl_config_move_done;
7257         ctl_datamove((union ctl_io *)ctsio);
7258         return(retval);
7259 }
7260
7261 int
7262 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7263 {
7264         struct ctl_lun *lun = CTL_LUN(ctsio);
7265         struct scsi_report_supported_opcodes *cdb;
7266         const struct ctl_cmd_entry *entry, *sentry;
7267         struct scsi_report_supported_opcodes_all *all;
7268         struct scsi_report_supported_opcodes_descr *descr;
7269         struct scsi_report_supported_opcodes_one *one;
7270         int retval;
7271         int alloc_len, total_len;
7272         int opcode, service_action, i, j, num;
7273
7274         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7275
7276         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7277         retval = CTL_RETVAL_COMPLETE;
7278
7279         opcode = cdb->requested_opcode;
7280         service_action = scsi_2btoul(cdb->requested_service_action);
7281         switch (cdb->options & RSO_OPTIONS_MASK) {
7282         case RSO_OPTIONS_ALL:
7283                 num = 0;
7284                 for (i = 0; i < 256; i++) {
7285                         entry = &ctl_cmd_table[i];
7286                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7287                                 for (j = 0; j < 32; j++) {
7288                                         sentry = &((const struct ctl_cmd_entry *)
7289                                             entry->execute)[j];
7290                                         if (ctl_cmd_applicable(
7291                                             lun->be_lun->lun_type, sentry))
7292                                                 num++;
7293                                 }
7294                         } else {
7295                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7296                                     entry))
7297                                         num++;
7298                         }
7299                 }
7300                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7301                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7302                 break;
7303         case RSO_OPTIONS_OC:
7304                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7305                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7306                                               /*sks_valid*/ 1,
7307                                               /*command*/ 1,
7308                                               /*field*/ 2,
7309                                               /*bit_valid*/ 1,
7310                                               /*bit*/ 2);
7311                         ctl_done((union ctl_io *)ctsio);
7312                         return (CTL_RETVAL_COMPLETE);
7313                 }
7314                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7315                 break;
7316         case RSO_OPTIONS_OC_SA:
7317                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7318                     service_action >= 32) {
7319                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7320                                               /*sks_valid*/ 1,
7321                                               /*command*/ 1,
7322                                               /*field*/ 2,
7323                                               /*bit_valid*/ 1,
7324                                               /*bit*/ 2);
7325                         ctl_done((union ctl_io *)ctsio);
7326                         return (CTL_RETVAL_COMPLETE);
7327                 }
7328                 /* FALLTHROUGH */
7329         case RSO_OPTIONS_OC_ASA:
7330                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7331                 break;
7332         default:
7333                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7334                                       /*sks_valid*/ 1,
7335                                       /*command*/ 1,
7336                                       /*field*/ 2,
7337                                       /*bit_valid*/ 1,
7338                                       /*bit*/ 2);
7339                 ctl_done((union ctl_io *)ctsio);
7340                 return (CTL_RETVAL_COMPLETE);
7341         }
7342
7343         alloc_len = scsi_4btoul(cdb->length);
7344
7345         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7346         ctsio->kern_sg_entries = 0;
7347         ctsio->kern_rel_offset = 0;
7348         ctsio->kern_data_len = min(total_len, alloc_len);
7349         ctsio->kern_total_len = ctsio->kern_data_len;
7350
7351         switch (cdb->options & RSO_OPTIONS_MASK) {
7352         case RSO_OPTIONS_ALL:
7353                 all = (struct scsi_report_supported_opcodes_all *)
7354                     ctsio->kern_data_ptr;
7355                 num = 0;
7356                 for (i = 0; i < 256; i++) {
7357                         entry = &ctl_cmd_table[i];
7358                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7359                                 for (j = 0; j < 32; j++) {
7360                                         sentry = &((const struct ctl_cmd_entry *)
7361                                             entry->execute)[j];
7362                                         if (!ctl_cmd_applicable(
7363                                             lun->be_lun->lun_type, sentry))
7364                                                 continue;
7365                                         descr = &all->descr[num++];
7366                                         descr->opcode = i;
7367                                         scsi_ulto2b(j, descr->service_action);
7368                                         descr->flags = RSO_SERVACTV;
7369                                         scsi_ulto2b(sentry->length,
7370                                             descr->cdb_length);
7371                                 }
7372                         } else {
7373                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7374                                     entry))
7375                                         continue;
7376                                 descr = &all->descr[num++];
7377                                 descr->opcode = i;
7378                                 scsi_ulto2b(0, descr->service_action);
7379                                 descr->flags = 0;
7380                                 scsi_ulto2b(entry->length, descr->cdb_length);
7381                         }
7382                 }
7383                 scsi_ulto4b(
7384                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7385                     all->length);
7386                 break;
7387         case RSO_OPTIONS_OC:
7388                 one = (struct scsi_report_supported_opcodes_one *)
7389                     ctsio->kern_data_ptr;
7390                 entry = &ctl_cmd_table[opcode];
7391                 goto fill_one;
7392         case RSO_OPTIONS_OC_SA:
7393                 one = (struct scsi_report_supported_opcodes_one *)
7394                     ctsio->kern_data_ptr;
7395                 entry = &ctl_cmd_table[opcode];
7396                 entry = &((const struct ctl_cmd_entry *)
7397                     entry->execute)[service_action];
7398 fill_one:
7399                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7400                         one->support = 3;
7401                         scsi_ulto2b(entry->length, one->cdb_length);
7402                         one->cdb_usage[0] = opcode;
7403                         memcpy(&one->cdb_usage[1], entry->usage,
7404                             entry->length - 1);
7405                 } else
7406                         one->support = 1;
7407                 break;
7408         case RSO_OPTIONS_OC_ASA:
7409                 one = (struct scsi_report_supported_opcodes_one *)
7410                     ctsio->kern_data_ptr;
7411                 entry = &ctl_cmd_table[opcode];
7412                 if (entry->flags & CTL_CMD_FLAG_SA5) {
7413                         entry = &((const struct ctl_cmd_entry *)
7414                             entry->execute)[service_action];
7415                 } else if (service_action != 0) {
7416                         one->support = 1;
7417                         break;
7418                 }
7419                 goto fill_one;
7420         }
7421
7422         ctl_set_success(ctsio);
7423         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7424         ctsio->be_move_done = ctl_config_move_done;
7425         ctl_datamove((union ctl_io *)ctsio);
7426         return(retval);
7427 }
7428
7429 int
7430 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7431 {
7432         struct scsi_report_supported_tmf *cdb;
7433         struct scsi_report_supported_tmf_ext_data *data;
7434         int retval;
7435         int alloc_len, total_len;
7436
7437         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7438
7439         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7440
7441         retval = CTL_RETVAL_COMPLETE;
7442
7443         if (cdb->options & RST_REPD)
7444                 total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7445         else
7446                 total_len = sizeof(struct scsi_report_supported_tmf_data);
7447         alloc_len = scsi_4btoul(cdb->length);
7448
7449         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7450         ctsio->kern_sg_entries = 0;
7451         ctsio->kern_rel_offset = 0;
7452         ctsio->kern_data_len = min(total_len, alloc_len);
7453         ctsio->kern_total_len = ctsio->kern_data_len;
7454
7455         data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7456         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7457             RST_TRS;
7458         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7459         data->length = total_len - 4;
7460
7461         ctl_set_success(ctsio);
7462         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7463         ctsio->be_move_done = ctl_config_move_done;
7464         ctl_datamove((union ctl_io *)ctsio);
7465         return (retval);
7466 }
7467
7468 int
7469 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7470 {
7471         struct scsi_report_timestamp *cdb;
7472         struct scsi_report_timestamp_data *data;
7473         struct timeval tv;
7474         int64_t timestamp;
7475         int retval;
7476         int alloc_len, total_len;
7477
7478         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7479
7480         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7481
7482         retval = CTL_RETVAL_COMPLETE;
7483
7484         total_len = sizeof(struct scsi_report_timestamp_data);
7485         alloc_len = scsi_4btoul(cdb->length);
7486
7487         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7488         ctsio->kern_sg_entries = 0;
7489         ctsio->kern_rel_offset = 0;
7490         ctsio->kern_data_len = min(total_len, alloc_len);
7491         ctsio->kern_total_len = ctsio->kern_data_len;
7492
7493         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7494         scsi_ulto2b(sizeof(*data) - 2, data->length);
7495         data->origin = RTS_ORIG_OUTSIDE;
7496         getmicrotime(&tv);
7497         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7498         scsi_ulto4b(timestamp >> 16, data->timestamp);
7499         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7500
7501         ctl_set_success(ctsio);
7502         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7503         ctsio->be_move_done = ctl_config_move_done;
7504         ctl_datamove((union ctl_io *)ctsio);
7505         return (retval);
7506 }
7507
7508 int
7509 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7510 {
7511         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7512         struct ctl_lun *lun = CTL_LUN(ctsio);
7513         struct scsi_per_res_in *cdb;
7514         int alloc_len, total_len = 0;
7515         /* struct scsi_per_res_in_rsrv in_data; */
7516         uint64_t key;
7517
7518         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7519
7520         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7521
7522         alloc_len = scsi_2btoul(cdb->length);
7523
7524 retry:
7525         mtx_lock(&lun->lun_lock);
7526         switch (cdb->action) {
7527         case SPRI_RK: /* read keys */
7528                 total_len = sizeof(struct scsi_per_res_in_keys) +
7529                         lun->pr_key_count *
7530                         sizeof(struct scsi_per_res_key);
7531                 break;
7532         case SPRI_RR: /* read reservation */
7533                 if (lun->flags & CTL_LUN_PR_RESERVED)
7534                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7535                 else
7536                         total_len = sizeof(struct scsi_per_res_in_header);
7537                 break;
7538         case SPRI_RC: /* report capabilities */
7539                 total_len = sizeof(struct scsi_per_res_cap);
7540                 break;
7541         case SPRI_RS: /* read full status */
7542                 total_len = sizeof(struct scsi_per_res_in_header) +
7543                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7544                     lun->pr_key_count;
7545                 break;
7546         default:
7547                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7548         }
7549         mtx_unlock(&lun->lun_lock);
7550
7551         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7552         ctsio->kern_rel_offset = 0;
7553         ctsio->kern_sg_entries = 0;
7554         ctsio->kern_data_len = min(total_len, alloc_len);
7555         ctsio->kern_total_len = ctsio->kern_data_len;
7556
7557         mtx_lock(&lun->lun_lock);
7558         switch (cdb->action) {
7559         case SPRI_RK: { // read keys
7560         struct scsi_per_res_in_keys *res_keys;
7561                 int i, key_count;
7562
7563                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7564
7565                 /*
7566                  * We had to drop the lock to allocate our buffer, which
7567                  * leaves time for someone to come in with another
7568                  * persistent reservation.  (That is unlikely, though,
7569                  * since this should be the only persistent reservation
7570                  * command active right now.)
7571                  */
7572                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7573                     (lun->pr_key_count *
7574                      sizeof(struct scsi_per_res_key)))){
7575                         mtx_unlock(&lun->lun_lock);
7576                         free(ctsio->kern_data_ptr, M_CTL);
7577                         printf("%s: reservation length changed, retrying\n",
7578                                __func__);
7579                         goto retry;
7580                 }
7581
7582                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7583
7584                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7585                              lun->pr_key_count, res_keys->header.length);
7586
7587                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7588                         if ((key = ctl_get_prkey(lun, i)) == 0)
7589                                 continue;
7590
7591                         /*
7592                          * We used lun->pr_key_count to calculate the
7593                          * size to allocate.  If it turns out the number of
7594                          * initiators with the registered flag set is
7595                          * larger than that (i.e. they haven't been kept in
7596                          * sync), we've got a problem.
7597                          */
7598                         if (key_count >= lun->pr_key_count) {
7599                                 key_count++;
7600                                 continue;
7601                         }
7602                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7603                         key_count++;
7604                 }
7605                 break;
7606         }
7607         case SPRI_RR: { // read reservation
7608                 struct scsi_per_res_in_rsrv *res;
7609                 int tmp_len, header_only;
7610
7611                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7612
7613                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7614
7615                 if (lun->flags & CTL_LUN_PR_RESERVED)
7616                 {
7617                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7618                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7619                                     res->header.length);
7620                         header_only = 0;
7621                 } else {
7622                         tmp_len = sizeof(struct scsi_per_res_in_header);
7623                         scsi_ulto4b(0, res->header.length);
7624                         header_only = 1;
7625                 }
7626
7627                 /*
7628                  * We had to drop the lock to allocate our buffer, which
7629                  * leaves time for someone to come in with another
7630                  * persistent reservation.  (That is unlikely, though,
7631                  * since this should be the only persistent reservation
7632                  * command active right now.)
7633                  */
7634                 if (tmp_len != total_len) {
7635                         mtx_unlock(&lun->lun_lock);
7636                         free(ctsio->kern_data_ptr, M_CTL);
7637                         printf("%s: reservation status changed, retrying\n",
7638                                __func__);
7639                         goto retry;
7640                 }
7641
7642                 /*
7643                  * No reservation held, so we're done.
7644                  */
7645                 if (header_only != 0)
7646                         break;
7647
7648                 /*
7649                  * If the registration is an All Registrants type, the key
7650                  * is 0, since it doesn't really matter.
7651                  */
7652                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7653                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7654                             res->data.reservation);
7655                 }
7656                 res->data.scopetype = lun->pr_res_type;
7657                 break;
7658         }
7659         case SPRI_RC:     //report capabilities
7660         {
7661                 struct scsi_per_res_cap *res_cap;
7662                 uint16_t type_mask;
7663
7664                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7665                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7666                 res_cap->flags1 = SPRI_CRH;
7667                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7668                 type_mask = SPRI_TM_WR_EX_AR |
7669                             SPRI_TM_EX_AC_RO |
7670                             SPRI_TM_WR_EX_RO |
7671                             SPRI_TM_EX_AC |
7672                             SPRI_TM_WR_EX |
7673                             SPRI_TM_EX_AC_AR;
7674                 scsi_ulto2b(type_mask, res_cap->type_mask);
7675                 break;
7676         }
7677         case SPRI_RS: { // read full status
7678                 struct scsi_per_res_in_full *res_status;
7679                 struct scsi_per_res_in_full_desc *res_desc;
7680                 struct ctl_port *port;
7681                 int i, len;
7682
7683                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7684
7685                 /*
7686                  * We had to drop the lock to allocate our buffer, which
7687                  * leaves time for someone to come in with another
7688                  * persistent reservation.  (That is unlikely, though,
7689                  * since this should be the only persistent reservation
7690                  * command active right now.)
7691                  */
7692                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7693                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7694                      lun->pr_key_count)){
7695                         mtx_unlock(&lun->lun_lock);
7696                         free(ctsio->kern_data_ptr, M_CTL);
7697                         printf("%s: reservation length changed, retrying\n",
7698                                __func__);
7699                         goto retry;
7700                 }
7701
7702                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7703
7704                 res_desc = &res_status->desc[0];
7705                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7706                         if ((key = ctl_get_prkey(lun, i)) == 0)
7707                                 continue;
7708
7709                         scsi_u64to8b(key, res_desc->res_key.key);
7710                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7711                             (lun->pr_res_idx == i ||
7712                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7713                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7714                                 res_desc->scopetype = lun->pr_res_type;
7715                         }
7716                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7717                             res_desc->rel_trgt_port_id);
7718                         len = 0;
7719                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7720                         if (port != NULL)
7721                                 len = ctl_create_iid(port,
7722                                     i % CTL_MAX_INIT_PER_PORT,
7723                                     res_desc->transport_id);
7724                         scsi_ulto4b(len, res_desc->additional_length);
7725                         res_desc = (struct scsi_per_res_in_full_desc *)
7726                             &res_desc->transport_id[len];
7727                 }
7728                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7729                     res_status->header.length);
7730                 break;
7731         }
7732         default:
7733                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7734         }
7735         mtx_unlock(&lun->lun_lock);
7736
7737         ctl_set_success(ctsio);
7738         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7739         ctsio->be_move_done = ctl_config_move_done;
7740         ctl_datamove((union ctl_io *)ctsio);
7741         return (CTL_RETVAL_COMPLETE);
7742 }
7743
7744 /*
7745  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7746  * it should return.
7747  */
7748 static int
7749 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7750                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7751                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7752                 struct scsi_per_res_out_parms* param)
7753 {
7754         union ctl_ha_msg persis_io;
7755         int i;
7756
7757         mtx_lock(&lun->lun_lock);
7758         if (sa_res_key == 0) {
7759                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7760                         /* validate scope and type */
7761                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7762                              SPR_LU_SCOPE) {
7763                                 mtx_unlock(&lun->lun_lock);
7764                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7765                                                       /*sks_valid*/ 1,
7766                                                       /*command*/ 1,
7767                                                       /*field*/ 2,
7768                                                       /*bit_valid*/ 1,
7769                                                       /*bit*/ 4);
7770                                 ctl_done((union ctl_io *)ctsio);
7771                                 return (1);
7772                         }
7773
7774                         if (type>8 || type==2 || type==4 || type==0) {
7775                                 mtx_unlock(&lun->lun_lock);
7776                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7777                                                       /*sks_valid*/ 1,
7778                                                       /*command*/ 1,
7779                                                       /*field*/ 2,
7780                                                       /*bit_valid*/ 1,
7781                                                       /*bit*/ 0);
7782                                 ctl_done((union ctl_io *)ctsio);
7783                                 return (1);
7784                         }
7785
7786                         /*
7787                          * Unregister everybody else and build UA for
7788                          * them
7789                          */
7790                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7791                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7792                                         continue;
7793
7794                                 ctl_clr_prkey(lun, i);
7795                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7796                         }
7797                         lun->pr_key_count = 1;
7798                         lun->pr_res_type = type;
7799                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7800                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7801                                 lun->pr_res_idx = residx;
7802                         lun->pr_generation++;
7803                         mtx_unlock(&lun->lun_lock);
7804
7805                         /* send msg to other side */
7806                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7807                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7808                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7809                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7810                         persis_io.pr.pr_info.res_type = type;
7811                         memcpy(persis_io.pr.pr_info.sa_res_key,
7812                                param->serv_act_res_key,
7813                                sizeof(param->serv_act_res_key));
7814                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7815                             sizeof(persis_io.pr), M_WAITOK);
7816                 } else {
7817                         /* not all registrants */
7818                         mtx_unlock(&lun->lun_lock);
7819                         free(ctsio->kern_data_ptr, M_CTL);
7820                         ctl_set_invalid_field(ctsio,
7821                                               /*sks_valid*/ 1,
7822                                               /*command*/ 0,
7823                                               /*field*/ 8,
7824                                               /*bit_valid*/ 0,
7825                                               /*bit*/ 0);
7826                         ctl_done((union ctl_io *)ctsio);
7827                         return (1);
7828                 }
7829         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7830                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7831                 int found = 0;
7832
7833                 if (res_key == sa_res_key) {
7834                         /* special case */
7835                         /*
7836                          * The spec implies this is not good but doesn't
7837                          * say what to do. There are two choices either
7838                          * generate a res conflict or check condition
7839                          * with illegal field in parameter data. Since
7840                          * that is what is done when the sa_res_key is
7841                          * zero I'll take that approach since this has
7842                          * to do with the sa_res_key.
7843                          */
7844                         mtx_unlock(&lun->lun_lock);
7845                         free(ctsio->kern_data_ptr, M_CTL);
7846                         ctl_set_invalid_field(ctsio,
7847                                               /*sks_valid*/ 1,
7848                                               /*command*/ 0,
7849                                               /*field*/ 8,
7850                                               /*bit_valid*/ 0,
7851                                               /*bit*/ 0);
7852                         ctl_done((union ctl_io *)ctsio);
7853                         return (1);
7854                 }
7855
7856                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7857                         if (ctl_get_prkey(lun, i) != sa_res_key)
7858                                 continue;
7859
7860                         found = 1;
7861                         ctl_clr_prkey(lun, i);
7862                         lun->pr_key_count--;
7863                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7864                 }
7865                 if (!found) {
7866                         mtx_unlock(&lun->lun_lock);
7867                         free(ctsio->kern_data_ptr, M_CTL);
7868                         ctl_set_reservation_conflict(ctsio);
7869                         ctl_done((union ctl_io *)ctsio);
7870                         return (CTL_RETVAL_COMPLETE);
7871                 }
7872                 lun->pr_generation++;
7873                 mtx_unlock(&lun->lun_lock);
7874
7875                 /* send msg to other side */
7876                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7877                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7878                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7879                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7880                 persis_io.pr.pr_info.res_type = type;
7881                 memcpy(persis_io.pr.pr_info.sa_res_key,
7882                        param->serv_act_res_key,
7883                        sizeof(param->serv_act_res_key));
7884                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7885                     sizeof(persis_io.pr), M_WAITOK);
7886         } else {
7887                 /* Reserved but not all registrants */
7888                 /* sa_res_key is res holder */
7889                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7890                         /* validate scope and type */
7891                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7892                              SPR_LU_SCOPE) {
7893                                 mtx_unlock(&lun->lun_lock);
7894                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7895                                                       /*sks_valid*/ 1,
7896                                                       /*command*/ 1,
7897                                                       /*field*/ 2,
7898                                                       /*bit_valid*/ 1,
7899                                                       /*bit*/ 4);
7900                                 ctl_done((union ctl_io *)ctsio);
7901                                 return (1);
7902                         }
7903
7904                         if (type>8 || type==2 || type==4 || type==0) {
7905                                 mtx_unlock(&lun->lun_lock);
7906                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7907                                                       /*sks_valid*/ 1,
7908                                                       /*command*/ 1,
7909                                                       /*field*/ 2,
7910                                                       /*bit_valid*/ 1,
7911                                                       /*bit*/ 0);
7912                                 ctl_done((union ctl_io *)ctsio);
7913                                 return (1);
7914                         }
7915
7916                         /*
7917                          * Do the following:
7918                          * if sa_res_key != res_key remove all
7919                          * registrants w/sa_res_key and generate UA
7920                          * for these registrants(Registrations
7921                          * Preempted) if it wasn't an exclusive
7922                          * reservation generate UA(Reservations
7923                          * Preempted) for all other registered nexuses
7924                          * if the type has changed. Establish the new
7925                          * reservation and holder. If res_key and
7926                          * sa_res_key are the same do the above
7927                          * except don't unregister the res holder.
7928                          */
7929
7930                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7931                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7932                                         continue;
7933
7934                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
7935                                         ctl_clr_prkey(lun, i);
7936                                         lun->pr_key_count--;
7937                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7938                                 } else if (type != lun->pr_res_type &&
7939                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
7940                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
7941                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
7942                                 }
7943                         }
7944                         lun->pr_res_type = type;
7945                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7946                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7947                                 lun->pr_res_idx = residx;
7948                         else
7949                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7950                         lun->pr_generation++;
7951                         mtx_unlock(&lun->lun_lock);
7952
7953                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7954                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7955                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7956                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7957                         persis_io.pr.pr_info.res_type = type;
7958                         memcpy(persis_io.pr.pr_info.sa_res_key,
7959                                param->serv_act_res_key,
7960                                sizeof(param->serv_act_res_key));
7961                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7962                             sizeof(persis_io.pr), M_WAITOK);
7963                 } else {
7964                         /*
7965                          * sa_res_key is not the res holder just
7966                          * remove registrants
7967                          */
7968                         int found=0;
7969
7970                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7971                                 if (sa_res_key != ctl_get_prkey(lun, i))
7972                                         continue;
7973
7974                                 found = 1;
7975                                 ctl_clr_prkey(lun, i);
7976                                 lun->pr_key_count--;
7977                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7978                         }
7979
7980                         if (!found) {
7981                                 mtx_unlock(&lun->lun_lock);
7982                                 free(ctsio->kern_data_ptr, M_CTL);
7983                                 ctl_set_reservation_conflict(ctsio);
7984                                 ctl_done((union ctl_io *)ctsio);
7985                                 return (1);
7986                         }
7987                         lun->pr_generation++;
7988                         mtx_unlock(&lun->lun_lock);
7989
7990                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7991                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7992                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7993                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7994                         persis_io.pr.pr_info.res_type = type;
7995                         memcpy(persis_io.pr.pr_info.sa_res_key,
7996                                param->serv_act_res_key,
7997                                sizeof(param->serv_act_res_key));
7998                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7999                             sizeof(persis_io.pr), M_WAITOK);
8000                 }
8001         }
8002         return (0);
8003 }
8004
8005 static void
8006 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8007 {
8008         uint64_t sa_res_key;
8009         int i;
8010
8011         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8012
8013         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8014          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8015          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8016                 if (sa_res_key == 0) {
8017                         /*
8018                          * Unregister everybody else and build UA for
8019                          * them
8020                          */
8021                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8022                                 if (i == msg->pr.pr_info.residx ||
8023                                     ctl_get_prkey(lun, i) == 0)
8024                                         continue;
8025
8026                                 ctl_clr_prkey(lun, i);
8027                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8028                         }
8029
8030                         lun->pr_key_count = 1;
8031                         lun->pr_res_type = msg->pr.pr_info.res_type;
8032                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8033                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8034                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8035                 } else {
8036                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8037                                 if (sa_res_key == ctl_get_prkey(lun, i))
8038                                         continue;
8039
8040                                 ctl_clr_prkey(lun, i);
8041                                 lun->pr_key_count--;
8042                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8043                         }
8044                 }
8045         } else {
8046                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8047                         if (i == msg->pr.pr_info.residx ||
8048                             ctl_get_prkey(lun, i) == 0)
8049                                 continue;
8050
8051                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8052                                 ctl_clr_prkey(lun, i);
8053                                 lun->pr_key_count--;
8054                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8055                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8056                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8057                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8058                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8059                         }
8060                 }
8061                 lun->pr_res_type = msg->pr.pr_info.res_type;
8062                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8063                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8064                         lun->pr_res_idx = msg->pr.pr_info.residx;
8065                 else
8066                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8067         }
8068         lun->pr_generation++;
8069
8070 }
8071
8072 int
8073 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8074 {
8075         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8076         struct ctl_lun *lun = CTL_LUN(ctsio);
8077         int retval;
8078         uint32_t param_len;
8079         struct scsi_per_res_out *cdb;
8080         struct scsi_per_res_out_parms* param;
8081         uint32_t residx;
8082         uint64_t res_key, sa_res_key, key;
8083         uint8_t type;
8084         union ctl_ha_msg persis_io;
8085         int    i;
8086
8087         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8088
8089         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8090         retval = CTL_RETVAL_COMPLETE;
8091
8092         /*
8093          * We only support whole-LUN scope.  The scope & type are ignored for
8094          * register, register and ignore existing key and clear.
8095          * We sometimes ignore scope and type on preempts too!!
8096          * Verify reservation type here as well.
8097          */
8098         type = cdb->scope_type & SPR_TYPE_MASK;
8099         if ((cdb->action == SPRO_RESERVE)
8100          || (cdb->action == SPRO_RELEASE)) {
8101                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8102                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8103                                               /*sks_valid*/ 1,
8104                                               /*command*/ 1,
8105                                               /*field*/ 2,
8106                                               /*bit_valid*/ 1,
8107                                               /*bit*/ 4);
8108                         ctl_done((union ctl_io *)ctsio);
8109                         return (CTL_RETVAL_COMPLETE);
8110                 }
8111
8112                 if (type>8 || type==2 || type==4 || type==0) {
8113                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8114                                               /*sks_valid*/ 1,
8115                                               /*command*/ 1,
8116                                               /*field*/ 2,
8117                                               /*bit_valid*/ 1,
8118                                               /*bit*/ 0);
8119                         ctl_done((union ctl_io *)ctsio);
8120                         return (CTL_RETVAL_COMPLETE);
8121                 }
8122         }
8123
8124         param_len = scsi_4btoul(cdb->length);
8125
8126         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8127                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8128                 ctsio->kern_data_len = param_len;
8129                 ctsio->kern_total_len = param_len;
8130                 ctsio->kern_rel_offset = 0;
8131                 ctsio->kern_sg_entries = 0;
8132                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8133                 ctsio->be_move_done = ctl_config_move_done;
8134                 ctl_datamove((union ctl_io *)ctsio);
8135
8136                 return (CTL_RETVAL_COMPLETE);
8137         }
8138
8139         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8140
8141         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8142         res_key = scsi_8btou64(param->res_key.key);
8143         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8144
8145         /*
8146          * Validate the reservation key here except for SPRO_REG_IGNO
8147          * This must be done for all other service actions
8148          */
8149         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8150                 mtx_lock(&lun->lun_lock);
8151                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8152                         if (res_key != key) {
8153                                 /*
8154                                  * The current key passed in doesn't match
8155                                  * the one the initiator previously
8156                                  * registered.
8157                                  */
8158                                 mtx_unlock(&lun->lun_lock);
8159                                 free(ctsio->kern_data_ptr, M_CTL);
8160                                 ctl_set_reservation_conflict(ctsio);
8161                                 ctl_done((union ctl_io *)ctsio);
8162                                 return (CTL_RETVAL_COMPLETE);
8163                         }
8164                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8165                         /*
8166                          * We are not registered
8167                          */
8168                         mtx_unlock(&lun->lun_lock);
8169                         free(ctsio->kern_data_ptr, M_CTL);
8170                         ctl_set_reservation_conflict(ctsio);
8171                         ctl_done((union ctl_io *)ctsio);
8172                         return (CTL_RETVAL_COMPLETE);
8173                 } else if (res_key != 0) {
8174                         /*
8175                          * We are not registered and trying to register but
8176                          * the register key isn't zero.
8177                          */
8178                         mtx_unlock(&lun->lun_lock);
8179                         free(ctsio->kern_data_ptr, M_CTL);
8180                         ctl_set_reservation_conflict(ctsio);
8181                         ctl_done((union ctl_io *)ctsio);
8182                         return (CTL_RETVAL_COMPLETE);
8183                 }
8184                 mtx_unlock(&lun->lun_lock);
8185         }
8186
8187         switch (cdb->action & SPRO_ACTION_MASK) {
8188         case SPRO_REGISTER:
8189         case SPRO_REG_IGNO: {
8190                 /*
8191                  * We don't support any of these options, as we report in
8192                  * the read capabilities request (see
8193                  * ctl_persistent_reserve_in(), above).
8194                  */
8195                 if ((param->flags & SPR_SPEC_I_PT)
8196                  || (param->flags & SPR_ALL_TG_PT)
8197                  || (param->flags & SPR_APTPL)) {
8198                         int bit_ptr;
8199
8200                         if (param->flags & SPR_APTPL)
8201                                 bit_ptr = 0;
8202                         else if (param->flags & SPR_ALL_TG_PT)
8203                                 bit_ptr = 2;
8204                         else /* SPR_SPEC_I_PT */
8205                                 bit_ptr = 3;
8206
8207                         free(ctsio->kern_data_ptr, M_CTL);
8208                         ctl_set_invalid_field(ctsio,
8209                                               /*sks_valid*/ 1,
8210                                               /*command*/ 0,
8211                                               /*field*/ 20,
8212                                               /*bit_valid*/ 1,
8213                                               /*bit*/ bit_ptr);
8214                         ctl_done((union ctl_io *)ctsio);
8215                         return (CTL_RETVAL_COMPLETE);
8216                 }
8217
8218                 mtx_lock(&lun->lun_lock);
8219
8220                 /*
8221                  * The initiator wants to clear the
8222                  * key/unregister.
8223                  */
8224                 if (sa_res_key == 0) {
8225                         if ((res_key == 0
8226                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8227                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8228                           && ctl_get_prkey(lun, residx) == 0)) {
8229                                 mtx_unlock(&lun->lun_lock);
8230                                 goto done;
8231                         }
8232
8233                         ctl_clr_prkey(lun, residx);
8234                         lun->pr_key_count--;
8235
8236                         if (residx == lun->pr_res_idx) {
8237                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8238                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8239
8240                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8241                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8242                                     lun->pr_key_count) {
8243                                         /*
8244                                          * If the reservation is a registrants
8245                                          * only type we need to generate a UA
8246                                          * for other registered inits.  The
8247                                          * sense code should be RESERVATIONS
8248                                          * RELEASED
8249                                          */
8250
8251                                         for (i = softc->init_min; i < softc->init_max; i++){
8252                                                 if (ctl_get_prkey(lun, i) == 0)
8253                                                         continue;
8254                                                 ctl_est_ua(lun, i,
8255                                                     CTL_UA_RES_RELEASE);
8256                                         }
8257                                 }
8258                                 lun->pr_res_type = 0;
8259                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8260                                 if (lun->pr_key_count==0) {
8261                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8262                                         lun->pr_res_type = 0;
8263                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8264                                 }
8265                         }
8266                         lun->pr_generation++;
8267                         mtx_unlock(&lun->lun_lock);
8268
8269                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8270                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8271                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8272                         persis_io.pr.pr_info.residx = residx;
8273                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8274                             sizeof(persis_io.pr), M_WAITOK);
8275                 } else /* sa_res_key != 0 */ {
8276                         /*
8277                          * If we aren't registered currently then increment
8278                          * the key count and set the registered flag.
8279                          */
8280                         ctl_alloc_prkey(lun, residx);
8281                         if (ctl_get_prkey(lun, residx) == 0)
8282                                 lun->pr_key_count++;
8283                         ctl_set_prkey(lun, residx, sa_res_key);
8284                         lun->pr_generation++;
8285                         mtx_unlock(&lun->lun_lock);
8286
8287                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8288                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8289                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8290                         persis_io.pr.pr_info.residx = residx;
8291                         memcpy(persis_io.pr.pr_info.sa_res_key,
8292                                param->serv_act_res_key,
8293                                sizeof(param->serv_act_res_key));
8294                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8295                             sizeof(persis_io.pr), M_WAITOK);
8296                 }
8297
8298                 break;
8299         }
8300         case SPRO_RESERVE:
8301                 mtx_lock(&lun->lun_lock);
8302                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8303                         /*
8304                          * if this isn't the reservation holder and it's
8305                          * not a "all registrants" type or if the type is
8306                          * different then we have a conflict
8307                          */
8308                         if ((lun->pr_res_idx != residx
8309                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8310                          || lun->pr_res_type != type) {
8311                                 mtx_unlock(&lun->lun_lock);
8312                                 free(ctsio->kern_data_ptr, M_CTL);
8313                                 ctl_set_reservation_conflict(ctsio);
8314                                 ctl_done((union ctl_io *)ctsio);
8315                                 return (CTL_RETVAL_COMPLETE);
8316                         }
8317                         mtx_unlock(&lun->lun_lock);
8318                 } else /* create a reservation */ {
8319                         /*
8320                          * If it's not an "all registrants" type record
8321                          * reservation holder
8322                          */
8323                         if (type != SPR_TYPE_WR_EX_AR
8324                          && type != SPR_TYPE_EX_AC_AR)
8325                                 lun->pr_res_idx = residx; /* Res holder */
8326                         else
8327                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8328
8329                         lun->flags |= CTL_LUN_PR_RESERVED;
8330                         lun->pr_res_type = type;
8331
8332                         mtx_unlock(&lun->lun_lock);
8333
8334                         /* send msg to other side */
8335                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8336                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8337                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8338                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8339                         persis_io.pr.pr_info.res_type = type;
8340                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8341                             sizeof(persis_io.pr), M_WAITOK);
8342                 }
8343                 break;
8344
8345         case SPRO_RELEASE:
8346                 mtx_lock(&lun->lun_lock);
8347                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8348                         /* No reservation exists return good status */
8349                         mtx_unlock(&lun->lun_lock);
8350                         goto done;
8351                 }
8352                 /*
8353                  * Is this nexus a reservation holder?
8354                  */
8355                 if (lun->pr_res_idx != residx
8356                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8357                         /*
8358                          * not a res holder return good status but
8359                          * do nothing
8360                          */
8361                         mtx_unlock(&lun->lun_lock);
8362                         goto done;
8363                 }
8364
8365                 if (lun->pr_res_type != type) {
8366                         mtx_unlock(&lun->lun_lock);
8367                         free(ctsio->kern_data_ptr, M_CTL);
8368                         ctl_set_illegal_pr_release(ctsio);
8369                         ctl_done((union ctl_io *)ctsio);
8370                         return (CTL_RETVAL_COMPLETE);
8371                 }
8372
8373                 /* okay to release */
8374                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8375                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8376                 lun->pr_res_type = 0;
8377
8378                 /*
8379                  * If this isn't an exclusive access reservation and NUAR
8380                  * is not set, generate UA for all other registrants.
8381                  */
8382                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8383                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8384                         for (i = softc->init_min; i < softc->init_max; i++) {
8385                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8386                                         continue;
8387                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8388                         }
8389                 }
8390                 mtx_unlock(&lun->lun_lock);
8391
8392                 /* Send msg to other side */
8393                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8394                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8395                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8396                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8397                      sizeof(persis_io.pr), M_WAITOK);
8398                 break;
8399
8400         case SPRO_CLEAR:
8401                 /* send msg to other side */
8402
8403                 mtx_lock(&lun->lun_lock);
8404                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8405                 lun->pr_res_type = 0;
8406                 lun->pr_key_count = 0;
8407                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8408
8409                 ctl_clr_prkey(lun, residx);
8410                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8411                         if (ctl_get_prkey(lun, i) != 0) {
8412                                 ctl_clr_prkey(lun, i);
8413                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8414                         }
8415                 lun->pr_generation++;
8416                 mtx_unlock(&lun->lun_lock);
8417
8418                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8419                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8420                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8421                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8422                      sizeof(persis_io.pr), M_WAITOK);
8423                 break;
8424
8425         case SPRO_PREEMPT:
8426         case SPRO_PRE_ABO: {
8427                 int nretval;
8428
8429                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8430                                           residx, ctsio, cdb, param);
8431                 if (nretval != 0)
8432                         return (CTL_RETVAL_COMPLETE);
8433                 break;
8434         }
8435         default:
8436                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8437         }
8438
8439 done:
8440         free(ctsio->kern_data_ptr, M_CTL);
8441         ctl_set_success(ctsio);
8442         ctl_done((union ctl_io *)ctsio);
8443
8444         return (retval);
8445 }
8446
8447 /*
8448  * This routine is for handling a message from the other SC pertaining to
8449  * persistent reserve out. All the error checking will have been done
8450  * so only performing the action need be done here to keep the two
8451  * in sync.
8452  */
8453 static void
8454 ctl_hndl_per_res_out_on_other_sc(union ctl_io *io)
8455 {
8456         struct ctl_softc *softc = CTL_SOFTC(io);
8457         union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg;
8458         struct ctl_lun *lun;
8459         int i;
8460         uint32_t residx, targ_lun;
8461
8462         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8463         mtx_lock(&softc->ctl_lock);
8464         if (targ_lun >= ctl_max_luns ||
8465             (lun = softc->ctl_luns[targ_lun]) == NULL) {
8466                 mtx_unlock(&softc->ctl_lock);
8467                 return;
8468         }
8469         mtx_lock(&lun->lun_lock);
8470         mtx_unlock(&softc->ctl_lock);
8471         if (lun->flags & CTL_LUN_DISABLED) {
8472                 mtx_unlock(&lun->lun_lock);
8473                 return;
8474         }
8475         residx = ctl_get_initindex(&msg->hdr.nexus);
8476         switch(msg->pr.pr_info.action) {
8477         case CTL_PR_REG_KEY:
8478                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8479                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8480                         lun->pr_key_count++;
8481                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8482                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8483                 lun->pr_generation++;
8484                 break;
8485
8486         case CTL_PR_UNREG_KEY:
8487                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8488                 lun->pr_key_count--;
8489
8490                 /* XXX Need to see if the reservation has been released */
8491                 /* if so do we need to generate UA? */
8492                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8493                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8494                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8495
8496                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8497                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8498                             lun->pr_key_count) {
8499                                 /*
8500                                  * If the reservation is a registrants
8501                                  * only type we need to generate a UA
8502                                  * for other registered inits.  The
8503                                  * sense code should be RESERVATIONS
8504                                  * RELEASED
8505                                  */
8506
8507                                 for (i = softc->init_min; i < softc->init_max; i++) {
8508                                         if (ctl_get_prkey(lun, i) == 0)
8509                                                 continue;
8510
8511                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8512                                 }
8513                         }
8514                         lun->pr_res_type = 0;
8515                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8516                         if (lun->pr_key_count==0) {
8517                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8518                                 lun->pr_res_type = 0;
8519                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8520                         }
8521                 }
8522                 lun->pr_generation++;
8523                 break;
8524
8525         case CTL_PR_RESERVE:
8526                 lun->flags |= CTL_LUN_PR_RESERVED;
8527                 lun->pr_res_type = msg->pr.pr_info.res_type;
8528                 lun->pr_res_idx = msg->pr.pr_info.residx;
8529
8530                 break;
8531
8532         case CTL_PR_RELEASE:
8533                 /*
8534                  * If this isn't an exclusive access reservation and NUAR
8535                  * is not set, generate UA for all other registrants.
8536                  */
8537                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8538                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8539                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8540                         for (i = softc->init_min; i < softc->init_max; i++) {
8541                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8542                                         continue;
8543                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8544                         }
8545                 }
8546
8547                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8548                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8549                 lun->pr_res_type = 0;
8550                 break;
8551
8552         case CTL_PR_PREEMPT:
8553                 ctl_pro_preempt_other(lun, msg);
8554                 break;
8555         case CTL_PR_CLEAR:
8556                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8557                 lun->pr_res_type = 0;
8558                 lun->pr_key_count = 0;
8559                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8560
8561                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8562                         if (ctl_get_prkey(lun, i) == 0)
8563                                 continue;
8564                         ctl_clr_prkey(lun, i);
8565                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8566                 }
8567                 lun->pr_generation++;
8568                 break;
8569         }
8570
8571         mtx_unlock(&lun->lun_lock);
8572 }
8573
8574 int
8575 ctl_read_write(struct ctl_scsiio *ctsio)
8576 {
8577         struct ctl_lun *lun = CTL_LUN(ctsio);
8578         struct ctl_lba_len_flags *lbalen;
8579         uint64_t lba;
8580         uint32_t num_blocks;
8581         int flags, retval;
8582         int isread;
8583
8584         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8585
8586         flags = 0;
8587         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8588               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8589         switch (ctsio->cdb[0]) {
8590         case READ_6:
8591         case WRITE_6: {
8592                 struct scsi_rw_6 *cdb;
8593
8594                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8595
8596                 lba = scsi_3btoul(cdb->addr);
8597                 /* only 5 bits are valid in the most significant address byte */
8598                 lba &= 0x1fffff;
8599                 num_blocks = cdb->length;
8600                 /*
8601                  * This is correct according to SBC-2.
8602                  */
8603                 if (num_blocks == 0)
8604                         num_blocks = 256;
8605                 break;
8606         }
8607         case READ_10:
8608         case WRITE_10: {
8609                 struct scsi_rw_10 *cdb;
8610
8611                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8612                 if (cdb->byte2 & SRW10_FUA)
8613                         flags |= CTL_LLF_FUA;
8614                 if (cdb->byte2 & SRW10_DPO)
8615                         flags |= CTL_LLF_DPO;
8616                 lba = scsi_4btoul(cdb->addr);
8617                 num_blocks = scsi_2btoul(cdb->length);
8618                 break;
8619         }
8620         case WRITE_VERIFY_10: {
8621                 struct scsi_write_verify_10 *cdb;
8622
8623                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8624                 flags |= CTL_LLF_FUA;
8625                 if (cdb->byte2 & SWV_DPO)
8626                         flags |= CTL_LLF_DPO;
8627                 lba = scsi_4btoul(cdb->addr);
8628                 num_blocks = scsi_2btoul(cdb->length);
8629                 break;
8630         }
8631         case READ_12:
8632         case WRITE_12: {
8633                 struct scsi_rw_12 *cdb;
8634
8635                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8636                 if (cdb->byte2 & SRW12_FUA)
8637                         flags |= CTL_LLF_FUA;
8638                 if (cdb->byte2 & SRW12_DPO)
8639                         flags |= CTL_LLF_DPO;
8640                 lba = scsi_4btoul(cdb->addr);
8641                 num_blocks = scsi_4btoul(cdb->length);
8642                 break;
8643         }
8644         case WRITE_VERIFY_12: {
8645                 struct scsi_write_verify_12 *cdb;
8646
8647                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8648                 flags |= CTL_LLF_FUA;
8649                 if (cdb->byte2 & SWV_DPO)
8650                         flags |= CTL_LLF_DPO;
8651                 lba = scsi_4btoul(cdb->addr);
8652                 num_blocks = scsi_4btoul(cdb->length);
8653                 break;
8654         }
8655         case READ_16:
8656         case WRITE_16: {
8657                 struct scsi_rw_16 *cdb;
8658
8659                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8660                 if (cdb->byte2 & SRW12_FUA)
8661                         flags |= CTL_LLF_FUA;
8662                 if (cdb->byte2 & SRW12_DPO)
8663                         flags |= CTL_LLF_DPO;
8664                 lba = scsi_8btou64(cdb->addr);
8665                 num_blocks = scsi_4btoul(cdb->length);
8666                 break;
8667         }
8668         case WRITE_ATOMIC_16: {
8669                 struct scsi_write_atomic_16 *cdb;
8670
8671                 if (lun->be_lun->atomicblock == 0) {
8672                         ctl_set_invalid_opcode(ctsio);
8673                         ctl_done((union ctl_io *)ctsio);
8674                         return (CTL_RETVAL_COMPLETE);
8675                 }
8676
8677                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8678                 if (cdb->byte2 & SRW12_FUA)
8679                         flags |= CTL_LLF_FUA;
8680                 if (cdb->byte2 & SRW12_DPO)
8681                         flags |= CTL_LLF_DPO;
8682                 lba = scsi_8btou64(cdb->addr);
8683                 num_blocks = scsi_2btoul(cdb->length);
8684                 if (num_blocks > lun->be_lun->atomicblock) {
8685                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8686                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8687                             /*bit*/ 0);
8688                         ctl_done((union ctl_io *)ctsio);
8689                         return (CTL_RETVAL_COMPLETE);
8690                 }
8691                 break;
8692         }
8693         case WRITE_VERIFY_16: {
8694                 struct scsi_write_verify_16 *cdb;
8695
8696                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8697                 flags |= CTL_LLF_FUA;
8698                 if (cdb->byte2 & SWV_DPO)
8699                         flags |= CTL_LLF_DPO;
8700                 lba = scsi_8btou64(cdb->addr);
8701                 num_blocks = scsi_4btoul(cdb->length);
8702                 break;
8703         }
8704         default:
8705                 /*
8706                  * We got a command we don't support.  This shouldn't
8707                  * happen, commands should be filtered out above us.
8708                  */
8709                 ctl_set_invalid_opcode(ctsio);
8710                 ctl_done((union ctl_io *)ctsio);
8711
8712                 return (CTL_RETVAL_COMPLETE);
8713                 break; /* NOTREACHED */
8714         }
8715
8716         /*
8717          * The first check is to make sure we're in bounds, the second
8718          * check is to catch wrap-around problems.  If the lba + num blocks
8719          * is less than the lba, then we've wrapped around and the block
8720          * range is invalid anyway.
8721          */
8722         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8723          || ((lba + num_blocks) < lba)) {
8724                 ctl_set_lba_out_of_range(ctsio,
8725                     MAX(lba, lun->be_lun->maxlba + 1));
8726                 ctl_done((union ctl_io *)ctsio);
8727                 return (CTL_RETVAL_COMPLETE);
8728         }
8729
8730         /*
8731          * According to SBC-3, a transfer length of 0 is not an error.
8732          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8733          * translates to 256 blocks for those commands.
8734          */
8735         if (num_blocks == 0) {
8736                 ctl_set_success(ctsio);
8737                 ctl_done((union ctl_io *)ctsio);
8738                 return (CTL_RETVAL_COMPLETE);
8739         }
8740
8741         /* Set FUA and/or DPO if caches are disabled. */
8742         if (isread) {
8743                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8744                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8745         } else {
8746                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8747                         flags |= CTL_LLF_FUA;
8748         }
8749
8750         lbalen = (struct ctl_lba_len_flags *)
8751             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8752         lbalen->lba = lba;
8753         lbalen->len = num_blocks;
8754         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8755
8756         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8757         ctsio->kern_rel_offset = 0;
8758
8759         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8760
8761         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8762         return (retval);
8763 }
8764
8765 static int
8766 ctl_cnw_cont(union ctl_io *io)
8767 {
8768         struct ctl_lun *lun = CTL_LUN(io);
8769         struct ctl_scsiio *ctsio;
8770         struct ctl_lba_len_flags *lbalen;
8771         int retval;
8772
8773         ctsio = &io->scsiio;
8774         ctsio->io_hdr.status = CTL_STATUS_NONE;
8775         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8776         lbalen = (struct ctl_lba_len_flags *)
8777             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8778         lbalen->flags &= ~CTL_LLF_COMPARE;
8779         lbalen->flags |= CTL_LLF_WRITE;
8780
8781         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8782         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8783         return (retval);
8784 }
8785
8786 int
8787 ctl_cnw(struct ctl_scsiio *ctsio)
8788 {
8789         struct ctl_lun *lun = CTL_LUN(ctsio);
8790         struct ctl_lba_len_flags *lbalen;
8791         uint64_t lba;
8792         uint32_t num_blocks;
8793         int flags, retval;
8794
8795         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8796
8797         flags = 0;
8798         switch (ctsio->cdb[0]) {
8799         case COMPARE_AND_WRITE: {
8800                 struct scsi_compare_and_write *cdb;
8801
8802                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8803                 if (cdb->byte2 & SRW10_FUA)
8804                         flags |= CTL_LLF_FUA;
8805                 if (cdb->byte2 & SRW10_DPO)
8806                         flags |= CTL_LLF_DPO;
8807                 lba = scsi_8btou64(cdb->addr);
8808                 num_blocks = cdb->length;
8809                 break;
8810         }
8811         default:
8812                 /*
8813                  * We got a command we don't support.  This shouldn't
8814                  * happen, commands should be filtered out above us.
8815                  */
8816                 ctl_set_invalid_opcode(ctsio);
8817                 ctl_done((union ctl_io *)ctsio);
8818
8819                 return (CTL_RETVAL_COMPLETE);
8820                 break; /* NOTREACHED */
8821         }
8822
8823         /*
8824          * The first check is to make sure we're in bounds, the second
8825          * check is to catch wrap-around problems.  If the lba + num blocks
8826          * is less than the lba, then we've wrapped around and the block
8827          * range is invalid anyway.
8828          */
8829         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8830          || ((lba + num_blocks) < lba)) {
8831                 ctl_set_lba_out_of_range(ctsio,
8832                     MAX(lba, lun->be_lun->maxlba + 1));
8833                 ctl_done((union ctl_io *)ctsio);
8834                 return (CTL_RETVAL_COMPLETE);
8835         }
8836
8837         /*
8838          * According to SBC-3, a transfer length of 0 is not an error.
8839          */
8840         if (num_blocks == 0) {
8841                 ctl_set_success(ctsio);
8842                 ctl_done((union ctl_io *)ctsio);
8843                 return (CTL_RETVAL_COMPLETE);
8844         }
8845
8846         /* Set FUA if write cache is disabled. */
8847         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8848                 flags |= CTL_LLF_FUA;
8849
8850         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8851         ctsio->kern_rel_offset = 0;
8852
8853         /*
8854          * Set the IO_CONT flag, so that if this I/O gets passed to
8855          * ctl_data_submit_done(), it'll get passed back to
8856          * ctl_ctl_cnw_cont() for further processing.
8857          */
8858         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8859         ctsio->io_cont = ctl_cnw_cont;
8860
8861         lbalen = (struct ctl_lba_len_flags *)
8862             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8863         lbalen->lba = lba;
8864         lbalen->len = num_blocks;
8865         lbalen->flags = CTL_LLF_COMPARE | flags;
8866
8867         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8868         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8869         return (retval);
8870 }
8871
8872 int
8873 ctl_verify(struct ctl_scsiio *ctsio)
8874 {
8875         struct ctl_lun *lun = CTL_LUN(ctsio);
8876         struct ctl_lba_len_flags *lbalen;
8877         uint64_t lba;
8878         uint32_t num_blocks;
8879         int bytchk, flags;
8880         int retval;
8881
8882         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8883
8884         bytchk = 0;
8885         flags = CTL_LLF_FUA;
8886         switch (ctsio->cdb[0]) {
8887         case VERIFY_10: {
8888                 struct scsi_verify_10 *cdb;
8889
8890                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8891                 if (cdb->byte2 & SVFY_BYTCHK)
8892                         bytchk = 1;
8893                 if (cdb->byte2 & SVFY_DPO)
8894                         flags |= CTL_LLF_DPO;
8895                 lba = scsi_4btoul(cdb->addr);
8896                 num_blocks = scsi_2btoul(cdb->length);
8897                 break;
8898         }
8899         case VERIFY_12: {
8900                 struct scsi_verify_12 *cdb;
8901
8902                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8903                 if (cdb->byte2 & SVFY_BYTCHK)
8904                         bytchk = 1;
8905                 if (cdb->byte2 & SVFY_DPO)
8906                         flags |= CTL_LLF_DPO;
8907                 lba = scsi_4btoul(cdb->addr);
8908                 num_blocks = scsi_4btoul(cdb->length);
8909                 break;
8910         }
8911         case VERIFY_16: {
8912                 struct scsi_rw_16 *cdb;
8913
8914                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8915                 if (cdb->byte2 & SVFY_BYTCHK)
8916                         bytchk = 1;
8917                 if (cdb->byte2 & SVFY_DPO)
8918                         flags |= CTL_LLF_DPO;
8919                 lba = scsi_8btou64(cdb->addr);
8920                 num_blocks = scsi_4btoul(cdb->length);
8921                 break;
8922         }
8923         default:
8924                 /*
8925                  * We got a command we don't support.  This shouldn't
8926                  * happen, commands should be filtered out above us.
8927                  */
8928                 ctl_set_invalid_opcode(ctsio);
8929                 ctl_done((union ctl_io *)ctsio);
8930                 return (CTL_RETVAL_COMPLETE);
8931         }
8932
8933         /*
8934          * The first check is to make sure we're in bounds, the second
8935          * check is to catch wrap-around problems.  If the lba + num blocks
8936          * is less than the lba, then we've wrapped around and the block
8937          * range is invalid anyway.
8938          */
8939         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8940          || ((lba + num_blocks) < lba)) {
8941                 ctl_set_lba_out_of_range(ctsio,
8942                     MAX(lba, lun->be_lun->maxlba + 1));
8943                 ctl_done((union ctl_io *)ctsio);
8944                 return (CTL_RETVAL_COMPLETE);
8945         }
8946
8947         /*
8948          * According to SBC-3, a transfer length of 0 is not an error.
8949          */
8950         if (num_blocks == 0) {
8951                 ctl_set_success(ctsio);
8952                 ctl_done((union ctl_io *)ctsio);
8953                 return (CTL_RETVAL_COMPLETE);
8954         }
8955
8956         lbalen = (struct ctl_lba_len_flags *)
8957             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8958         lbalen->lba = lba;
8959         lbalen->len = num_blocks;
8960         if (bytchk) {
8961                 lbalen->flags = CTL_LLF_COMPARE | flags;
8962                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8963         } else {
8964                 lbalen->flags = CTL_LLF_VERIFY | flags;
8965                 ctsio->kern_total_len = 0;
8966         }
8967         ctsio->kern_rel_offset = 0;
8968
8969         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
8970         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8971         return (retval);
8972 }
8973
8974 int
8975 ctl_report_luns(struct ctl_scsiio *ctsio)
8976 {
8977         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8978         struct ctl_port *port = CTL_PORT(ctsio);
8979         struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio);
8980         struct scsi_report_luns *cdb;
8981         struct scsi_report_luns_data *lun_data;
8982         int num_filled, num_luns, num_port_luns, retval;
8983         uint32_t alloc_len, lun_datalen;
8984         uint32_t initidx, targ_lun_id, lun_id;
8985
8986         retval = CTL_RETVAL_COMPLETE;
8987         cdb = (struct scsi_report_luns *)ctsio->cdb;
8988
8989         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
8990
8991         num_luns = 0;
8992         num_port_luns = port->lun_map ? port->lun_map_size : ctl_max_luns;
8993         mtx_lock(&softc->ctl_lock);
8994         for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) {
8995                 if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX)
8996                         num_luns++;
8997         }
8998         mtx_unlock(&softc->ctl_lock);
8999
9000         switch (cdb->select_report) {
9001         case RPL_REPORT_DEFAULT:
9002         case RPL_REPORT_ALL:
9003         case RPL_REPORT_NONSUBSID:
9004                 break;
9005         case RPL_REPORT_WELLKNOWN:
9006         case RPL_REPORT_ADMIN:
9007         case RPL_REPORT_CONGLOM:
9008                 num_luns = 0;
9009                 break;
9010         default:
9011                 ctl_set_invalid_field(ctsio,
9012                                       /*sks_valid*/ 1,
9013                                       /*command*/ 1,
9014                                       /*field*/ 2,
9015                                       /*bit_valid*/ 0,
9016                                       /*bit*/ 0);
9017                 ctl_done((union ctl_io *)ctsio);
9018                 return (retval);
9019                 break; /* NOTREACHED */
9020         }
9021
9022         alloc_len = scsi_4btoul(cdb->length);
9023         /*
9024          * The initiator has to allocate at least 16 bytes for this request,
9025          * so he can at least get the header and the first LUN.  Otherwise
9026          * we reject the request (per SPC-3 rev 14, section 6.21).
9027          */
9028         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9029             sizeof(struct scsi_report_luns_lundata))) {
9030                 ctl_set_invalid_field(ctsio,
9031                                       /*sks_valid*/ 1,
9032                                       /*command*/ 1,
9033                                       /*field*/ 6,
9034                                       /*bit_valid*/ 0,
9035                                       /*bit*/ 0);
9036                 ctl_done((union ctl_io *)ctsio);
9037                 return (retval);
9038         }
9039
9040         lun_datalen = sizeof(*lun_data) +
9041                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9042
9043         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9044         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9045         ctsio->kern_sg_entries = 0;
9046
9047         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9048
9049         mtx_lock(&softc->ctl_lock);
9050         for (targ_lun_id = 0, num_filled = 0;
9051             targ_lun_id < num_port_luns && num_filled < num_luns;
9052             targ_lun_id++) {
9053                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9054                 if (lun_id == UINT32_MAX)
9055                         continue;
9056                 lun = softc->ctl_luns[lun_id];
9057                 if (lun == NULL)
9058                         continue;
9059
9060                 be64enc(lun_data->luns[num_filled++].lundata,
9061                     ctl_encode_lun(targ_lun_id));
9062
9063                 /*
9064                  * According to SPC-3, rev 14 section 6.21:
9065                  *
9066                  * "The execution of a REPORT LUNS command to any valid and
9067                  * installed logical unit shall clear the REPORTED LUNS DATA
9068                  * HAS CHANGED unit attention condition for all logical
9069                  * units of that target with respect to the requesting
9070                  * initiator. A valid and installed logical unit is one
9071                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9072                  * INQUIRY data (see 6.4.2)."
9073                  *
9074                  * If request_lun is NULL, the LUN this report luns command
9075                  * was issued to is either disabled or doesn't exist. In that
9076                  * case, we shouldn't clear any pending lun change unit
9077                  * attention.
9078                  */
9079                 if (request_lun != NULL) {
9080                         mtx_lock(&lun->lun_lock);
9081                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9082                         mtx_unlock(&lun->lun_lock);
9083                 }
9084         }
9085         mtx_unlock(&softc->ctl_lock);
9086
9087         /*
9088          * It's quite possible that we've returned fewer LUNs than we allocated
9089          * space for.  Trim it.
9090          */
9091         lun_datalen = sizeof(*lun_data) +
9092                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9093         ctsio->kern_rel_offset = 0;
9094         ctsio->kern_sg_entries = 0;
9095         ctsio->kern_data_len = min(lun_datalen, alloc_len);
9096         ctsio->kern_total_len = ctsio->kern_data_len;
9097
9098         /*
9099          * We set this to the actual data length, regardless of how much
9100          * space we actually have to return results.  If the user looks at
9101          * this value, he'll know whether or not he allocated enough space
9102          * and reissue the command if necessary.  We don't support well
9103          * known logical units, so if the user asks for that, return none.
9104          */
9105         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9106
9107         /*
9108          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9109          * this request.
9110          */
9111         ctl_set_success(ctsio);
9112         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9113         ctsio->be_move_done = ctl_config_move_done;
9114         ctl_datamove((union ctl_io *)ctsio);
9115         return (retval);
9116 }
9117
9118 int
9119 ctl_request_sense(struct ctl_scsiio *ctsio)
9120 {
9121         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9122         struct ctl_lun *lun = CTL_LUN(ctsio);
9123         struct scsi_request_sense *cdb;
9124         struct scsi_sense_data *sense_ptr, *ps;
9125         uint32_t initidx;
9126         int have_error;
9127         u_int sense_len = SSD_FULL_SIZE;
9128         scsi_sense_data_type sense_format;
9129         ctl_ua_type ua_type;
9130         uint8_t asc = 0, ascq = 0;
9131
9132         cdb = (struct scsi_request_sense *)ctsio->cdb;
9133
9134         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9135
9136         /*
9137          * Determine which sense format the user wants.
9138          */
9139         if (cdb->byte2 & SRS_DESC)
9140                 sense_format = SSD_TYPE_DESC;
9141         else
9142                 sense_format = SSD_TYPE_FIXED;
9143
9144         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9145         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9146         ctsio->kern_sg_entries = 0;
9147         ctsio->kern_rel_offset = 0;
9148
9149         /*
9150          * struct scsi_sense_data, which is currently set to 256 bytes, is
9151          * larger than the largest allowed value for the length field in the
9152          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9153          */
9154         ctsio->kern_data_len = cdb->length;
9155         ctsio->kern_total_len = cdb->length;
9156
9157         /*
9158          * If we don't have a LUN, we don't have any pending sense.
9159          */
9160         if (lun == NULL ||
9161             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9162              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9163                 /* "Logical unit not supported" */
9164                 ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9165                     /*current_error*/ 1,
9166                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9167                     /*asc*/ 0x25,
9168                     /*ascq*/ 0x00,
9169                     SSD_ELEM_NONE);
9170                 goto send;
9171         }
9172
9173         have_error = 0;
9174         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9175         /*
9176          * Check for pending sense, and then for pending unit attentions.
9177          * Pending sense gets returned first, then pending unit attentions.
9178          */
9179         mtx_lock(&lun->lun_lock);
9180         ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
9181         if (ps != NULL)
9182                 ps += initidx % CTL_MAX_INIT_PER_PORT;
9183         if (ps != NULL && ps->error_code != 0) {
9184                 scsi_sense_data_type stored_format;
9185
9186                 /*
9187                  * Check to see which sense format was used for the stored
9188                  * sense data.
9189                  */
9190                 stored_format = scsi_sense_type(ps);
9191
9192                 /*
9193                  * If the user requested a different sense format than the
9194                  * one we stored, then we need to convert it to the other
9195                  * format.  If we're going from descriptor to fixed format
9196                  * sense data, we may lose things in translation, depending
9197                  * on what options were used.
9198                  *
9199                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9200                  * for some reason we'll just copy it out as-is.
9201                  */
9202                 if ((stored_format == SSD_TYPE_FIXED)
9203                  && (sense_format == SSD_TYPE_DESC))
9204                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9205                             ps, (struct scsi_sense_data_desc *)sense_ptr);
9206                 else if ((stored_format == SSD_TYPE_DESC)
9207                       && (sense_format == SSD_TYPE_FIXED))
9208                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9209                             ps, (struct scsi_sense_data_fixed *)sense_ptr);
9210                 else
9211                         memcpy(sense_ptr, ps, sizeof(*sense_ptr));
9212
9213                 ps->error_code = 0;
9214                 have_error = 1;
9215         } else {
9216                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9217                     sense_format);
9218                 if (ua_type != CTL_UA_NONE)
9219                         have_error = 1;
9220         }
9221         if (have_error == 0) {
9222                 /*
9223                  * Report informational exception if have one and allowed.
9224                  */
9225                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9226                         asc = lun->ie_asc;
9227                         ascq = lun->ie_ascq;
9228                 }
9229                 ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9230                     /*current_error*/ 1,
9231                     /*sense_key*/ SSD_KEY_NO_SENSE,
9232                     /*asc*/ asc,
9233                     /*ascq*/ ascq,
9234                     SSD_ELEM_NONE);
9235         }
9236         mtx_unlock(&lun->lun_lock);
9237
9238 send:
9239         /*
9240          * We report the SCSI status as OK, since the status of the command
9241          * itself is OK.  We're reporting sense as parameter data.
9242          */
9243         ctl_set_success(ctsio);
9244         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9245         ctsio->be_move_done = ctl_config_move_done;
9246         ctl_datamove((union ctl_io *)ctsio);
9247         return (CTL_RETVAL_COMPLETE);
9248 }
9249
9250 int
9251 ctl_tur(struct ctl_scsiio *ctsio)
9252 {
9253
9254         CTL_DEBUG_PRINT(("ctl_tur\n"));
9255
9256         ctl_set_success(ctsio);
9257         ctl_done((union ctl_io *)ctsio);
9258
9259         return (CTL_RETVAL_COMPLETE);
9260 }
9261
9262 /*
9263  * SCSI VPD page 0x00, the Supported VPD Pages page.
9264  */
9265 static int
9266 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9267 {
9268         struct ctl_lun *lun = CTL_LUN(ctsio);
9269         struct scsi_vpd_supported_pages *pages;
9270         int sup_page_size;
9271         int p;
9272
9273         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9274             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9275         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9276         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9277         ctsio->kern_rel_offset = 0;
9278         ctsio->kern_sg_entries = 0;
9279         ctsio->kern_data_len = min(sup_page_size, alloc_len);
9280         ctsio->kern_total_len = ctsio->kern_data_len;
9281
9282         /*
9283          * The control device is always connected.  The disk device, on the
9284          * other hand, may not be online all the time.  Need to change this
9285          * to figure out whether the disk device is actually online or not.
9286          */
9287         if (lun != NULL)
9288                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9289                                 lun->be_lun->lun_type;
9290         else
9291                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9292
9293         p = 0;
9294         /* Supported VPD pages */
9295         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9296         /* Serial Number */
9297         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9298         /* Device Identification */
9299         pages->page_list[p++] = SVPD_DEVICE_ID;
9300         /* Extended INQUIRY Data */
9301         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9302         /* Mode Page Policy */
9303         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9304         /* SCSI Ports */
9305         pages->page_list[p++] = SVPD_SCSI_PORTS;
9306         /* Third-party Copy */
9307         pages->page_list[p++] = SVPD_SCSI_TPC;
9308         /* SCSI Feature Sets */
9309         pages->page_list[p++] = SVPD_SCSI_SFS;
9310         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9311                 /* Block limits */
9312                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9313                 /* Block Device Characteristics */
9314                 pages->page_list[p++] = SVPD_BDC;
9315                 /* Logical Block Provisioning */
9316                 pages->page_list[p++] = SVPD_LBP;
9317         }
9318         pages->length = p;
9319
9320         ctl_set_success(ctsio);
9321         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9322         ctsio->be_move_done = ctl_config_move_done;
9323         ctl_datamove((union ctl_io *)ctsio);
9324         return (CTL_RETVAL_COMPLETE);
9325 }
9326
9327 /*
9328  * SCSI VPD page 0x80, the Unit Serial Number page.
9329  */
9330 static int
9331 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9332 {
9333         struct ctl_lun *lun = CTL_LUN(ctsio);
9334         struct scsi_vpd_unit_serial_number *sn_ptr;
9335         int data_len;
9336
9337         data_len = 4 + CTL_SN_LEN;
9338         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9339         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9340         ctsio->kern_rel_offset = 0;
9341         ctsio->kern_sg_entries = 0;
9342         ctsio->kern_data_len = min(data_len, alloc_len);
9343         ctsio->kern_total_len = ctsio->kern_data_len;
9344
9345         /*
9346          * The control device is always connected.  The disk device, on the
9347          * other hand, may not be online all the time.  Need to change this
9348          * to figure out whether the disk device is actually online or not.
9349          */
9350         if (lun != NULL)
9351                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9352                                   lun->be_lun->lun_type;
9353         else
9354                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9355
9356         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9357         sn_ptr->length = CTL_SN_LEN;
9358         /*
9359          * If we don't have a LUN, we just leave the serial number as
9360          * all spaces.
9361          */
9362         if (lun != NULL) {
9363                 strncpy((char *)sn_ptr->serial_num,
9364                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9365         } else
9366                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9367
9368         ctl_set_success(ctsio);
9369         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9370         ctsio->be_move_done = ctl_config_move_done;
9371         ctl_datamove((union ctl_io *)ctsio);
9372         return (CTL_RETVAL_COMPLETE);
9373 }
9374
9375 /*
9376  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9377  */
9378 static int
9379 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9380 {
9381         struct ctl_lun *lun = CTL_LUN(ctsio);
9382         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9383         int data_len;
9384
9385         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9386         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9387         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9388         ctsio->kern_sg_entries = 0;
9389         ctsio->kern_rel_offset = 0;
9390         ctsio->kern_data_len = min(data_len, alloc_len);
9391         ctsio->kern_total_len = ctsio->kern_data_len;
9392
9393         /*
9394          * The control device is always connected.  The disk device, on the
9395          * other hand, may not be online all the time.
9396          */
9397         if (lun != NULL)
9398                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9399                                      lun->be_lun->lun_type;
9400         else
9401                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9402         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9403         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9404         /*
9405          * We support head of queue, ordered and simple tags.
9406          */
9407         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9408         /*
9409          * Volatile cache supported.
9410          */
9411         eid_ptr->flags3 = SVPD_EID_V_SUP;
9412
9413         /*
9414          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9415          * attention for a particular IT nexus on all LUNs once we report
9416          * it to that nexus once.  This bit is required as of SPC-4.
9417          */
9418         eid_ptr->flags4 = SVPD_EID_LUICLR;
9419
9420         /*
9421          * We support revert to defaults (RTD) bit in MODE SELECT.
9422          */
9423         eid_ptr->flags5 = SVPD_EID_RTD_SUP;
9424
9425         /*
9426          * XXX KDM in order to correctly answer this, we would need
9427          * information from the SIM to determine how much sense data it
9428          * can send.  So this would really be a path inquiry field, most
9429          * likely.  This can be set to a maximum of 252 according to SPC-4,
9430          * but the hardware may or may not be able to support that much.
9431          * 0 just means that the maximum sense data length is not reported.
9432          */
9433         eid_ptr->max_sense_length = 0;
9434
9435         ctl_set_success(ctsio);
9436         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9437         ctsio->be_move_done = ctl_config_move_done;
9438         ctl_datamove((union ctl_io *)ctsio);
9439         return (CTL_RETVAL_COMPLETE);
9440 }
9441
9442 static int
9443 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9444 {
9445         struct ctl_lun *lun = CTL_LUN(ctsio);
9446         struct scsi_vpd_mode_page_policy *mpp_ptr;
9447         int data_len;
9448
9449         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9450             sizeof(struct scsi_vpd_mode_page_policy_descr);
9451
9452         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9453         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9454         ctsio->kern_rel_offset = 0;
9455         ctsio->kern_sg_entries = 0;
9456         ctsio->kern_data_len = min(data_len, alloc_len);
9457         ctsio->kern_total_len = ctsio->kern_data_len;
9458
9459         /*
9460          * The control device is always connected.  The disk device, on the
9461          * other hand, may not be online all the time.
9462          */
9463         if (lun != NULL)
9464                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9465                                      lun->be_lun->lun_type;
9466         else
9467                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9468         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9469         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9470         mpp_ptr->descr[0].page_code = 0x3f;
9471         mpp_ptr->descr[0].subpage_code = 0xff;
9472         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9473
9474         ctl_set_success(ctsio);
9475         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9476         ctsio->be_move_done = ctl_config_move_done;
9477         ctl_datamove((union ctl_io *)ctsio);
9478         return (CTL_RETVAL_COMPLETE);
9479 }
9480
9481 /*
9482  * SCSI VPD page 0x83, the Device Identification page.
9483  */
9484 static int
9485 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9486 {
9487         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9488         struct ctl_port *port = CTL_PORT(ctsio);
9489         struct ctl_lun *lun = CTL_LUN(ctsio);
9490         struct scsi_vpd_device_id *devid_ptr;
9491         struct scsi_vpd_id_descriptor *desc;
9492         int data_len, g;
9493         uint8_t proto;
9494
9495         data_len = sizeof(struct scsi_vpd_device_id) +
9496             sizeof(struct scsi_vpd_id_descriptor) +
9497                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9498             sizeof(struct scsi_vpd_id_descriptor) +
9499                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9500         if (lun && lun->lun_devid)
9501                 data_len += lun->lun_devid->len;
9502         if (port && port->port_devid)
9503                 data_len += port->port_devid->len;
9504         if (port && port->target_devid)
9505                 data_len += port->target_devid->len;
9506
9507         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9508         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9509         ctsio->kern_sg_entries = 0;
9510         ctsio->kern_rel_offset = 0;
9511         ctsio->kern_sg_entries = 0;
9512         ctsio->kern_data_len = min(data_len, alloc_len);
9513         ctsio->kern_total_len = ctsio->kern_data_len;
9514
9515         /*
9516          * The control device is always connected.  The disk device, on the
9517          * other hand, may not be online all the time.
9518          */
9519         if (lun != NULL)
9520                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9521                                      lun->be_lun->lun_type;
9522         else
9523                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9524         devid_ptr->page_code = SVPD_DEVICE_ID;
9525         scsi_ulto2b(data_len - 4, devid_ptr->length);
9526
9527         if (port && port->port_type == CTL_PORT_FC)
9528                 proto = SCSI_PROTO_FC << 4;
9529         else if (port && port->port_type == CTL_PORT_SAS)
9530                 proto = SCSI_PROTO_SAS << 4;
9531         else if (port && port->port_type == CTL_PORT_ISCSI)
9532                 proto = SCSI_PROTO_ISCSI << 4;
9533         else
9534                 proto = SCSI_PROTO_SPI << 4;
9535         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9536
9537         /*
9538          * We're using a LUN association here.  i.e., this device ID is a
9539          * per-LUN identifier.
9540          */
9541         if (lun && lun->lun_devid) {
9542                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9543                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9544                     lun->lun_devid->len);
9545         }
9546
9547         /*
9548          * This is for the WWPN which is a port association.
9549          */
9550         if (port && port->port_devid) {
9551                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9552                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9553                     port->port_devid->len);
9554         }
9555
9556         /*
9557          * This is for the Relative Target Port(type 4h) identifier
9558          */
9559         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9560         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9561             SVPD_ID_TYPE_RELTARG;
9562         desc->length = 4;
9563         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9564         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9565             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9566
9567         /*
9568          * This is for the Target Port Group(type 5h) identifier
9569          */
9570         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9571         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9572             SVPD_ID_TYPE_TPORTGRP;
9573         desc->length = 4;
9574         if (softc->is_single ||
9575             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9576                 g = 1;
9577         else
9578                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9579         scsi_ulto2b(g, &desc->identifier[2]);
9580         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9581             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9582
9583         /*
9584          * This is for the Target identifier
9585          */
9586         if (port && port->target_devid) {
9587                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9588         }
9589
9590         ctl_set_success(ctsio);
9591         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9592         ctsio->be_move_done = ctl_config_move_done;
9593         ctl_datamove((union ctl_io *)ctsio);
9594         return (CTL_RETVAL_COMPLETE);
9595 }
9596
9597 static int
9598 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9599 {
9600         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9601         struct ctl_lun *lun = CTL_LUN(ctsio);
9602         struct scsi_vpd_scsi_ports *sp;
9603         struct scsi_vpd_port_designation *pd;
9604         struct scsi_vpd_port_designation_cont *pdc;
9605         struct ctl_port *port;
9606         int data_len, num_target_ports, iid_len, id_len;
9607
9608         num_target_ports = 0;
9609         iid_len = 0;
9610         id_len = 0;
9611         mtx_lock(&softc->ctl_lock);
9612         STAILQ_FOREACH(port, &softc->port_list, links) {
9613                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9614                         continue;
9615                 if (lun != NULL &&
9616                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9617                         continue;
9618                 num_target_ports++;
9619                 if (port->init_devid)
9620                         iid_len += port->init_devid->len;
9621                 if (port->port_devid)
9622                         id_len += port->port_devid->len;
9623         }
9624         mtx_unlock(&softc->ctl_lock);
9625
9626         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9627             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9628              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9629         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9630         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9631         ctsio->kern_sg_entries = 0;
9632         ctsio->kern_rel_offset = 0;
9633         ctsio->kern_sg_entries = 0;
9634         ctsio->kern_data_len = min(data_len, alloc_len);
9635         ctsio->kern_total_len = ctsio->kern_data_len;
9636
9637         /*
9638          * The control device is always connected.  The disk device, on the
9639          * other hand, may not be online all the time.  Need to change this
9640          * to figure out whether the disk device is actually online or not.
9641          */
9642         if (lun != NULL)
9643                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9644                                   lun->be_lun->lun_type;
9645         else
9646                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9647
9648         sp->page_code = SVPD_SCSI_PORTS;
9649         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9650             sp->page_length);
9651         pd = &sp->design[0];
9652
9653         mtx_lock(&softc->ctl_lock);
9654         STAILQ_FOREACH(port, &softc->port_list, links) {
9655                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9656                         continue;
9657                 if (lun != NULL &&
9658                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9659                         continue;
9660                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9661                 if (port->init_devid) {
9662                         iid_len = port->init_devid->len;
9663                         memcpy(pd->initiator_transportid,
9664                             port->init_devid->data, port->init_devid->len);
9665                 } else
9666                         iid_len = 0;
9667                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9668                 pdc = (struct scsi_vpd_port_designation_cont *)
9669                     (&pd->initiator_transportid[iid_len]);
9670                 if (port->port_devid) {
9671                         id_len = port->port_devid->len;
9672                         memcpy(pdc->target_port_descriptors,
9673                             port->port_devid->data, port->port_devid->len);
9674                 } else
9675                         id_len = 0;
9676                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9677                 pd = (struct scsi_vpd_port_designation *)
9678                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9679         }
9680         mtx_unlock(&softc->ctl_lock);
9681
9682         ctl_set_success(ctsio);
9683         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9684         ctsio->be_move_done = ctl_config_move_done;
9685         ctl_datamove((union ctl_io *)ctsio);
9686         return (CTL_RETVAL_COMPLETE);
9687 }
9688
9689 static int
9690 ctl_inquiry_evpd_sfs(struct ctl_scsiio *ctsio, int alloc_len)
9691 {
9692         struct ctl_lun *lun = CTL_LUN(ctsio);
9693         struct scsi_vpd_sfs *sfs_ptr;
9694         int sfs_page_size, n;
9695
9696         sfs_page_size = sizeof(*sfs_ptr) + 5 * 2;
9697         ctsio->kern_data_ptr = malloc(sfs_page_size, M_CTL, M_WAITOK | M_ZERO);
9698         sfs_ptr = (struct scsi_vpd_sfs *)ctsio->kern_data_ptr;
9699         ctsio->kern_sg_entries = 0;
9700         ctsio->kern_rel_offset = 0;
9701         ctsio->kern_sg_entries = 0;
9702         ctsio->kern_data_len = min(sfs_page_size, alloc_len);
9703         ctsio->kern_total_len = ctsio->kern_data_len;
9704
9705         /*
9706          * The control device is always connected.  The disk device, on the
9707          * other hand, may not be online all the time.  Need to change this
9708          * to figure out whether the disk device is actually online or not.
9709          */
9710         if (lun != NULL)
9711                 sfs_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9712                                   lun->be_lun->lun_type;
9713         else
9714                 sfs_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9715
9716         sfs_ptr->page_code = SVPD_SCSI_SFS;
9717         n = 0;
9718         /* Discovery 2016 */
9719         scsi_ulto2b(0x0001, &sfs_ptr->codes[2 * n++]);
9720         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9721                  /* SBC Base 2016 */
9722                 scsi_ulto2b(0x0101, &sfs_ptr->codes[2 * n++]);
9723                  /* SBC Base 2010 */
9724                 scsi_ulto2b(0x0102, &sfs_ptr->codes[2 * n++]);
9725                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9726                         /* Basic Provisioning 2016 */
9727                         scsi_ulto2b(0x0103, &sfs_ptr->codes[2 * n++]);
9728                 }
9729                 /* Drive Maintenance 2016 */
9730                 //scsi_ulto2b(0x0104, &sfs_ptr->codes[2 * n++]);
9731         }
9732         scsi_ulto2b(4 + 2 * n, sfs_ptr->page_length);
9733
9734         ctl_set_success(ctsio);
9735         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9736         ctsio->be_move_done = ctl_config_move_done;
9737         ctl_datamove((union ctl_io *)ctsio);
9738         return (CTL_RETVAL_COMPLETE);
9739 }
9740
9741 static int
9742 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9743 {
9744         struct ctl_lun *lun = CTL_LUN(ctsio);
9745         struct scsi_vpd_block_limits *bl_ptr;
9746         const char *val;
9747         uint64_t ival;
9748
9749         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9750         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9751         ctsio->kern_sg_entries = 0;
9752         ctsio->kern_rel_offset = 0;
9753         ctsio->kern_sg_entries = 0;
9754         ctsio->kern_data_len = min(sizeof(*bl_ptr), alloc_len);
9755         ctsio->kern_total_len = ctsio->kern_data_len;
9756
9757         /*
9758          * The control device is always connected.  The disk device, on the
9759          * other hand, may not be online all the time.  Need to change this
9760          * to figure out whether the disk device is actually online or not.
9761          */
9762         if (lun != NULL)
9763                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9764                                   lun->be_lun->lun_type;
9765         else
9766                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9767
9768         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9769         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9770         bl_ptr->max_cmp_write_len = 0xff;
9771         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9772         if (lun != NULL) {
9773                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9774                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9775                         ival = 0xffffffff;
9776                         val = dnvlist_get_string(lun->be_lun->options,
9777                             "unmap_max_lba", NULL);
9778                         if (val != NULL)
9779                                 ctl_expand_number(val, &ival);
9780                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9781                         ival = 0xffffffff;
9782                         val = dnvlist_get_string(lun->be_lun->options,
9783                             "unmap_max_descr", NULL);
9784                         if (val != NULL)
9785                                 ctl_expand_number(val, &ival);
9786                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9787                         if (lun->be_lun->ublockexp != 0) {
9788                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9789                                     bl_ptr->opt_unmap_grain);
9790                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9791                                     bl_ptr->unmap_grain_align);
9792                         }
9793                 }
9794                 scsi_ulto4b(lun->be_lun->atomicblock,
9795                     bl_ptr->max_atomic_transfer_length);
9796                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9797                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9798                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9799                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9800                 ival = UINT64_MAX;
9801                 val = dnvlist_get_string(lun->be_lun->options,
9802                     "write_same_max_lba", NULL);
9803                 if (val != NULL)
9804                         ctl_expand_number(val, &ival);
9805                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9806                 if (lun->be_lun->maxlba + 1 > ival)
9807                         bl_ptr->flags |= SVPD_BL_WSNZ;
9808         }
9809
9810         ctl_set_success(ctsio);
9811         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9812         ctsio->be_move_done = ctl_config_move_done;
9813         ctl_datamove((union ctl_io *)ctsio);
9814         return (CTL_RETVAL_COMPLETE);
9815 }
9816
9817 static int
9818 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9819 {
9820         struct ctl_lun *lun = CTL_LUN(ctsio);
9821         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9822         const char *value;
9823         u_int i;
9824
9825         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9826         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9827         ctsio->kern_sg_entries = 0;
9828         ctsio->kern_rel_offset = 0;
9829         ctsio->kern_data_len = min(sizeof(*bdc_ptr), alloc_len);
9830         ctsio->kern_total_len = ctsio->kern_data_len;
9831
9832         /*
9833          * The control device is always connected.  The disk device, on the
9834          * other hand, may not be online all the time.  Need to change this
9835          * to figure out whether the disk device is actually online or not.
9836          */
9837         if (lun != NULL)
9838                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9839                                   lun->be_lun->lun_type;
9840         else
9841                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9842         bdc_ptr->page_code = SVPD_BDC;
9843         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9844         if (lun != NULL &&
9845             (value = dnvlist_get_string(lun->be_lun->options, "rpm", NULL)) != NULL)
9846                 i = strtol(value, NULL, 0);
9847         else
9848                 i = CTL_DEFAULT_ROTATION_RATE;
9849         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9850         if (lun != NULL &&
9851             (value = dnvlist_get_string(lun->be_lun->options, "formfactor", NULL)) != NULL)
9852                 i = strtol(value, NULL, 0);
9853         else
9854                 i = 0;
9855         bdc_ptr->wab_wac_ff = (i & 0x0f);
9856         bdc_ptr->flags = SVPD_RBWZ | SVPD_FUAB | SVPD_VBULS;
9857
9858         ctl_set_success(ctsio);
9859         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9860         ctsio->be_move_done = ctl_config_move_done;
9861         ctl_datamove((union ctl_io *)ctsio);
9862         return (CTL_RETVAL_COMPLETE);
9863 }
9864
9865 static int
9866 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9867 {
9868         struct ctl_lun *lun = CTL_LUN(ctsio);
9869         struct scsi_vpd_logical_block_prov *lbp_ptr;
9870         const char *value;
9871
9872         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9873         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9874         ctsio->kern_sg_entries = 0;
9875         ctsio->kern_rel_offset = 0;
9876         ctsio->kern_data_len = min(sizeof(*lbp_ptr), alloc_len);
9877         ctsio->kern_total_len = ctsio->kern_data_len;
9878
9879         /*
9880          * The control device is always connected.  The disk device, on the
9881          * other hand, may not be online all the time.  Need to change this
9882          * to figure out whether the disk device is actually online or not.
9883          */
9884         if (lun != NULL)
9885                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9886                                   lun->be_lun->lun_type;
9887         else
9888                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9889
9890         lbp_ptr->page_code = SVPD_LBP;
9891         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9892         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
9893         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9894                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
9895                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
9896                 value = dnvlist_get_string(lun->be_lun->options,
9897                     "provisioning_type", NULL);
9898                 if (value != NULL) {
9899                         if (strcmp(value, "resource") == 0)
9900                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
9901                         else if (strcmp(value, "thin") == 0)
9902                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
9903                 } else
9904                         lbp_ptr->prov_type = SVPD_LBP_THIN;
9905         }
9906
9907         ctl_set_success(ctsio);
9908         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9909         ctsio->be_move_done = ctl_config_move_done;
9910         ctl_datamove((union ctl_io *)ctsio);
9911         return (CTL_RETVAL_COMPLETE);
9912 }
9913
9914 /*
9915  * INQUIRY with the EVPD bit set.
9916  */
9917 static int
9918 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9919 {
9920         struct ctl_lun *lun = CTL_LUN(ctsio);
9921         struct scsi_inquiry *cdb;
9922         int alloc_len, retval;
9923
9924         cdb = (struct scsi_inquiry *)ctsio->cdb;
9925         alloc_len = scsi_2btoul(cdb->length);
9926
9927         switch (cdb->page_code) {
9928         case SVPD_SUPPORTED_PAGES:
9929                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9930                 break;
9931         case SVPD_UNIT_SERIAL_NUMBER:
9932                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9933                 break;
9934         case SVPD_DEVICE_ID:
9935                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9936                 break;
9937         case SVPD_EXTENDED_INQUIRY_DATA:
9938                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
9939                 break;
9940         case SVPD_MODE_PAGE_POLICY:
9941                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
9942                 break;
9943         case SVPD_SCSI_PORTS:
9944                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
9945                 break;
9946         case SVPD_SCSI_TPC:
9947                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
9948                 break;
9949         case SVPD_SCSI_SFS:
9950                 retval = ctl_inquiry_evpd_sfs(ctsio, alloc_len);
9951                 break;
9952         case SVPD_BLOCK_LIMITS:
9953                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9954                         goto err;
9955                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9956                 break;
9957         case SVPD_BDC:
9958                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9959                         goto err;
9960                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
9961                 break;
9962         case SVPD_LBP:
9963                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9964                         goto err;
9965                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9966                 break;
9967         default:
9968 err:
9969                 ctl_set_invalid_field(ctsio,
9970                                       /*sks_valid*/ 1,
9971                                       /*command*/ 1,
9972                                       /*field*/ 2,
9973                                       /*bit_valid*/ 0,
9974                                       /*bit*/ 0);
9975                 ctl_done((union ctl_io *)ctsio);
9976                 retval = CTL_RETVAL_COMPLETE;
9977                 break;
9978         }
9979
9980         return (retval);
9981 }
9982
9983 /*
9984  * Standard INQUIRY data.
9985  */
9986 static int
9987 ctl_inquiry_std(struct ctl_scsiio *ctsio)
9988 {
9989         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9990         struct ctl_port *port = CTL_PORT(ctsio);
9991         struct ctl_lun *lun = CTL_LUN(ctsio);
9992         struct scsi_inquiry_data *inq_ptr;
9993         struct scsi_inquiry *cdb;
9994         const char *val;
9995         uint32_t alloc_len, data_len;
9996         ctl_port_type port_type;
9997
9998         port_type = port->port_type;
9999         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10000                 port_type = CTL_PORT_SCSI;
10001
10002         cdb = (struct scsi_inquiry *)ctsio->cdb;
10003         alloc_len = scsi_2btoul(cdb->length);
10004
10005         /*
10006          * We malloc the full inquiry data size here and fill it
10007          * in.  If the user only asks for less, we'll give him
10008          * that much.
10009          */
10010         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10011         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10012         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10013         ctsio->kern_sg_entries = 0;
10014         ctsio->kern_rel_offset = 0;
10015         ctsio->kern_data_len = min(data_len, alloc_len);
10016         ctsio->kern_total_len = ctsio->kern_data_len;
10017
10018         if (lun != NULL) {
10019                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10020                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10021                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10022                             lun->be_lun->lun_type;
10023                 } else {
10024                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10025                             lun->be_lun->lun_type;
10026                 }
10027                 if (lun->flags & CTL_LUN_REMOVABLE)
10028                         inq_ptr->dev_qual2 |= SID_RMB;
10029         } else
10030                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10031
10032         /* RMB in byte 2 is 0 */
10033         inq_ptr->version = SCSI_REV_SPC5;
10034
10035         /*
10036          * According to SAM-3, even if a device only supports a single
10037          * level of LUN addressing, it should still set the HISUP bit:
10038          *
10039          * 4.9.1 Logical unit numbers overview
10040          *
10041          * All logical unit number formats described in this standard are
10042          * hierarchical in structure even when only a single level in that
10043          * hierarchy is used. The HISUP bit shall be set to one in the
10044          * standard INQUIRY data (see SPC-2) when any logical unit number
10045          * format described in this standard is used.  Non-hierarchical
10046          * formats are outside the scope of this standard.
10047          *
10048          * Therefore we set the HiSup bit here.
10049          *
10050          * The response format is 2, per SPC-3.
10051          */
10052         inq_ptr->response_format = SID_HiSup | 2;
10053
10054         inq_ptr->additional_length = data_len -
10055             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10056         CTL_DEBUG_PRINT(("additional_length = %d\n",
10057                          inq_ptr->additional_length));
10058
10059         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10060         if (port_type == CTL_PORT_SCSI)
10061                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10062         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10063         inq_ptr->flags = SID_CmdQue;
10064         if (port_type == CTL_PORT_SCSI)
10065                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10066
10067         /*
10068          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10069          * We have 8 bytes for the vendor name, and 16 bytes for the device
10070          * name and 4 bytes for the revision.
10071          */
10072         if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options,
10073             "vendor", NULL)) == NULL) {
10074                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10075         } else {
10076                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10077                 strncpy(inq_ptr->vendor, val,
10078                     min(sizeof(inq_ptr->vendor), strlen(val)));
10079         }
10080         if (lun == NULL) {
10081                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10082                     sizeof(inq_ptr->product));
10083         } else if ((val = dnvlist_get_string(lun->be_lun->options, "product",
10084             NULL)) == NULL) {
10085                 switch (lun->be_lun->lun_type) {
10086                 case T_DIRECT:
10087                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10088                             sizeof(inq_ptr->product));
10089                         break;
10090                 case T_PROCESSOR:
10091                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10092                             sizeof(inq_ptr->product));
10093                         break;
10094                 case T_CDROM:
10095                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10096                             sizeof(inq_ptr->product));
10097                         break;
10098                 default:
10099                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10100                             sizeof(inq_ptr->product));
10101                         break;
10102                 }
10103         } else {
10104                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10105                 strncpy(inq_ptr->product, val,
10106                     min(sizeof(inq_ptr->product), strlen(val)));
10107         }
10108
10109         /*
10110          * XXX make this a macro somewhere so it automatically gets
10111          * incremented when we make changes.
10112          */
10113         if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options,
10114             "revision", NULL)) == NULL) {
10115                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10116         } else {
10117                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10118                 strncpy(inq_ptr->revision, val,
10119                     min(sizeof(inq_ptr->revision), strlen(val)));
10120         }
10121
10122         /*
10123          * For parallel SCSI, we support double transition and single
10124          * transition clocking.  We also support QAS (Quick Arbitration
10125          * and Selection) and Information Unit transfers on both the
10126          * control and array devices.
10127          */
10128         if (port_type == CTL_PORT_SCSI)
10129                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10130                                     SID_SPI_IUS;
10131
10132         /* SAM-6 (no version claimed) */
10133         scsi_ulto2b(0x00C0, inq_ptr->version1);
10134         /* SPC-5 (no version claimed) */
10135         scsi_ulto2b(0x05C0, inq_ptr->version2);
10136         if (port_type == CTL_PORT_FC) {
10137                 /* FCP-2 ANSI INCITS.350:2003 */
10138                 scsi_ulto2b(0x0917, inq_ptr->version3);
10139         } else if (port_type == CTL_PORT_SCSI) {
10140                 /* SPI-4 ANSI INCITS.362:200x */
10141                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10142         } else if (port_type == CTL_PORT_ISCSI) {
10143                 /* iSCSI (no version claimed) */
10144                 scsi_ulto2b(0x0960, inq_ptr->version3);
10145         } else if (port_type == CTL_PORT_SAS) {
10146                 /* SAS (no version claimed) */
10147                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10148         } else if (port_type == CTL_PORT_UMASS) {
10149                 /* USB Mass Storage Class Bulk-Only Transport, Revision 1.0 */
10150                 scsi_ulto2b(0x1730, inq_ptr->version3);
10151         }
10152
10153         if (lun == NULL) {
10154                 /* SBC-4 (no version claimed) */
10155                 scsi_ulto2b(0x0600, inq_ptr->version4);
10156         } else {
10157                 switch (lun->be_lun->lun_type) {
10158                 case T_DIRECT:
10159                         /* SBC-4 (no version claimed) */
10160                         scsi_ulto2b(0x0600, inq_ptr->version4);
10161                         break;
10162                 case T_PROCESSOR:
10163                         break;
10164                 case T_CDROM:
10165                         /* MMC-6 (no version claimed) */
10166                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10167                         break;
10168                 default:
10169                         break;
10170                 }
10171         }
10172
10173         ctl_set_success(ctsio);
10174         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10175         ctsio->be_move_done = ctl_config_move_done;
10176         ctl_datamove((union ctl_io *)ctsio);
10177         return (CTL_RETVAL_COMPLETE);
10178 }
10179
10180 int
10181 ctl_inquiry(struct ctl_scsiio *ctsio)
10182 {
10183         struct scsi_inquiry *cdb;
10184         int retval;
10185
10186         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10187
10188         cdb = (struct scsi_inquiry *)ctsio->cdb;
10189         if (cdb->byte2 & SI_EVPD)
10190                 retval = ctl_inquiry_evpd(ctsio);
10191         else if (cdb->page_code == 0)
10192                 retval = ctl_inquiry_std(ctsio);
10193         else {
10194                 ctl_set_invalid_field(ctsio,
10195                                       /*sks_valid*/ 1,
10196                                       /*command*/ 1,
10197                                       /*field*/ 2,
10198                                       /*bit_valid*/ 0,
10199                                       /*bit*/ 0);
10200                 ctl_done((union ctl_io *)ctsio);
10201                 return (CTL_RETVAL_COMPLETE);
10202         }
10203
10204         return (retval);
10205 }
10206
10207 int
10208 ctl_get_config(struct ctl_scsiio *ctsio)
10209 {
10210         struct ctl_lun *lun = CTL_LUN(ctsio);
10211         struct scsi_get_config_header *hdr;
10212         struct scsi_get_config_feature *feature;
10213         struct scsi_get_config *cdb;
10214         uint32_t alloc_len, data_len;
10215         int rt, starting;
10216
10217         cdb = (struct scsi_get_config *)ctsio->cdb;
10218         rt = (cdb->rt & SGC_RT_MASK);
10219         starting = scsi_2btoul(cdb->starting_feature);
10220         alloc_len = scsi_2btoul(cdb->length);
10221
10222         data_len = sizeof(struct scsi_get_config_header) +
10223             sizeof(struct scsi_get_config_feature) + 8 +
10224             sizeof(struct scsi_get_config_feature) + 8 +
10225             sizeof(struct scsi_get_config_feature) + 4 +
10226             sizeof(struct scsi_get_config_feature) + 4 +
10227             sizeof(struct scsi_get_config_feature) + 8 +
10228             sizeof(struct scsi_get_config_feature) +
10229             sizeof(struct scsi_get_config_feature) + 4 +
10230             sizeof(struct scsi_get_config_feature) + 4 +
10231             sizeof(struct scsi_get_config_feature) + 4 +
10232             sizeof(struct scsi_get_config_feature) + 4 +
10233             sizeof(struct scsi_get_config_feature) + 4 +
10234             sizeof(struct scsi_get_config_feature) + 4;
10235         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10236         ctsio->kern_sg_entries = 0;
10237         ctsio->kern_rel_offset = 0;
10238
10239         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10240         if (lun->flags & CTL_LUN_NO_MEDIA)
10241                 scsi_ulto2b(0x0000, hdr->current_profile);
10242         else
10243                 scsi_ulto2b(0x0010, hdr->current_profile);
10244         feature = (struct scsi_get_config_feature *)(hdr + 1);
10245
10246         if (starting > 0x003b)
10247                 goto done;
10248         if (starting > 0x003a)
10249                 goto f3b;
10250         if (starting > 0x002b)
10251                 goto f3a;
10252         if (starting > 0x002a)
10253                 goto f2b;
10254         if (starting > 0x001f)
10255                 goto f2a;
10256         if (starting > 0x001e)
10257                 goto f1f;
10258         if (starting > 0x001d)
10259                 goto f1e;
10260         if (starting > 0x0010)
10261                 goto f1d;
10262         if (starting > 0x0003)
10263                 goto f10;
10264         if (starting > 0x0002)
10265                 goto f3;
10266         if (starting > 0x0001)
10267                 goto f2;
10268         if (starting > 0x0000)
10269                 goto f1;
10270
10271         /* Profile List */
10272         scsi_ulto2b(0x0000, feature->feature_code);
10273         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10274         feature->add_length = 8;
10275         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10276         feature->feature_data[2] = 0x00;
10277         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10278         feature->feature_data[6] = 0x01;
10279         feature = (struct scsi_get_config_feature *)
10280             &feature->feature_data[feature->add_length];
10281
10282 f1:     /* Core */
10283         scsi_ulto2b(0x0001, feature->feature_code);
10284         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10285         feature->add_length = 8;
10286         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10287         feature->feature_data[4] = 0x03;
10288         feature = (struct scsi_get_config_feature *)
10289             &feature->feature_data[feature->add_length];
10290
10291 f2:     /* Morphing */
10292         scsi_ulto2b(0x0002, feature->feature_code);
10293         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10294         feature->add_length = 4;
10295         feature->feature_data[0] = 0x02;
10296         feature = (struct scsi_get_config_feature *)
10297             &feature->feature_data[feature->add_length];
10298
10299 f3:     /* Removable Medium */
10300         scsi_ulto2b(0x0003, feature->feature_code);
10301         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10302         feature->add_length = 4;
10303         feature->feature_data[0] = 0x39;
10304         feature = (struct scsi_get_config_feature *)
10305             &feature->feature_data[feature->add_length];
10306
10307         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10308                 goto done;
10309
10310 f10:    /* Random Read */
10311         scsi_ulto2b(0x0010, feature->feature_code);
10312         feature->flags = 0x00;
10313         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10314                 feature->flags |= SGC_F_CURRENT;
10315         feature->add_length = 8;
10316         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10317         scsi_ulto2b(1, &feature->feature_data[4]);
10318         feature->feature_data[6] = 0x00;
10319         feature = (struct scsi_get_config_feature *)
10320             &feature->feature_data[feature->add_length];
10321
10322 f1d:    /* Multi-Read */
10323         scsi_ulto2b(0x001D, feature->feature_code);
10324         feature->flags = 0x00;
10325         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10326                 feature->flags |= SGC_F_CURRENT;
10327         feature->add_length = 0;
10328         feature = (struct scsi_get_config_feature *)
10329             &feature->feature_data[feature->add_length];
10330
10331 f1e:    /* CD Read */
10332         scsi_ulto2b(0x001E, feature->feature_code);
10333         feature->flags = 0x00;
10334         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10335                 feature->flags |= SGC_F_CURRENT;
10336         feature->add_length = 4;
10337         feature->feature_data[0] = 0x00;
10338         feature = (struct scsi_get_config_feature *)
10339             &feature->feature_data[feature->add_length];
10340
10341 f1f:    /* DVD Read */
10342         scsi_ulto2b(0x001F, feature->feature_code);
10343         feature->flags = 0x08;
10344         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10345                 feature->flags |= SGC_F_CURRENT;
10346         feature->add_length = 4;
10347         feature->feature_data[0] = 0x01;
10348         feature->feature_data[2] = 0x03;
10349         feature = (struct scsi_get_config_feature *)
10350             &feature->feature_data[feature->add_length];
10351
10352 f2a:    /* DVD+RW */
10353         scsi_ulto2b(0x002A, feature->feature_code);
10354         feature->flags = 0x04;
10355         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10356                 feature->flags |= SGC_F_CURRENT;
10357         feature->add_length = 4;
10358         feature->feature_data[0] = 0x00;
10359         feature->feature_data[1] = 0x00;
10360         feature = (struct scsi_get_config_feature *)
10361             &feature->feature_data[feature->add_length];
10362
10363 f2b:    /* DVD+R */
10364         scsi_ulto2b(0x002B, feature->feature_code);
10365         feature->flags = 0x00;
10366         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10367                 feature->flags |= SGC_F_CURRENT;
10368         feature->add_length = 4;
10369         feature->feature_data[0] = 0x00;
10370         feature = (struct scsi_get_config_feature *)
10371             &feature->feature_data[feature->add_length];
10372
10373 f3a:    /* DVD+RW Dual Layer */
10374         scsi_ulto2b(0x003A, feature->feature_code);
10375         feature->flags = 0x00;
10376         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10377                 feature->flags |= SGC_F_CURRENT;
10378         feature->add_length = 4;
10379         feature->feature_data[0] = 0x00;
10380         feature->feature_data[1] = 0x00;
10381         feature = (struct scsi_get_config_feature *)
10382             &feature->feature_data[feature->add_length];
10383
10384 f3b:    /* DVD+R Dual Layer */
10385         scsi_ulto2b(0x003B, feature->feature_code);
10386         feature->flags = 0x00;
10387         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10388                 feature->flags |= SGC_F_CURRENT;
10389         feature->add_length = 4;
10390         feature->feature_data[0] = 0x00;
10391         feature = (struct scsi_get_config_feature *)
10392             &feature->feature_data[feature->add_length];
10393
10394 done:
10395         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10396         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10397                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10398                 if (scsi_2btoul(feature->feature_code) == starting)
10399                         feature = (struct scsi_get_config_feature *)
10400                             &feature->feature_data[feature->add_length];
10401                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10402         }
10403         scsi_ulto4b(data_len - 4, hdr->data_length);
10404         ctsio->kern_data_len = min(data_len, alloc_len);
10405         ctsio->kern_total_len = ctsio->kern_data_len;
10406
10407         ctl_set_success(ctsio);
10408         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10409         ctsio->be_move_done = ctl_config_move_done;
10410         ctl_datamove((union ctl_io *)ctsio);
10411         return (CTL_RETVAL_COMPLETE);
10412 }
10413
10414 int
10415 ctl_get_event_status(struct ctl_scsiio *ctsio)
10416 {
10417         struct scsi_get_event_status_header *hdr;
10418         struct scsi_get_event_status *cdb;
10419         uint32_t alloc_len, data_len;
10420
10421         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10422         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10423                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10424                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10425                 ctl_done((union ctl_io *)ctsio);
10426                 return (CTL_RETVAL_COMPLETE);
10427         }
10428         alloc_len = scsi_2btoul(cdb->length);
10429
10430         data_len = sizeof(struct scsi_get_event_status_header);
10431         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10432         ctsio->kern_sg_entries = 0;
10433         ctsio->kern_rel_offset = 0;
10434         ctsio->kern_data_len = min(data_len, alloc_len);
10435         ctsio->kern_total_len = ctsio->kern_data_len;
10436
10437         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10438         scsi_ulto2b(0, hdr->descr_length);
10439         hdr->nea_class = SGESN_NEA;
10440         hdr->supported_class = 0;
10441
10442         ctl_set_success(ctsio);
10443         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10444         ctsio->be_move_done = ctl_config_move_done;
10445         ctl_datamove((union ctl_io *)ctsio);
10446         return (CTL_RETVAL_COMPLETE);
10447 }
10448
10449 int
10450 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10451 {
10452         struct scsi_mechanism_status_header *hdr;
10453         struct scsi_mechanism_status *cdb;
10454         uint32_t alloc_len, data_len;
10455
10456         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10457         alloc_len = scsi_2btoul(cdb->length);
10458
10459         data_len = sizeof(struct scsi_mechanism_status_header);
10460         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10461         ctsio->kern_sg_entries = 0;
10462         ctsio->kern_rel_offset = 0;
10463         ctsio->kern_data_len = min(data_len, alloc_len);
10464         ctsio->kern_total_len = ctsio->kern_data_len;
10465
10466         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10467         hdr->state1 = 0x00;
10468         hdr->state2 = 0xe0;
10469         scsi_ulto3b(0, hdr->lba);
10470         hdr->slots_num = 0;
10471         scsi_ulto2b(0, hdr->slots_length);
10472
10473         ctl_set_success(ctsio);
10474         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10475         ctsio->be_move_done = ctl_config_move_done;
10476         ctl_datamove((union ctl_io *)ctsio);
10477         return (CTL_RETVAL_COMPLETE);
10478 }
10479
10480 static void
10481 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10482 {
10483
10484         lba += 150;
10485         buf[0] = 0;
10486         buf[1] = bin2bcd((lba / 75) / 60);
10487         buf[2] = bin2bcd((lba / 75) % 60);
10488         buf[3] = bin2bcd(lba % 75);
10489 }
10490
10491 int
10492 ctl_read_toc(struct ctl_scsiio *ctsio)
10493 {
10494         struct ctl_lun *lun = CTL_LUN(ctsio);
10495         struct scsi_read_toc_hdr *hdr;
10496         struct scsi_read_toc_type01_descr *descr;
10497         struct scsi_read_toc *cdb;
10498         uint32_t alloc_len, data_len;
10499         int format, msf;
10500
10501         cdb = (struct scsi_read_toc *)ctsio->cdb;
10502         msf = (cdb->byte2 & CD_MSF) != 0;
10503         format = cdb->format;
10504         alloc_len = scsi_2btoul(cdb->data_len);
10505
10506         data_len = sizeof(struct scsi_read_toc_hdr);
10507         if (format == 0)
10508                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10509         else
10510                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10511         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10512         ctsio->kern_sg_entries = 0;
10513         ctsio->kern_rel_offset = 0;
10514         ctsio->kern_data_len = min(data_len, alloc_len);
10515         ctsio->kern_total_len = ctsio->kern_data_len;
10516
10517         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10518         if (format == 0) {
10519                 scsi_ulto2b(0x12, hdr->data_length);
10520                 hdr->first = 1;
10521                 hdr->last = 1;
10522                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10523                 descr->addr_ctl = 0x14;
10524                 descr->track_number = 1;
10525                 if (msf)
10526                         ctl_ultomsf(0, descr->track_start);
10527                 else
10528                         scsi_ulto4b(0, descr->track_start);
10529                 descr++;
10530                 descr->addr_ctl = 0x14;
10531                 descr->track_number = 0xaa;
10532                 if (msf)
10533                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10534                 else
10535                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10536         } else {
10537                 scsi_ulto2b(0x0a, hdr->data_length);
10538                 hdr->first = 1;
10539                 hdr->last = 1;
10540                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10541                 descr->addr_ctl = 0x14;
10542                 descr->track_number = 1;
10543                 if (msf)
10544                         ctl_ultomsf(0, descr->track_start);
10545                 else
10546                         scsi_ulto4b(0, descr->track_start);
10547         }
10548
10549         ctl_set_success(ctsio);
10550         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10551         ctsio->be_move_done = ctl_config_move_done;
10552         ctl_datamove((union ctl_io *)ctsio);
10553         return (CTL_RETVAL_COMPLETE);
10554 }
10555
10556 /*
10557  * For known CDB types, parse the LBA and length.
10558  */
10559 static int
10560 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10561 {
10562
10563         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
10564             ("%s: unexpected I/O type %x", __func__, io->io_hdr.io_type));
10565
10566         switch (io->scsiio.cdb[0]) {
10567         case COMPARE_AND_WRITE: {
10568                 struct scsi_compare_and_write *cdb;
10569
10570                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10571
10572                 *lba = scsi_8btou64(cdb->addr);
10573                 *len = cdb->length;
10574                 break;
10575         }
10576         case READ_6:
10577         case WRITE_6: {
10578                 struct scsi_rw_6 *cdb;
10579
10580                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10581
10582                 *lba = scsi_3btoul(cdb->addr);
10583                 /* only 5 bits are valid in the most significant address byte */
10584                 *lba &= 0x1fffff;
10585                 *len = cdb->length;
10586                 break;
10587         }
10588         case READ_10:
10589         case WRITE_10: {
10590                 struct scsi_rw_10 *cdb;
10591
10592                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10593
10594                 *lba = scsi_4btoul(cdb->addr);
10595                 *len = scsi_2btoul(cdb->length);
10596                 break;
10597         }
10598         case WRITE_VERIFY_10: {
10599                 struct scsi_write_verify_10 *cdb;
10600
10601                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10602
10603                 *lba = scsi_4btoul(cdb->addr);
10604                 *len = scsi_2btoul(cdb->length);
10605                 break;
10606         }
10607         case READ_12:
10608         case WRITE_12: {
10609                 struct scsi_rw_12 *cdb;
10610
10611                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10612
10613                 *lba = scsi_4btoul(cdb->addr);
10614                 *len = scsi_4btoul(cdb->length);
10615                 break;
10616         }
10617         case WRITE_VERIFY_12: {
10618                 struct scsi_write_verify_12 *cdb;
10619
10620                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10621
10622                 *lba = scsi_4btoul(cdb->addr);
10623                 *len = scsi_4btoul(cdb->length);
10624                 break;
10625         }
10626         case READ_16:
10627         case WRITE_16: {
10628                 struct scsi_rw_16 *cdb;
10629
10630                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10631
10632                 *lba = scsi_8btou64(cdb->addr);
10633                 *len = scsi_4btoul(cdb->length);
10634                 break;
10635         }
10636         case WRITE_ATOMIC_16: {
10637                 struct scsi_write_atomic_16 *cdb;
10638
10639                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10640
10641                 *lba = scsi_8btou64(cdb->addr);
10642                 *len = scsi_2btoul(cdb->length);
10643                 break;
10644         }
10645         case WRITE_VERIFY_16: {
10646                 struct scsi_write_verify_16 *cdb;
10647
10648                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10649
10650                 *lba = scsi_8btou64(cdb->addr);
10651                 *len = scsi_4btoul(cdb->length);
10652                 break;
10653         }
10654         case WRITE_SAME_10: {
10655                 struct scsi_write_same_10 *cdb;
10656
10657                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10658
10659                 *lba = scsi_4btoul(cdb->addr);
10660                 *len = scsi_2btoul(cdb->length);
10661                 break;
10662         }
10663         case WRITE_SAME_16: {
10664                 struct scsi_write_same_16 *cdb;
10665
10666                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10667
10668                 *lba = scsi_8btou64(cdb->addr);
10669                 *len = scsi_4btoul(cdb->length);
10670                 break;
10671         }
10672         case VERIFY_10: {
10673                 struct scsi_verify_10 *cdb;
10674
10675                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10676
10677                 *lba = scsi_4btoul(cdb->addr);
10678                 *len = scsi_2btoul(cdb->length);
10679                 break;
10680         }
10681         case VERIFY_12: {
10682                 struct scsi_verify_12 *cdb;
10683
10684                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10685
10686                 *lba = scsi_4btoul(cdb->addr);
10687                 *len = scsi_4btoul(cdb->length);
10688                 break;
10689         }
10690         case VERIFY_16: {
10691                 struct scsi_verify_16 *cdb;
10692
10693                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10694
10695                 *lba = scsi_8btou64(cdb->addr);
10696                 *len = scsi_4btoul(cdb->length);
10697                 break;
10698         }
10699         case UNMAP: {
10700                 *lba = 0;
10701                 *len = UINT64_MAX;
10702                 break;
10703         }
10704         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10705                 struct scsi_get_lba_status *cdb;
10706
10707                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10708                 *lba = scsi_8btou64(cdb->addr);
10709                 *len = UINT32_MAX;
10710                 break;
10711         }
10712         default:
10713                 *lba = 0;
10714                 *len = UINT64_MAX;
10715                 return (1);
10716         }
10717
10718         return (0);
10719 }
10720
10721 static ctl_action
10722 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10723     bool seq)
10724 {
10725         uint64_t endlba1, endlba2;
10726
10727         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10728         endlba2 = lba2 + len2 - 1;
10729
10730         if ((endlba1 < lba2) || (endlba2 < lba1))
10731                 return (CTL_ACTION_PASS);
10732         else
10733                 return (CTL_ACTION_BLOCK);
10734 }
10735
10736 static int
10737 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10738 {
10739         struct ctl_ptr_len_flags *ptrlen;
10740         struct scsi_unmap_desc *buf, *end, *range;
10741         uint64_t lba;
10742         uint32_t len;
10743
10744         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
10745             ("%s: unexpected I/O type %x", __func__, io->io_hdr.io_type));
10746
10747         /* If not UNMAP -- go other way. */
10748         if (io->scsiio.cdb[0] != UNMAP)
10749                 return (CTL_ACTION_SKIP);
10750
10751         /* If UNMAP without data -- block and wait for data. */
10752         ptrlen = (struct ctl_ptr_len_flags *)
10753             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10754         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10755             ptrlen->ptr == NULL)
10756                 return (CTL_ACTION_BLOCK);
10757
10758         /* UNMAP with data -- check for collision. */
10759         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10760         end = buf + ptrlen->len / sizeof(*buf);
10761         for (range = buf; range < end; range++) {
10762                 lba = scsi_8btou64(range->lba);
10763                 len = scsi_4btoul(range->length);
10764                 if ((lba < lba2 + len2) && (lba + len > lba2))
10765                         return (CTL_ACTION_BLOCK);
10766         }
10767         return (CTL_ACTION_PASS);
10768 }
10769
10770 static ctl_action
10771 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10772 {
10773         uint64_t lba1, lba2;
10774         uint64_t len1, len2;
10775         int retval;
10776
10777         retval = ctl_get_lba_len(io2, &lba2, &len2);
10778         KASSERT(retval == 0, ("ctl_get_lba_len() error"));
10779
10780         retval = ctl_extent_check_unmap(io1, lba2, len2);
10781         if (retval != CTL_ACTION_SKIP)
10782                 return (retval);
10783
10784         retval = ctl_get_lba_len(io1, &lba1, &len1);
10785         KASSERT(retval == 0, ("ctl_get_lba_len() error"));
10786
10787         if (seq && (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE))
10788                 seq = FALSE;
10789         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10790 }
10791
10792 static ctl_action
10793 ctl_seq_check(union ctl_io *io1, union ctl_io *io2)
10794 {
10795         uint64_t lba1, lba2;
10796         uint64_t len1, len2;
10797         int retval __diagused;
10798
10799         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10800                 return (CTL_ACTION_PASS);
10801         retval = ctl_get_lba_len(io1, &lba1, &len1);
10802         KASSERT(retval == 0, ("ctl_get_lba_len() error"));
10803         retval = ctl_get_lba_len(io2, &lba2, &len2);
10804         KASSERT(retval == 0, ("ctl_get_lba_len() error"));
10805
10806         if (lba1 + len1 == lba2)
10807                 return (CTL_ACTION_BLOCK);
10808         return (CTL_ACTION_PASS);
10809 }
10810
10811 static ctl_action
10812 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10813     const uint8_t *serialize_row, union ctl_io *ooa_io)
10814 {
10815
10816         /*
10817          * The initiator attempted multiple untagged commands at the same
10818          * time.  Can't do that.
10819          */
10820         if (__predict_false(pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10821          && __predict_false(ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10822          && ((pending_io->io_hdr.nexus.targ_port ==
10823               ooa_io->io_hdr.nexus.targ_port)
10824           && (pending_io->io_hdr.nexus.initid ==
10825               ooa_io->io_hdr.nexus.initid))
10826          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10827               CTL_FLAG_STATUS_SENT)) == 0))
10828                 return (CTL_ACTION_OVERLAP);
10829
10830         /*
10831          * The initiator attempted to send multiple tagged commands with
10832          * the same ID.  (It's fine if different initiators have the same
10833          * tag ID.)
10834          *
10835          * Even if all of those conditions are true, we don't kill the I/O
10836          * if the command ahead of us has been aborted.  We won't end up
10837          * sending it to the FETD, and it's perfectly legal to resend a
10838          * command with the same tag number as long as the previous
10839          * instance of this tag number has been aborted somehow.
10840          */
10841         if (__predict_true(pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10842          && __predict_true(ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10843          && __predict_false(pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10844          && ((pending_io->io_hdr.nexus.targ_port ==
10845               ooa_io->io_hdr.nexus.targ_port)
10846           && (pending_io->io_hdr.nexus.initid ==
10847               ooa_io->io_hdr.nexus.initid))
10848          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10849               CTL_FLAG_STATUS_SENT)) == 0))
10850                 return (CTL_ACTION_OVERLAP_TAG);
10851
10852         /*
10853          * If we get a head of queue tag, SAM-3 says that we should
10854          * immediately execute it.
10855          *
10856          * What happens if this command would normally block for some other
10857          * reason?  e.g. a request sense with a head of queue tag
10858          * immediately after a write.  Normally that would block, but this
10859          * will result in its getting executed immediately...
10860          *
10861          * We currently return "pass" instead of "skip", so we'll end up
10862          * going through the rest of the queue to check for overlapped tags.
10863          *
10864          * XXX KDM check for other types of blockage first??
10865          */
10866         if (__predict_false(pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE))
10867                 return (CTL_ACTION_PASS);
10868
10869         /*
10870          * Simple tags get blocked until all head of queue and ordered tags
10871          * ahead of them have completed.  I'm lumping untagged commands in
10872          * with simple tags here.  XXX KDM is that the right thing to do?
10873          */
10874         if (__predict_false(ooa_io->scsiio.tag_type == CTL_TAG_ORDERED) ||
10875             __predict_false(ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE))
10876                 return (CTL_ACTION_BLOCK);
10877
10878         /* Unsupported command in OOA queue. */
10879         if (__predict_false(ooa_io->scsiio.seridx == CTL_SERIDX_INVLD))
10880                 return (CTL_ACTION_PASS);
10881
10882         switch (serialize_row[ooa_io->scsiio.seridx]) {
10883         case CTL_SER_SEQ:
10884                 if (lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10885                         return (ctl_seq_check(ooa_io, pending_io));
10886                 /* FALLTHROUGH */
10887         case CTL_SER_PASS:
10888                 return (CTL_ACTION_PASS);
10889         case CTL_SER_EXTENTOPT:
10890                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) ==
10891                     SCP_QUEUE_ALG_UNRESTRICTED)
10892                         return (CTL_ACTION_PASS);
10893                 /* FALLTHROUGH */
10894         case CTL_SER_EXTENT:
10895                 return (ctl_extent_check(ooa_io, pending_io,
10896                     (lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10897         case CTL_SER_BLOCKOPT:
10898                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) ==
10899                     SCP_QUEUE_ALG_UNRESTRICTED)
10900                         return (CTL_ACTION_PASS);
10901                 /* FALLTHROUGH */
10902         case CTL_SER_BLOCK:
10903                 return (CTL_ACTION_BLOCK);
10904         default:
10905                 __assert_unreachable();
10906         }
10907 }
10908
10909 /*
10910  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10911  * Assumptions:
10912  * - pending_io is generally either incoming, or on the blocked queue
10913  * - starting I/O is the I/O we want to start the check with.
10914  */
10915 static ctl_action
10916 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10917               union ctl_io **starting_io)
10918 {
10919         union ctl_io *ooa_io = *starting_io;
10920         const uint8_t *serialize_row;
10921         ctl_action action;
10922
10923         mtx_assert(&lun->lun_lock, MA_OWNED);
10924
10925         /*
10926          * Aborted commands are not going to be executed and may even
10927          * not report completion, so we don't care about their order.
10928          * Let them complete ASAP to clean the OOA queue.
10929          */
10930         if (__predict_false(pending_io->io_hdr.flags & CTL_FLAG_ABORT))
10931                 return (CTL_ACTION_SKIP);
10932
10933         /*
10934          * Ordered tags have to block until all items ahead of them have
10935          * completed.  If we get called with an ordered tag, we always
10936          * block, if something else is ahead of us in the queue.
10937          */
10938         if ((pending_io->scsiio.tag_type == CTL_TAG_ORDERED) &&
10939             (ooa_io != NULL))
10940                 return (CTL_ACTION_BLOCK);
10941
10942         serialize_row = ctl_serialize_table[pending_io->scsiio.seridx];
10943
10944         /*
10945          * Run back along the OOA queue, starting with the current
10946          * blocked I/O and going through every I/O before it on the
10947          * queue.  If starting_io is NULL, we'll just end up returning
10948          * CTL_ACTION_PASS.
10949          */
10950         for (; ooa_io != NULL;
10951              ooa_io = (union ctl_io *)LIST_NEXT(&ooa_io->io_hdr, ooa_links)) {
10952                 action = ctl_check_for_blockage(lun, pending_io, serialize_row,
10953                     ooa_io);
10954                 if (action != CTL_ACTION_PASS) {
10955                         *starting_io = ooa_io;
10956                         return (action);
10957                 }
10958         }
10959
10960         *starting_io = NULL;
10961         return (CTL_ACTION_PASS);
10962 }
10963
10964 /*
10965  * Try to unblock the specified I/O.
10966  *
10967  * skip parameter allows explicitly skip present blocker of the I/O,
10968  * starting from the previous one on OOA queue.  It can be used when
10969  * we know for sure that the blocker I/O does no longer count.
10970  */
10971 static void
10972 ctl_try_unblock_io(struct ctl_lun *lun, union ctl_io *io, bool skip)
10973 {
10974         struct ctl_softc *softc = lun->ctl_softc;
10975         union ctl_io *bio, *obio;
10976         const struct ctl_cmd_entry *entry;
10977         union ctl_ha_msg msg_info;
10978         ctl_action action;
10979
10980         mtx_assert(&lun->lun_lock, MA_OWNED);
10981
10982         if (io->io_hdr.blocker == NULL)
10983                 return;
10984
10985         obio = bio = io->io_hdr.blocker;
10986         if (skip)
10987                 bio = (union ctl_io *)LIST_NEXT(&bio->io_hdr, ooa_links);
10988         action = ctl_check_ooa(lun, io, &bio);
10989         if (action == CTL_ACTION_BLOCK) {
10990                 /* Still blocked, but may be by different I/O now. */
10991                 if (bio != obio) {
10992                         TAILQ_REMOVE(&obio->io_hdr.blocked_queue,
10993                             &io->io_hdr, blocked_links);
10994                         TAILQ_INSERT_TAIL(&bio->io_hdr.blocked_queue,
10995                             &io->io_hdr, blocked_links);
10996                         io->io_hdr.blocker = bio;
10997                 }
10998                 return;
10999         }
11000
11001         /* No longer blocked, one way or another. */
11002         TAILQ_REMOVE(&obio->io_hdr.blocked_queue, &io->io_hdr, blocked_links);
11003         io->io_hdr.blocker = NULL;
11004
11005         switch (action) {
11006         case CTL_ACTION_PASS:
11007         case CTL_ACTION_SKIP:
11008
11009                 /* Serializing commands from the other SC retire there. */
11010                 if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) &&
11011                     (softc->ha_mode != CTL_HA_MODE_XFER)) {
11012                         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11013                         msg_info.hdr.original_sc = io->io_hdr.remote_io;
11014                         msg_info.hdr.serializing_sc = io;
11015                         msg_info.hdr.msg_type = CTL_MSG_R2R;
11016                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11017                             sizeof(msg_info.hdr), M_NOWAIT);
11018                         break;
11019                 }
11020
11021                 /*
11022                  * Check this I/O for LUN state changes that may have happened
11023                  * while this command was blocked. The LUN state may have been
11024                  * changed by a command ahead of us in the queue.
11025                  */
11026                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
11027                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
11028                         ctl_done(io);
11029                         break;
11030                 }
11031
11032                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11033                 ctl_enqueue_rtr(io);
11034                 break;
11035         default:
11036                 __assert_unreachable();
11037         case CTL_ACTION_OVERLAP:
11038                 ctl_set_overlapped_cmd(&io->scsiio);
11039                 goto error;
11040         case CTL_ACTION_OVERLAP_TAG:
11041                 ctl_set_overlapped_tag(&io->scsiio,
11042                     io->scsiio.tag_num & 0xff);
11043 error:
11044                 /* Serializing commands from the other SC are done here. */
11045                 if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) &&
11046                     (softc->ha_mode != CTL_HA_MODE_XFER)) {
11047                         ctl_try_unblock_others(lun, io, TRUE);
11048                         LIST_REMOVE(&io->io_hdr, ooa_links);
11049
11050                         ctl_copy_sense_data_back(io, &msg_info);
11051                         msg_info.hdr.original_sc = io->io_hdr.remote_io;
11052                         msg_info.hdr.serializing_sc = NULL;
11053                         msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
11054                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11055                             sizeof(msg_info.scsi), M_WAITOK);
11056                         ctl_free_io(io);
11057                         break;
11058                 }
11059
11060                 ctl_done(io);
11061                 break;
11062         }
11063 }
11064
11065 /*
11066  * Try to unblock I/Os blocked by the specified I/O.
11067  *
11068  * skip parameter allows explicitly skip the specified I/O as blocker,
11069  * starting from the previous one on the OOA queue.  It can be used when
11070  * we know for sure that the specified I/O does no longer count (done).
11071  * It has to be still on OOA queue though so that we know where to start.
11072  */
11073 static void
11074 ctl_try_unblock_others(struct ctl_lun *lun, union ctl_io *bio, bool skip)
11075 {
11076         union ctl_io *io, *next_io;
11077
11078         mtx_assert(&lun->lun_lock, MA_OWNED);
11079
11080         for (io = (union ctl_io *)TAILQ_FIRST(&bio->io_hdr.blocked_queue);
11081              io != NULL; io = next_io) {
11082                 next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, blocked_links);
11083
11084                 KASSERT(io->io_hdr.blocker != NULL,
11085                     ("I/O %p on blocked list without blocker", io));
11086                 ctl_try_unblock_io(lun, io, skip);
11087         }
11088         KASSERT(!skip || TAILQ_EMPTY(&bio->io_hdr.blocked_queue),
11089             ("blocked_queue is not empty after skipping %p", bio));
11090 }
11091
11092 /*
11093  * This routine (with one exception) checks LUN flags that can be set by
11094  * commands ahead of us in the OOA queue.  These flags have to be checked
11095  * when a command initially comes in, and when we pull a command off the
11096  * blocked queue and are preparing to execute it.  The reason we have to
11097  * check these flags for commands on the blocked queue is that the LUN
11098  * state may have been changed by a command ahead of us while we're on the
11099  * blocked queue.
11100  *
11101  * Ordering is somewhat important with these checks, so please pay
11102  * careful attention to the placement of any new checks.
11103  */
11104 static int
11105 ctl_scsiio_lun_check(struct ctl_lun *lun,
11106     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11107 {
11108         struct ctl_softc *softc = lun->ctl_softc;
11109         int retval;
11110         uint32_t residx;
11111
11112         retval = 0;
11113
11114         mtx_assert(&lun->lun_lock, MA_OWNED);
11115
11116         /*
11117          * If this shelf is a secondary shelf controller, we may have to
11118          * reject some commands disallowed by HA mode and link state.
11119          */
11120         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11121                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11122                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11123                         ctl_set_lun_unavail(ctsio);
11124                         retval = 1;
11125                         goto bailout;
11126                 }
11127                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11128                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11129                         ctl_set_lun_transit(ctsio);
11130                         retval = 1;
11131                         goto bailout;
11132                 }
11133                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11134                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11135                         ctl_set_lun_standby(ctsio);
11136                         retval = 1;
11137                         goto bailout;
11138                 }
11139
11140                 /* The rest of checks are only done on executing side */
11141                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11142                         goto bailout;
11143         }
11144
11145         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11146                 if (lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11147                         ctl_set_hw_write_protected(ctsio);
11148                         retval = 1;
11149                         goto bailout;
11150                 }
11151                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11152                         ctl_set_sense(ctsio, /*current_error*/ 1,
11153                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11154                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11155                         retval = 1;
11156                         goto bailout;
11157                 }
11158         }
11159
11160         /*
11161          * Check for a reservation conflict.  If this command isn't allowed
11162          * even on reserved LUNs, and if this initiator isn't the one who
11163          * reserved us, reject the command with a reservation conflict.
11164          */
11165         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11166         if ((lun->flags & CTL_LUN_RESERVED)
11167          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11168                 if (lun->res_idx != residx) {
11169                         ctl_set_reservation_conflict(ctsio);
11170                         retval = 1;
11171                         goto bailout;
11172                 }
11173         }
11174
11175         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11176             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11177                 /* No reservation or command is allowed. */;
11178         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11179             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11180              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11181              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11182                 /* The command is allowed for Write Exclusive resv. */;
11183         } else {
11184                 /*
11185                  * if we aren't registered or it's a res holder type
11186                  * reservation and this isn't the res holder then set a
11187                  * conflict.
11188                  */
11189                 if (ctl_get_prkey(lun, residx) == 0 ||
11190                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11191                         ctl_set_reservation_conflict(ctsio);
11192                         retval = 1;
11193                         goto bailout;
11194                 }
11195         }
11196
11197         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11198                 if (lun->flags & CTL_LUN_EJECTED)
11199                         ctl_set_lun_ejected(ctsio);
11200                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11201                         if (lun->flags & CTL_LUN_REMOVABLE)
11202                                 ctl_set_lun_no_media(ctsio);
11203                         else
11204                                 ctl_set_lun_int_reqd(ctsio);
11205                 } else if (lun->flags & CTL_LUN_STOPPED)
11206                         ctl_set_lun_stopped(ctsio);
11207                 else
11208                         goto bailout;
11209                 retval = 1;
11210                 goto bailout;
11211         }
11212
11213 bailout:
11214         return (retval);
11215 }
11216
11217 static void
11218 ctl_failover_io(union ctl_io *io, int have_lock)
11219 {
11220         ctl_set_busy(&io->scsiio);
11221         ctl_done(io);
11222 }
11223
11224 static void
11225 ctl_failover_lun(union ctl_io *rio)
11226 {
11227         struct ctl_softc *softc = CTL_SOFTC(rio);
11228         struct ctl_lun *lun;
11229         struct ctl_io_hdr *io, *next_io;
11230         uint32_t targ_lun;
11231
11232         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11233         CTL_DEBUG_PRINT(("FAILOVER for lun %u\n", targ_lun));
11234
11235         /* Find and lock the LUN. */
11236         mtx_lock(&softc->ctl_lock);
11237         if (targ_lun > ctl_max_luns ||
11238             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11239                 mtx_unlock(&softc->ctl_lock);
11240                 return;
11241         }
11242         mtx_lock(&lun->lun_lock);
11243         mtx_unlock(&softc->ctl_lock);
11244         if (lun->flags & CTL_LUN_DISABLED) {
11245                 mtx_unlock(&lun->lun_lock);
11246                 return;
11247         }
11248
11249         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11250                 LIST_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11251                         /* We are master */
11252                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11253                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11254                                         io->flags |= CTL_FLAG_ABORT |
11255                                             CTL_FLAG_FAILOVER;
11256                                         ctl_try_unblock_io(lun,
11257                                             (union ctl_io *)io, FALSE);
11258                                 } else { /* This can be only due to DATAMOVE */
11259                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11260                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11261                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11262                                         io->port_status = 31340;
11263                                         ctl_enqueue_isc((union ctl_io *)io);
11264                                 }
11265                         } else
11266                         /* We are slave */
11267                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11268                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11269                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11270                                         io->flags |= CTL_FLAG_FAILOVER;
11271                                 } else {
11272                                         ctl_set_busy(&((union ctl_io *)io)->
11273                                             scsiio);
11274                                         ctl_done((union ctl_io *)io);
11275                                 }
11276                         }
11277                 }
11278         } else { /* SERIALIZE modes */
11279                 LIST_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11280                         /* We are master */
11281                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11282                                 if (io->blocker != NULL) {
11283                                         TAILQ_REMOVE(&io->blocker->io_hdr.blocked_queue,
11284                                             io, blocked_links);
11285                                         io->blocker = NULL;
11286                                 }
11287                                 ctl_try_unblock_others(lun, (union ctl_io *)io,
11288                                     TRUE);
11289                                 LIST_REMOVE(io, ooa_links);
11290                                 ctl_free_io((union ctl_io *)io);
11291                         } else
11292                         /* We are slave */
11293                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11294                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11295                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11296                                         ctl_set_busy(&((union ctl_io *)io)->
11297                                             scsiio);
11298                                         ctl_done((union ctl_io *)io);
11299                                 }
11300                         }
11301                 }
11302         }
11303         mtx_unlock(&lun->lun_lock);
11304 }
11305
11306 static void
11307 ctl_scsiio_precheck(struct ctl_scsiio *ctsio)
11308 {
11309         struct ctl_softc *softc = CTL_SOFTC(ctsio);
11310         struct ctl_lun *lun;
11311         const struct ctl_cmd_entry *entry;
11312         union ctl_io *bio;
11313         uint32_t initidx, targ_lun;
11314
11315         lun = NULL;
11316         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11317         if (targ_lun < ctl_max_luns)
11318                 lun = softc->ctl_luns[targ_lun];
11319         if (lun) {
11320                 /*
11321                  * If the LUN is invalid, pretend that it doesn't exist.
11322                  * It will go away as soon as all pending I/O has been
11323                  * completed.
11324                  */
11325                 mtx_lock(&lun->lun_lock);
11326                 if (lun->flags & CTL_LUN_DISABLED) {
11327                         mtx_unlock(&lun->lun_lock);
11328                         lun = NULL;
11329                 }
11330         }
11331         CTL_LUN(ctsio) = lun;
11332         if (lun) {
11333                 CTL_BACKEND_LUN(ctsio) = lun->be_lun;
11334
11335                 /*
11336                  * Every I/O goes into the OOA queue for a particular LUN,
11337                  * and stays there until completion.
11338                  */
11339 #ifdef CTL_TIME_IO
11340                 if (LIST_EMPTY(&lun->ooa_queue))
11341                         lun->idle_time += getsbinuptime() - lun->last_busy;
11342 #endif
11343                 LIST_INSERT_HEAD(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11344         }
11345
11346         /* Get command entry and return error if it is unsuppotyed. */
11347         entry = ctl_validate_command(ctsio);
11348         if (entry == NULL) {
11349                 if (lun)
11350                         mtx_unlock(&lun->lun_lock);
11351                 return;
11352         }
11353
11354         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11355         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11356
11357         /*
11358          * Check to see whether we can send this command to LUNs that don't
11359          * exist.  This should pretty much only be the case for inquiry
11360          * and request sense.  Further checks, below, really require having
11361          * a LUN, so we can't really check the command anymore.  Just put
11362          * it on the rtr queue.
11363          */
11364         if (lun == NULL) {
11365                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11366                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11367                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11368                         return;
11369                 }
11370
11371                 ctl_set_unsupported_lun(ctsio);
11372                 ctl_done((union ctl_io *)ctsio);
11373                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11374                 return;
11375         } else {
11376                 /*
11377                  * Make sure we support this particular command on this LUN.
11378                  * e.g., we don't support writes to the control LUN.
11379                  */
11380                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11381                         mtx_unlock(&lun->lun_lock);
11382                         ctl_set_invalid_opcode(ctsio);
11383                         ctl_done((union ctl_io *)ctsio);
11384                         return;
11385                 }
11386         }
11387
11388         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11389
11390         /*
11391          * If we've got a request sense, it'll clear the contingent
11392          * allegiance condition.  Otherwise, if we have a CA condition for
11393          * this initiator, clear it, because it sent down a command other
11394          * than request sense.
11395          */
11396         if (ctsio->cdb[0] != REQUEST_SENSE) {
11397                 struct scsi_sense_data *ps;
11398
11399                 ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
11400                 if (ps != NULL)
11401                         ps[initidx % CTL_MAX_INIT_PER_PORT].error_code = 0;
11402         }
11403
11404         /*
11405          * If the command has this flag set, it handles its own unit
11406          * attention reporting, we shouldn't do anything.  Otherwise we
11407          * check for any pending unit attentions, and send them back to the
11408          * initiator.  We only do this when a command initially comes in,
11409          * not when we pull it off the blocked queue.
11410          *
11411          * According to SAM-3, section 5.3.2, the order that things get
11412          * presented back to the host is basically unit attentions caused
11413          * by some sort of reset event, busy status, reservation conflicts
11414          * or task set full, and finally any other status.
11415          *
11416          * One issue here is that some of the unit attentions we report
11417          * don't fall into the "reset" category (e.g. "reported luns data
11418          * has changed").  So reporting it here, before the reservation
11419          * check, may be technically wrong.  I guess the only thing to do
11420          * would be to check for and report the reset events here, and then
11421          * check for the other unit attention types after we check for a
11422          * reservation conflict.
11423          *
11424          * XXX KDM need to fix this
11425          */
11426         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11427                 ctl_ua_type ua_type;
11428                 u_int sense_len = 0;
11429
11430                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11431                     &sense_len, SSD_TYPE_NONE);
11432                 if (ua_type != CTL_UA_NONE) {
11433                         mtx_unlock(&lun->lun_lock);
11434                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11435                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11436                         ctsio->sense_len = sense_len;
11437                         ctl_done((union ctl_io *)ctsio);
11438                         return;
11439                 }
11440         }
11441
11442         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11443                 mtx_unlock(&lun->lun_lock);
11444                 ctl_done((union ctl_io *)ctsio);
11445                 return;
11446         }
11447
11448         /*
11449          * XXX CHD this is where we want to send IO to other side if
11450          * this LUN is secondary on this SC. We will need to make a copy
11451          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11452          * the copy we send as FROM_OTHER.
11453          * We also need to stuff the address of the original IO so we can
11454          * find it easily. Something similar will need be done on the other
11455          * side so when we are done we can find the copy.
11456          */
11457         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11458             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11459             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11460                 union ctl_ha_msg msg_info;
11461                 int isc_retval;
11462
11463                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11464                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11465                 mtx_unlock(&lun->lun_lock);
11466
11467                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11468                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11469                 msg_info.hdr.serializing_sc = NULL;
11470                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11471                 msg_info.scsi.tag_num = ctsio->tag_num;
11472                 msg_info.scsi.tag_type = ctsio->tag_type;
11473                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11474                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11475                 msg_info.scsi.priority = ctsio->priority;
11476
11477                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11478                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11479                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11480                         ctsio->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11481                         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
11482                         ctl_set_busy(ctsio);
11483                         ctl_done((union ctl_io *)ctsio);
11484                         return;
11485                 }
11486                 return;
11487         }
11488
11489         bio = (union ctl_io *)LIST_NEXT(&ctsio->io_hdr, ooa_links);
11490         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio, &bio)) {
11491         case CTL_ACTION_PASS:
11492         case CTL_ACTION_SKIP:
11493                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11494                 mtx_unlock(&lun->lun_lock);
11495                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11496                 break;
11497         case CTL_ACTION_BLOCK:
11498                 ctsio->io_hdr.blocker = bio;
11499                 TAILQ_INSERT_TAIL(&bio->io_hdr.blocked_queue, &ctsio->io_hdr,
11500                                   blocked_links);
11501                 mtx_unlock(&lun->lun_lock);
11502                 break;
11503         case CTL_ACTION_OVERLAP:
11504                 mtx_unlock(&lun->lun_lock);
11505                 ctl_set_overlapped_cmd(ctsio);
11506                 ctl_done((union ctl_io *)ctsio);
11507                 break;
11508         case CTL_ACTION_OVERLAP_TAG:
11509                 mtx_unlock(&lun->lun_lock);
11510                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11511                 ctl_done((union ctl_io *)ctsio);
11512                 break;
11513         default:
11514                 __assert_unreachable();
11515         }
11516 }
11517
11518 const struct ctl_cmd_entry *
11519 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11520 {
11521         const struct ctl_cmd_entry *entry;
11522         int service_action;
11523
11524         entry = &ctl_cmd_table[ctsio->cdb[0]];
11525         if (sa)
11526                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11527         if (entry->flags & CTL_CMD_FLAG_SA5) {
11528                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11529                 entry = &((const struct ctl_cmd_entry *)
11530                     entry->execute)[service_action];
11531         }
11532         return (entry);
11533 }
11534
11535 const struct ctl_cmd_entry *
11536 ctl_validate_command(struct ctl_scsiio *ctsio)
11537 {
11538         const struct ctl_cmd_entry *entry;
11539         int i, sa;
11540         uint8_t diff;
11541
11542         entry = ctl_get_cmd_entry(ctsio, &sa);
11543         ctsio->seridx = entry->seridx;
11544         if (entry->execute == NULL) {
11545                 if (sa)
11546                         ctl_set_invalid_field(ctsio,
11547                                               /*sks_valid*/ 1,
11548                                               /*command*/ 1,
11549                                               /*field*/ 1,
11550                                               /*bit_valid*/ 1,
11551                                               /*bit*/ 4);
11552                 else
11553                         ctl_set_invalid_opcode(ctsio);
11554                 ctl_done((union ctl_io *)ctsio);
11555                 return (NULL);
11556         }
11557         KASSERT(entry->length > 0,
11558             ("Not defined length for command 0x%02x/0x%02x",
11559              ctsio->cdb[0], ctsio->cdb[1]));
11560         for (i = 1; i < entry->length; i++) {
11561                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11562                 if (diff == 0)
11563                         continue;
11564                 ctl_set_invalid_field(ctsio,
11565                                       /*sks_valid*/ 1,
11566                                       /*command*/ 1,
11567                                       /*field*/ i,
11568                                       /*bit_valid*/ 1,
11569                                       /*bit*/ fls(diff) - 1);
11570                 ctl_done((union ctl_io *)ctsio);
11571                 return (NULL);
11572         }
11573         return (entry);
11574 }
11575
11576 static int
11577 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11578 {
11579
11580         switch (lun_type) {
11581         case T_DIRECT:
11582                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11583                         return (0);
11584                 break;
11585         case T_PROCESSOR:
11586                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11587                         return (0);
11588                 break;
11589         case T_CDROM:
11590                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11591                         return (0);
11592                 break;
11593         default:
11594                 return (0);
11595         }
11596         return (1);
11597 }
11598
11599 static int
11600 ctl_scsiio(struct ctl_scsiio *ctsio)
11601 {
11602         int retval;
11603         const struct ctl_cmd_entry *entry;
11604
11605         retval = CTL_RETVAL_COMPLETE;
11606
11607         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11608
11609         entry = ctl_get_cmd_entry(ctsio, NULL);
11610
11611         /*
11612          * If this I/O has been aborted, just send it straight to
11613          * ctl_done() without executing it.
11614          */
11615         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11616                 ctl_done((union ctl_io *)ctsio);
11617                 goto bailout;
11618         }
11619
11620         /*
11621          * All the checks should have been handled by ctl_scsiio_precheck().
11622          * We should be clear now to just execute the I/O.
11623          */
11624         retval = entry->execute(ctsio);
11625
11626 bailout:
11627         return (retval);
11628 }
11629
11630 static int
11631 ctl_target_reset(union ctl_io *io)
11632 {
11633         struct ctl_softc *softc = CTL_SOFTC(io);
11634         struct ctl_port *port = CTL_PORT(io);
11635         struct ctl_lun *lun;
11636         uint32_t initidx;
11637         ctl_ua_type ua_type;
11638
11639         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11640                 union ctl_ha_msg msg_info;
11641
11642                 msg_info.hdr.nexus = io->io_hdr.nexus;
11643                 msg_info.task.task_action = io->taskio.task_action;
11644                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11645                 msg_info.hdr.original_sc = NULL;
11646                 msg_info.hdr.serializing_sc = NULL;
11647                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11648                     sizeof(msg_info.task), M_WAITOK);
11649         }
11650
11651         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11652         if (io->taskio.task_action == CTL_TASK_TARGET_RESET)
11653                 ua_type = CTL_UA_TARG_RESET;
11654         else
11655                 ua_type = CTL_UA_BUS_RESET;
11656         mtx_lock(&softc->ctl_lock);
11657         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11658                 if (port != NULL &&
11659                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
11660                         continue;
11661                 ctl_do_lun_reset(lun, initidx, ua_type);
11662         }
11663         mtx_unlock(&softc->ctl_lock);
11664         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11665         return (0);
11666 }
11667
11668 /*
11669  * The LUN should always be set.  The I/O is optional, and is used to
11670  * distinguish between I/Os sent by this initiator, and by other
11671  * initiators.  We set unit attention for initiators other than this one.
11672  * SAM-3 is vague on this point.  It does say that a unit attention should
11673  * be established for other initiators when a LUN is reset (see section
11674  * 5.7.3), but it doesn't specifically say that the unit attention should
11675  * be established for this particular initiator when a LUN is reset.  Here
11676  * is the relevant text, from SAM-3 rev 8:
11677  *
11678  * 5.7.2 When a SCSI initiator port aborts its own tasks
11679  *
11680  * When a SCSI initiator port causes its own task(s) to be aborted, no
11681  * notification that the task(s) have been aborted shall be returned to
11682  * the SCSI initiator port other than the completion response for the
11683  * command or task management function action that caused the task(s) to
11684  * be aborted and notification(s) associated with related effects of the
11685  * action (e.g., a reset unit attention condition).
11686  *
11687  * XXX KDM for now, we're setting unit attention for all initiators.
11688  */
11689 static void
11690 ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua_type)
11691 {
11692         struct ctl_io_hdr *xioh;
11693         int i;
11694
11695         mtx_lock(&lun->lun_lock);
11696         /* Abort tasks. */
11697         LIST_FOREACH(xioh, &lun->ooa_queue, ooa_links) {
11698                 xioh->flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11699                 ctl_try_unblock_io(lun, (union ctl_io *)xioh, FALSE);
11700         }
11701         /* Clear CA. */
11702         for (i = 0; i < ctl_max_ports; i++) {
11703                 free(lun->pending_sense[i], M_CTL);
11704                 lun->pending_sense[i] = NULL;
11705         }
11706         /* Clear reservation. */
11707         lun->flags &= ~CTL_LUN_RESERVED;
11708         /* Clear prevent media removal. */
11709         if (lun->prevent) {
11710                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
11711                         ctl_clear_mask(lun->prevent, i);
11712                 lun->prevent_count = 0;
11713         }
11714         /* Clear TPC status */
11715         ctl_tpc_lun_clear(lun, -1);
11716         /* Establish UA. */
11717 #if 0
11718         ctl_est_ua_all(lun, initidx, ua_type);
11719 #else
11720         ctl_est_ua_all(lun, -1, ua_type);
11721 #endif
11722         mtx_unlock(&lun->lun_lock);
11723 }
11724
11725 static int
11726 ctl_lun_reset(union ctl_io *io)
11727 {
11728         struct ctl_softc *softc = CTL_SOFTC(io);
11729         struct ctl_lun *lun;
11730         uint32_t targ_lun, initidx;
11731
11732         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11733         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11734         mtx_lock(&softc->ctl_lock);
11735         if (targ_lun >= ctl_max_luns ||
11736             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11737                 mtx_unlock(&softc->ctl_lock);
11738                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11739                 return (1);
11740         }
11741         ctl_do_lun_reset(lun, initidx, CTL_UA_LUN_RESET);
11742         mtx_unlock(&softc->ctl_lock);
11743         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11744
11745         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11746                 union ctl_ha_msg msg_info;
11747
11748                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11749                 msg_info.hdr.nexus = io->io_hdr.nexus;
11750                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11751                 msg_info.hdr.original_sc = NULL;
11752                 msg_info.hdr.serializing_sc = NULL;
11753                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11754                     sizeof(msg_info.task), M_WAITOK);
11755         }
11756         return (0);
11757 }
11758
11759 static void
11760 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11761     int other_sc)
11762 {
11763         struct ctl_io_hdr *xioh;
11764
11765         mtx_assert(&lun->lun_lock, MA_OWNED);
11766
11767         /*
11768          * Run through the OOA queue and attempt to find the given I/O.
11769          * The target port, initiator ID, tag type and tag number have to
11770          * match the values that we got from the initiator.  If we have an
11771          * untagged command to abort, simply abort the first untagged command
11772          * we come to.  We only allow one untagged command at a time of course.
11773          */
11774         LIST_FOREACH(xioh, &lun->ooa_queue, ooa_links) {
11775                 union ctl_io *xio = (union ctl_io *)xioh;
11776                 if ((targ_port == UINT32_MAX ||
11777                      targ_port == xioh->nexus.targ_port) &&
11778                     (init_id == UINT32_MAX ||
11779                      init_id == xioh->nexus.initid)) {
11780                         if (targ_port != xioh->nexus.targ_port ||
11781                             init_id != xioh->nexus.initid)
11782                                 xioh->flags |= CTL_FLAG_ABORT_STATUS;
11783                         xioh->flags |= CTL_FLAG_ABORT;
11784                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11785                                 union ctl_ha_msg msg_info;
11786
11787                                 msg_info.hdr.nexus = xioh->nexus;
11788                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11789                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11790                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11791                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11792                                 msg_info.hdr.original_sc = NULL;
11793                                 msg_info.hdr.serializing_sc = NULL;
11794                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11795                                     sizeof(msg_info.task), M_NOWAIT);
11796                         }
11797                         ctl_try_unblock_io(lun, xio, FALSE);
11798                 }
11799         }
11800 }
11801
11802 static int
11803 ctl_abort_task_set(union ctl_io *io)
11804 {
11805         struct ctl_softc *softc = CTL_SOFTC(io);
11806         struct ctl_lun *lun;
11807         uint32_t targ_lun;
11808
11809         /*
11810          * Look up the LUN.
11811          */
11812         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11813         mtx_lock(&softc->ctl_lock);
11814         if (targ_lun >= ctl_max_luns ||
11815             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11816                 mtx_unlock(&softc->ctl_lock);
11817                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11818                 return (1);
11819         }
11820
11821         mtx_lock(&lun->lun_lock);
11822         mtx_unlock(&softc->ctl_lock);
11823         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11824                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11825                     io->io_hdr.nexus.initid,
11826                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11827         } else { /* CTL_TASK_CLEAR_TASK_SET */
11828                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11829                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11830         }
11831         mtx_unlock(&lun->lun_lock);
11832         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11833         return (0);
11834 }
11835
11836 static void
11837 ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
11838     ctl_ua_type ua_type)
11839 {
11840         struct ctl_lun *lun;
11841         struct scsi_sense_data *ps;
11842         uint32_t p, i;
11843
11844         p = initidx / CTL_MAX_INIT_PER_PORT;
11845         i = initidx % CTL_MAX_INIT_PER_PORT;
11846         mtx_lock(&softc->ctl_lock);
11847         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11848                 mtx_lock(&lun->lun_lock);
11849                 /* Abort tasks. */
11850                 ctl_abort_tasks_lun(lun, p, i, 1);
11851                 /* Clear CA. */
11852                 ps = lun->pending_sense[p];
11853                 if (ps != NULL)
11854                         ps[i].error_code = 0;
11855                 /* Clear reservation. */
11856                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11857                         lun->flags &= ~CTL_LUN_RESERVED;
11858                 /* Clear prevent media removal. */
11859                 if (lun->prevent && ctl_is_set(lun->prevent, initidx)) {
11860                         ctl_clear_mask(lun->prevent, initidx);
11861                         lun->prevent_count--;
11862                 }
11863                 /* Clear TPC status */
11864                 ctl_tpc_lun_clear(lun, initidx);
11865                 /* Establish UA. */
11866                 ctl_est_ua(lun, initidx, ua_type);
11867                 mtx_unlock(&lun->lun_lock);
11868         }
11869         mtx_unlock(&softc->ctl_lock);
11870 }
11871
11872 static int
11873 ctl_i_t_nexus_reset(union ctl_io *io)
11874 {
11875         struct ctl_softc *softc = CTL_SOFTC(io);
11876         uint32_t initidx;
11877
11878         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11879                 union ctl_ha_msg msg_info;
11880
11881                 msg_info.hdr.nexus = io->io_hdr.nexus;
11882                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
11883                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11884                 msg_info.hdr.original_sc = NULL;
11885                 msg_info.hdr.serializing_sc = NULL;
11886                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11887                     sizeof(msg_info.task), M_WAITOK);
11888         }
11889
11890         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11891         ctl_i_t_nexus_loss(softc, initidx, CTL_UA_I_T_NEXUS_LOSS);
11892         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11893         return (0);
11894 }
11895
11896 static int
11897 ctl_abort_task(union ctl_io *io)
11898 {
11899         struct ctl_softc *softc = CTL_SOFTC(io);
11900         struct ctl_io_hdr *xioh;
11901         struct ctl_lun *lun;
11902         uint32_t targ_lun;
11903
11904         /*
11905          * Look up the LUN.
11906          */
11907         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11908         mtx_lock(&softc->ctl_lock);
11909         if (targ_lun >= ctl_max_luns ||
11910             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11911                 mtx_unlock(&softc->ctl_lock);
11912                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11913                 return (1);
11914         }
11915
11916         mtx_lock(&lun->lun_lock);
11917         mtx_unlock(&softc->ctl_lock);
11918         /*
11919          * Run through the OOA queue and attempt to find the given I/O.
11920          * The target port, initiator ID, tag type and tag number have to
11921          * match the values that we got from the initiator.  If we have an
11922          * untagged command to abort, simply abort the first untagged command
11923          * we come to.  We only allow one untagged command at a time of course.
11924          */
11925         LIST_FOREACH(xioh, &lun->ooa_queue, ooa_links) {
11926                 union ctl_io *xio = (union ctl_io *)xioh;
11927                 if ((xioh->nexus.targ_port != io->io_hdr.nexus.targ_port)
11928                  || (xioh->nexus.initid != io->io_hdr.nexus.initid)
11929                  || (xioh->flags & CTL_FLAG_ABORT))
11930                         continue;
11931
11932                 /*
11933                  * If the abort says that the task is untagged, the
11934                  * task in the queue must be untagged.  Otherwise,
11935                  * we just check to see whether the tag numbers
11936                  * match.  This is because the QLogic firmware
11937                  * doesn't pass back the tag type in an abort
11938                  * request.
11939                  */
11940 #if 0
11941                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11942                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11943                  || (xio->scsiio.tag_num == io->taskio.tag_num)) {
11944 #else
11945                 /*
11946                  * XXX KDM we've got problems with FC, because it
11947                  * doesn't send down a tag type with aborts.  So we
11948                  * can only really go by the tag number...
11949                  * This may cause problems with parallel SCSI.
11950                  * Need to figure that out!!
11951                  */
11952                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
11953 #endif
11954                         xioh->flags |= CTL_FLAG_ABORT;
11955                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
11956                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11957                                 union ctl_ha_msg msg_info;
11958
11959                                 msg_info.hdr.nexus = io->io_hdr.nexus;
11960                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11961                                 msg_info.task.tag_num = io->taskio.tag_num;
11962                                 msg_info.task.tag_type = io->taskio.tag_type;
11963                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11964                                 msg_info.hdr.original_sc = NULL;
11965                                 msg_info.hdr.serializing_sc = NULL;
11966                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11967                                     sizeof(msg_info.task), M_NOWAIT);
11968                         }
11969                         ctl_try_unblock_io(lun, xio, FALSE);
11970                 }
11971         }
11972         mtx_unlock(&lun->lun_lock);
11973         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11974         return (0);
11975 }
11976
11977 static int
11978 ctl_query_task(union ctl_io *io, int task_set)
11979 {
11980         struct ctl_softc *softc = CTL_SOFTC(io);
11981         struct ctl_io_hdr *xioh;
11982         struct ctl_lun *lun;
11983         int found = 0;
11984         uint32_t targ_lun;
11985
11986         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11987         mtx_lock(&softc->ctl_lock);
11988         if (targ_lun >= ctl_max_luns ||
11989             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11990                 mtx_unlock(&softc->ctl_lock);
11991                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11992                 return (1);
11993         }
11994         mtx_lock(&lun->lun_lock);
11995         mtx_unlock(&softc->ctl_lock);
11996         LIST_FOREACH(xioh, &lun->ooa_queue, ooa_links) {
11997                 union ctl_io *xio = (union ctl_io *)xioh;
11998                 if ((xioh->nexus.targ_port != io->io_hdr.nexus.targ_port)
11999                  || (xioh->nexus.initid != io->io_hdr.nexus.initid)
12000                  || (xioh->flags & CTL_FLAG_ABORT))
12001                         continue;
12002
12003                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12004                         found = 1;
12005                         break;
12006                 }
12007         }
12008         mtx_unlock(&lun->lun_lock);
12009         if (found)
12010                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12011         else
12012                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12013         return (0);
12014 }
12015
12016 static int
12017 ctl_query_async_event(union ctl_io *io)
12018 {
12019         struct ctl_softc *softc = CTL_SOFTC(io);
12020         struct ctl_lun *lun;
12021         ctl_ua_type ua;
12022         uint32_t targ_lun, initidx;
12023
12024         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12025         mtx_lock(&softc->ctl_lock);
12026         if (targ_lun >= ctl_max_luns ||
12027             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12028                 mtx_unlock(&softc->ctl_lock);
12029                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12030                 return (1);
12031         }
12032         mtx_lock(&lun->lun_lock);
12033         mtx_unlock(&softc->ctl_lock);
12034         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12035         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12036         mtx_unlock(&lun->lun_lock);
12037         if (ua != CTL_UA_NONE)
12038                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12039         else
12040                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12041         return (0);
12042 }
12043
12044 static void
12045 ctl_run_task(union ctl_io *io)
12046 {
12047         int retval = 1;
12048
12049         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12050         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12051             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12052         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12053         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12054         switch (io->taskio.task_action) {
12055         case CTL_TASK_ABORT_TASK:
12056                 retval = ctl_abort_task(io);
12057                 break;
12058         case CTL_TASK_ABORT_TASK_SET:
12059         case CTL_TASK_CLEAR_TASK_SET:
12060                 retval = ctl_abort_task_set(io);
12061                 break;
12062         case CTL_TASK_CLEAR_ACA:
12063                 break;
12064         case CTL_TASK_I_T_NEXUS_RESET:
12065                 retval = ctl_i_t_nexus_reset(io);
12066                 break;
12067         case CTL_TASK_LUN_RESET:
12068                 retval = ctl_lun_reset(io);
12069                 break;
12070         case CTL_TASK_TARGET_RESET:
12071         case CTL_TASK_BUS_RESET:
12072                 retval = ctl_target_reset(io);
12073                 break;
12074         case CTL_TASK_PORT_LOGIN:
12075                 break;
12076         case CTL_TASK_PORT_LOGOUT:
12077                 break;
12078         case CTL_TASK_QUERY_TASK:
12079                 retval = ctl_query_task(io, 0);
12080                 break;
12081         case CTL_TASK_QUERY_TASK_SET:
12082                 retval = ctl_query_task(io, 1);
12083                 break;
12084         case CTL_TASK_QUERY_ASYNC_EVENT:
12085                 retval = ctl_query_async_event(io);
12086                 break;
12087         default:
12088                 printf("%s: got unknown task management event %d\n",
12089                        __func__, io->taskio.task_action);
12090                 break;
12091         }
12092         if (retval == 0)
12093                 io->io_hdr.status = CTL_SUCCESS;
12094         else
12095                 io->io_hdr.status = CTL_ERROR;
12096         ctl_done(io);
12097 }
12098
12099 /*
12100  * For HA operation.  Handle commands that come in from the other
12101  * controller.
12102  */
12103 static void
12104 ctl_handle_isc(union ctl_io *io)
12105 {
12106         struct ctl_softc *softc = CTL_SOFTC(io);
12107         struct ctl_lun *lun;
12108         const struct ctl_cmd_entry *entry;
12109         uint32_t targ_lun;
12110
12111         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12112         switch (io->io_hdr.msg_type) {
12113         case CTL_MSG_SERIALIZE:
12114                 ctl_serialize_other_sc_cmd(&io->scsiio);
12115                 break;
12116         case CTL_MSG_R2R:               /* Only used in SER_ONLY mode. */
12117                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12118                 if (targ_lun >= ctl_max_luns ||
12119                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12120                         ctl_done(io);
12121                         break;
12122                 }
12123                 mtx_lock(&lun->lun_lock);
12124                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
12125                         mtx_unlock(&lun->lun_lock);
12126                         ctl_done(io);
12127                         break;
12128                 }
12129                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12130                 mtx_unlock(&lun->lun_lock);
12131                 ctl_enqueue_rtr(io);
12132                 break;
12133         case CTL_MSG_FINISH_IO:
12134                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12135                         ctl_done(io);
12136                         break;
12137                 }
12138                 if (targ_lun >= ctl_max_luns ||
12139                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12140                         ctl_free_io(io);
12141                         break;
12142                 }
12143                 mtx_lock(&lun->lun_lock);
12144                 ctl_try_unblock_others(lun, io, TRUE);
12145                 LIST_REMOVE(&io->io_hdr, ooa_links);
12146                 mtx_unlock(&lun->lun_lock);
12147                 ctl_free_io(io);
12148                 break;
12149         case CTL_MSG_PERS_ACTION:
12150                 ctl_hndl_per_res_out_on_other_sc(io);
12151                 ctl_free_io(io);
12152                 break;
12153         case CTL_MSG_BAD_JUJU:
12154                 ctl_done(io);
12155                 break;
12156         case CTL_MSG_DATAMOVE:          /* Only used in XFER mode */
12157                 ctl_datamove_remote(io);
12158                 break;
12159         case CTL_MSG_DATAMOVE_DONE:     /* Only used in XFER mode */
12160                 ctl_datamove_done(io, false);
12161                 break;
12162         case CTL_MSG_FAILOVER:
12163                 ctl_failover_lun(io);
12164                 ctl_free_io(io);
12165                 break;
12166         default:
12167                 printf("%s: Invalid message type %d\n",
12168                        __func__, io->io_hdr.msg_type);
12169                 ctl_free_io(io);
12170                 break;
12171         }
12172
12173 }
12174
12175 /*
12176  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12177  * there is no match.
12178  */
12179 static ctl_lun_error_pattern
12180 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12181 {
12182         const struct ctl_cmd_entry *entry;
12183         ctl_lun_error_pattern filtered_pattern, pattern;
12184
12185         pattern = desc->error_pattern;
12186
12187         /*
12188          * XXX KDM we need more data passed into this function to match a
12189          * custom pattern, and we actually need to implement custom pattern
12190          * matching.
12191          */
12192         if (pattern & CTL_LUN_PAT_CMD)
12193                 return (CTL_LUN_PAT_CMD);
12194
12195         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12196                 return (CTL_LUN_PAT_ANY);
12197
12198         entry = ctl_get_cmd_entry(ctsio, NULL);
12199
12200         filtered_pattern = entry->pattern & pattern;
12201
12202         /*
12203          * If the user requested specific flags in the pattern (e.g.
12204          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12205          * flags.
12206          *
12207          * If the user did not specify any flags, it doesn't matter whether
12208          * or not the command supports the flags.
12209          */
12210         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12211              (pattern & ~CTL_LUN_PAT_MASK))
12212                 return (CTL_LUN_PAT_NONE);
12213
12214         /*
12215          * If the user asked for a range check, see if the requested LBA
12216          * range overlaps with this command's LBA range.
12217          */
12218         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12219                 uint64_t lba1;
12220                 uint64_t len1;
12221                 ctl_action action;
12222                 int retval;
12223
12224                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12225                 if (retval != 0)
12226                         return (CTL_LUN_PAT_NONE);
12227
12228                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12229                                               desc->lba_range.len, FALSE);
12230                 /*
12231                  * A "pass" means that the LBA ranges don't overlap, so
12232                  * this doesn't match the user's range criteria.
12233                  */
12234                 if (action == CTL_ACTION_PASS)
12235                         return (CTL_LUN_PAT_NONE);
12236         }
12237
12238         return (filtered_pattern);
12239 }
12240
12241 static void
12242 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12243 {
12244         struct ctl_error_desc *desc, *desc2;
12245
12246         mtx_assert(&lun->lun_lock, MA_OWNED);
12247
12248         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12249                 ctl_lun_error_pattern pattern;
12250                 /*
12251                  * Check to see whether this particular command matches
12252                  * the pattern in the descriptor.
12253                  */
12254                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12255                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12256                         continue;
12257
12258                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12259                 case CTL_LUN_INJ_ABORTED:
12260                         ctl_set_aborted(&io->scsiio);
12261                         break;
12262                 case CTL_LUN_INJ_MEDIUM_ERR:
12263                         ctl_set_medium_error(&io->scsiio,
12264                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12265                              CTL_FLAG_DATA_OUT);
12266                         break;
12267                 case CTL_LUN_INJ_UA:
12268                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12269                          * OCCURRED */
12270                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12271                         break;
12272                 case CTL_LUN_INJ_CUSTOM:
12273                         /*
12274                          * We're assuming the user knows what he is doing.
12275                          * Just copy the sense information without doing
12276                          * checks.
12277                          */
12278                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12279                               MIN(sizeof(desc->custom_sense),
12280                                   sizeof(io->scsiio.sense_data)));
12281                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12282                         io->scsiio.sense_len = SSD_FULL_SIZE;
12283                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12284                         break;
12285                 case CTL_LUN_INJ_NONE:
12286                 default:
12287                         /*
12288                          * If this is an error injection type we don't know
12289                          * about, clear the continuous flag (if it is set)
12290                          * so it will get deleted below.
12291                          */
12292                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12293                         break;
12294                 }
12295                 /*
12296                  * By default, each error injection action is a one-shot
12297                  */
12298                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12299                         continue;
12300
12301                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12302
12303                 free(desc, M_CTL);
12304         }
12305 }
12306
12307 #ifdef CTL_IO_DELAY
12308 static void
12309 ctl_datamove_timer_wakeup(void *arg)
12310 {
12311         union ctl_io *io;
12312
12313         io = (union ctl_io *)arg;
12314
12315         ctl_datamove(io);
12316 }
12317 #endif /* CTL_IO_DELAY */
12318
12319 static void
12320 ctl_datamove_done_process(union ctl_io *io)
12321 {
12322 #ifdef CTL_TIME_IO
12323         struct bintime cur_bt;
12324 #endif
12325
12326         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
12327             ("%s: unexpected I/O type %x", __func__, io->io_hdr.io_type));
12328
12329 #ifdef CTL_TIME_IO
12330         getbinuptime(&cur_bt);
12331         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12332         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12333 #endif
12334         io->io_hdr.num_dmas++;
12335
12336         if ((io->io_hdr.port_status != 0) &&
12337             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
12338              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
12339                 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
12340                     /*retry_count*/ io->io_hdr.port_status);
12341         } else if (io->scsiio.kern_data_resid != 0 &&
12342             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
12343             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
12344              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
12345                 ctl_set_invalid_field_ciu(&io->scsiio);
12346         } else if (ctl_debug & CTL_DEBUG_CDB_DATA)
12347                 ctl_data_print(io);
12348 }
12349
12350 void
12351 ctl_datamove_done(union ctl_io *io, bool samethr)
12352 {
12353
12354         ctl_datamove_done_process(io);
12355         io->scsiio.be_move_done(io, samethr);
12356 }
12357
12358 void
12359 ctl_datamove(union ctl_io *io)
12360 {
12361         void (*fe_datamove)(union ctl_io *io);
12362
12363         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12364
12365         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12366
12367         /* No data transferred yet.  Frontend must update this when done. */
12368         io->scsiio.kern_data_resid = io->scsiio.kern_data_len;
12369
12370 #ifdef CTL_TIME_IO
12371         getbinuptime(&io->io_hdr.dma_start_bt);
12372 #endif /* CTL_TIME_IO */
12373
12374 #ifdef CTL_IO_DELAY
12375         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12376                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12377         } else {
12378                 struct ctl_lun *lun;
12379
12380                 lun = CTL_LUN(io);
12381                 if ((lun != NULL)
12382                  && (lun->delay_info.datamove_delay > 0)) {
12383                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12384                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12385                         callout_reset(&io->io_hdr.delay_callout,
12386                                       lun->delay_info.datamove_delay * hz,
12387                                       ctl_datamove_timer_wakeup, io);
12388                         if (lun->delay_info.datamove_type ==
12389                             CTL_DELAY_TYPE_ONESHOT)
12390                                 lun->delay_info.datamove_delay = 0;
12391                         return;
12392                 }
12393         }
12394 #endif
12395
12396         /*
12397          * This command has been aborted.  Set the port status, so we fail
12398          * the data move.
12399          */
12400         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12401                 printf("ctl_datamove: tag 0x%jx on (%u:%u:%u) aborted\n",
12402                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12403                        io->io_hdr.nexus.targ_port,
12404                        io->io_hdr.nexus.targ_lun);
12405                 io->io_hdr.port_status = 31337;
12406                 ctl_datamove_done_process(io);
12407                 io->scsiio.be_move_done(io, true);
12408                 return;
12409         }
12410
12411         /* Don't confuse frontend with zero length data move. */
12412         if (io->scsiio.kern_data_len == 0) {
12413                 ctl_datamove_done_process(io);
12414                 io->scsiio.be_move_done(io, true);
12415                 return;
12416         }
12417
12418         fe_datamove = CTL_PORT(io)->fe_datamove;
12419         fe_datamove(io);
12420 }
12421
12422 static void
12423 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12424 {
12425         union ctl_ha_msg msg;
12426 #ifdef CTL_TIME_IO
12427         struct bintime cur_bt;
12428 #endif
12429
12430         memset(&msg, 0, sizeof(msg));
12431         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12432         msg.hdr.original_sc = io;
12433         msg.hdr.serializing_sc = io->io_hdr.remote_io;
12434         msg.hdr.nexus = io->io_hdr.nexus;
12435         msg.hdr.status = io->io_hdr.status;
12436         msg.scsi.kern_data_resid = io->scsiio.kern_data_resid;
12437         msg.scsi.tag_num = io->scsiio.tag_num;
12438         msg.scsi.tag_type = io->scsiio.tag_type;
12439         msg.scsi.scsi_status = io->scsiio.scsi_status;
12440         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12441                io->scsiio.sense_len);
12442         msg.scsi.sense_len = io->scsiio.sense_len;
12443         msg.scsi.port_status = io->io_hdr.port_status;
12444         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12445         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12446                 ctl_failover_io(io, /*have_lock*/ have_lock);
12447                 return;
12448         }
12449         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12450             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12451             msg.scsi.sense_len, M_WAITOK);
12452
12453 #ifdef CTL_TIME_IO
12454         getbinuptime(&cur_bt);
12455         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12456         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12457 #endif
12458         io->io_hdr.num_dmas++;
12459 }
12460
12461 /*
12462  * The DMA to the remote side is done, now we need to tell the other side
12463  * we're done so it can continue with its data movement.
12464  */
12465 static void
12466 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12467 {
12468         union ctl_io *io;
12469         uint32_t i;
12470
12471         io = rq->context;
12472
12473         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12474                 printf("%s: ISC DMA write failed with error %d", __func__,
12475                        rq->ret);
12476                 ctl_set_internal_failure(&io->scsiio,
12477                                          /*sks_valid*/ 1,
12478                                          /*retry_count*/ rq->ret);
12479         }
12480
12481         ctl_dt_req_free(rq);
12482
12483         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12484                 free(CTL_LSGLT(io)[i].addr, M_CTL);
12485         free(CTL_RSGL(io), M_CTL);
12486         CTL_RSGL(io) = NULL;
12487         CTL_LSGL(io) = NULL;
12488
12489         /*
12490          * The data is in local and remote memory, so now we need to send
12491          * status (good or back) back to the other side.
12492          */
12493         ctl_send_datamove_done(io, /*have_lock*/ 0);
12494 }
12495
12496 /*
12497  * We've moved the data from the host/controller into local memory.  Now we
12498  * need to push it over to the remote controller's memory.
12499  */
12500 static int
12501 ctl_datamove_remote_dm_write_cb(union ctl_io *io, bool samethr)
12502 {
12503         int retval;
12504
12505         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12506                                           ctl_datamove_remote_write_cb);
12507         return (retval);
12508 }
12509
12510 static void
12511 ctl_datamove_remote_write(union ctl_io *io)
12512 {
12513         int retval;
12514         void (*fe_datamove)(union ctl_io *io);
12515
12516         /*
12517          * - Get the data from the host/HBA into local memory.
12518          * - DMA memory from the local controller to the remote controller.
12519          * - Send status back to the remote controller.
12520          */
12521
12522         retval = ctl_datamove_remote_sgl_setup(io);
12523         if (retval != 0)
12524                 return;
12525
12526         /* Switch the pointer over so the FETD knows what to do */
12527         io->scsiio.kern_data_ptr = (uint8_t *)CTL_LSGL(io);
12528
12529         /*
12530          * Use a custom move done callback, since we need to send completion
12531          * back to the other controller, not to the backend on this side.
12532          */
12533         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12534
12535         fe_datamove = CTL_PORT(io)->fe_datamove;
12536         fe_datamove(io);
12537 }
12538
12539 static int
12540 ctl_datamove_remote_dm_read_cb(union ctl_io *io, bool samethr)
12541 {
12542         uint32_t i;
12543
12544         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12545                 free(CTL_LSGLT(io)[i].addr, M_CTL);
12546         free(CTL_RSGL(io), M_CTL);
12547         CTL_RSGL(io) = NULL;
12548         CTL_LSGL(io) = NULL;
12549
12550         /*
12551          * The read is done, now we need to send status (good or bad) back
12552          * to the other side.
12553          */
12554         ctl_send_datamove_done(io, /*have_lock*/ 0);
12555
12556         return (0);
12557 }
12558
12559 static void
12560 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12561 {
12562         union ctl_io *io;
12563         void (*fe_datamove)(union ctl_io *io);
12564
12565         io = rq->context;
12566
12567         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12568                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12569                        rq->ret);
12570                 ctl_set_internal_failure(&io->scsiio,
12571                                          /*sks_valid*/ 1,
12572                                          /*retry_count*/ rq->ret);
12573         }
12574
12575         ctl_dt_req_free(rq);
12576
12577         /* Switch the pointer over so the FETD knows what to do */
12578         io->scsiio.kern_data_ptr = (uint8_t *)CTL_LSGL(io);
12579
12580         /*
12581          * Use a custom move done callback, since we need to send completion
12582          * back to the other controller, not to the backend on this side.
12583          */
12584         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12585
12586         /* XXX KDM add checks like the ones in ctl_datamove? */
12587
12588         fe_datamove = CTL_PORT(io)->fe_datamove;
12589         fe_datamove(io);
12590 }
12591
12592 static int
12593 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12594 {
12595         struct ctl_sg_entry *local_sglist;
12596         uint32_t len_to_go;
12597         int retval;
12598         int i;
12599
12600         retval = 0;
12601         local_sglist = CTL_LSGL(io);
12602         len_to_go = io->scsiio.kern_data_len;
12603
12604         /*
12605          * The difficult thing here is that the size of the various
12606          * S/G segments may be different than the size from the
12607          * remote controller.  That'll make it harder when DMAing
12608          * the data back to the other side.
12609          */
12610         for (i = 0; len_to_go > 0; i++) {
12611                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12612                 local_sglist[i].addr =
12613                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12614
12615                 len_to_go -= local_sglist[i].len;
12616         }
12617         /*
12618          * Reset the number of S/G entries accordingly.  The original
12619          * number of S/G entries is available in rem_sg_entries.
12620          */
12621         io->scsiio.kern_sg_entries = i;
12622
12623         return (retval);
12624 }
12625
12626 static int
12627 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12628                          ctl_ha_dt_cb callback)
12629 {
12630         struct ctl_ha_dt_req *rq;
12631         struct ctl_sg_entry *remote_sglist, *local_sglist;
12632         uint32_t local_used, remote_used, total_used;
12633         int i, j, isc_ret;
12634
12635         rq = ctl_dt_req_alloc();
12636
12637         /*
12638          * If we failed to allocate the request, and if the DMA didn't fail
12639          * anyway, set busy status.  This is just a resource allocation
12640          * failure.
12641          */
12642         if ((rq == NULL)
12643          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12644              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12645                 ctl_set_busy(&io->scsiio);
12646
12647         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12648             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12649                 if (rq != NULL)
12650                         ctl_dt_req_free(rq);
12651
12652                 /*
12653                  * The data move failed.  We need to return status back
12654                  * to the other controller.  No point in trying to DMA
12655                  * data to the remote controller.
12656                  */
12657
12658                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12659
12660                 return (1);
12661         }
12662
12663         local_sglist = CTL_LSGL(io);
12664         remote_sglist = CTL_RSGL(io);
12665         local_used = 0;
12666         remote_used = 0;
12667         total_used = 0;
12668
12669         /*
12670          * Pull/push the data over the wire from/to the other controller.
12671          * This takes into account the possibility that the local and
12672          * remote sglists may not be identical in terms of the size of
12673          * the elements and the number of elements.
12674          *
12675          * One fundamental assumption here is that the length allocated for
12676          * both the local and remote sglists is identical.  Otherwise, we've
12677          * essentially got a coding error of some sort.
12678          */
12679         isc_ret = CTL_HA_STATUS_SUCCESS;
12680         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12681                 uint32_t cur_len;
12682                 uint8_t *tmp_ptr;
12683
12684                 rq->command = command;
12685                 rq->context = io;
12686
12687                 /*
12688                  * Both pointers should be aligned.  But it is possible
12689                  * that the allocation length is not.  They should both
12690                  * also have enough slack left over at the end, though,
12691                  * to round up to the next 8 byte boundary.
12692                  */
12693                 cur_len = MIN(local_sglist[i].len - local_used,
12694                               remote_sglist[j].len - remote_used);
12695                 rq->size = cur_len;
12696
12697                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12698                 tmp_ptr += local_used;
12699
12700 #if 0
12701                 /* Use physical addresses when talking to ISC hardware */
12702                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12703                         /* XXX KDM use busdma */
12704                         rq->local = vtophys(tmp_ptr);
12705                 } else
12706                         rq->local = tmp_ptr;
12707 #else
12708                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12709                     ("HA does not support BUS_ADDR"));
12710                 rq->local = tmp_ptr;
12711 #endif
12712
12713                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12714                 tmp_ptr += remote_used;
12715                 rq->remote = tmp_ptr;
12716
12717                 rq->callback = NULL;
12718
12719                 local_used += cur_len;
12720                 if (local_used >= local_sglist[i].len) {
12721                         i++;
12722                         local_used = 0;
12723                 }
12724
12725                 remote_used += cur_len;
12726                 if (remote_used >= remote_sglist[j].len) {
12727                         j++;
12728                         remote_used = 0;
12729                 }
12730                 total_used += cur_len;
12731
12732                 if (total_used >= io->scsiio.kern_data_len)
12733                         rq->callback = callback;
12734
12735                 isc_ret = ctl_dt_single(rq);
12736                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
12737                         break;
12738         }
12739         if (isc_ret != CTL_HA_STATUS_WAIT) {
12740                 rq->ret = isc_ret;
12741                 callback(rq);
12742         }
12743
12744         return (0);
12745 }
12746
12747 static void
12748 ctl_datamove_remote_read(union ctl_io *io)
12749 {
12750         int retval;
12751         uint32_t i;
12752
12753         /*
12754          * This will send an error to the other controller in the case of a
12755          * failure.
12756          */
12757         retval = ctl_datamove_remote_sgl_setup(io);
12758         if (retval != 0)
12759                 return;
12760
12761         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12762                                           ctl_datamove_remote_read_cb);
12763         if (retval != 0) {
12764                 /*
12765                  * Make sure we free memory if there was an error..  The
12766                  * ctl_datamove_remote_xfer() function will send the
12767                  * datamove done message, or call the callback with an
12768                  * error if there is a problem.
12769                  */
12770                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12771                         free(CTL_LSGLT(io)[i].addr, M_CTL);
12772                 free(CTL_RSGL(io), M_CTL);
12773                 CTL_RSGL(io) = NULL;
12774                 CTL_LSGL(io) = NULL;
12775         }
12776 }
12777
12778 /*
12779  * Process a datamove request from the other controller.  This is used for
12780  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12781  * first.  Once that is complete, the data gets DMAed into the remote
12782  * controller's memory.  For reads, we DMA from the remote controller's
12783  * memory into our memory first, and then move it out to the FETD.
12784  */
12785 static void
12786 ctl_datamove_remote(union ctl_io *io)
12787 {
12788
12789         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12790
12791         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12792                 ctl_failover_io(io, /*have_lock*/ 0);
12793                 return;
12794         }
12795
12796         /*
12797          * Note that we look for an aborted I/O here, but don't do some of
12798          * the other checks that ctl_datamove() normally does.
12799          * We don't need to run the datamove delay code, since that should
12800          * have been done if need be on the other controller.
12801          */
12802         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12803                 printf("%s: tag 0x%jx on (%u:%u:%u) aborted\n", __func__,
12804                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12805                        io->io_hdr.nexus.targ_port,
12806                        io->io_hdr.nexus.targ_lun);
12807                 io->io_hdr.port_status = 31338;
12808                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12809                 return;
12810         }
12811
12812         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
12813                 ctl_datamove_remote_write(io);
12814         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
12815                 ctl_datamove_remote_read(io);
12816         else {
12817                 io->io_hdr.port_status = 31339;
12818                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12819         }
12820 }
12821
12822 static void
12823 ctl_process_done(union ctl_io *io)
12824 {
12825         struct ctl_softc *softc = CTL_SOFTC(io);
12826         struct ctl_port *port = CTL_PORT(io);
12827         struct ctl_lun *lun = CTL_LUN(io);
12828         void (*fe_done)(union ctl_io *io);
12829         union ctl_ha_msg msg;
12830
12831         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12832         fe_done = port->fe_done;
12833
12834 #ifdef CTL_TIME_IO
12835         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12836                 char str[256];
12837                 char path_str[64];
12838                 struct sbuf sb;
12839
12840                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12841                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12842
12843                 ctl_io_sbuf(io, &sb);
12844                 sbuf_cat(&sb, path_str);
12845                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12846                             (intmax_t)time_uptime - io->io_hdr.start_time);
12847                 sbuf_finish(&sb);
12848                 printf("%s", sbuf_data(&sb));
12849         }
12850 #endif /* CTL_TIME_IO */
12851
12852         switch (io->io_hdr.io_type) {
12853         case CTL_IO_SCSI:
12854                 break;
12855         case CTL_IO_TASK:
12856                 if (ctl_debug & CTL_DEBUG_INFO)
12857                         ctl_io_error_print(io, NULL);
12858                 fe_done(io);
12859                 return;
12860         default:
12861                 panic("%s: Invalid CTL I/O type %d\n",
12862                     __func__, io->io_hdr.io_type);
12863         }
12864
12865         if (lun == NULL) {
12866                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12867                                  io->io_hdr.nexus.targ_mapped_lun));
12868                 goto bailout;
12869         }
12870
12871         mtx_lock(&lun->lun_lock);
12872
12873         /*
12874          * Check to see if we have any informational exception and status
12875          * of this command can be modified to report it in form of either
12876          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
12877          */
12878         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
12879             io->io_hdr.status == CTL_SUCCESS &&
12880             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
12881                 uint8_t mrie = lun->MODE_IE.mrie;
12882                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
12883                     (lun->MODE_VER.byte3 & SMS_VER_PER));
12884                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
12885                      mrie == SIEP_MRIE_REC_UNCOND ||
12886                      mrie == SIEP_MRIE_NO_SENSE) &&
12887                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
12888                      CTL_CMD_FLAG_NO_SENSE) == 0) {
12889                         ctl_set_sense(&io->scsiio,
12890                               /*current_error*/ 1,
12891                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
12892                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
12893                               /*asc*/ lun->ie_asc,
12894                               /*ascq*/ lun->ie_ascq,
12895                               SSD_ELEM_NONE);
12896                         lun->ie_reported = 1;
12897                 }
12898         } else if (lun->ie_reported < 0)
12899                 lun->ie_reported = 0;
12900
12901         /*
12902          * Check to see if we have any errors to inject here.  We only
12903          * inject errors for commands that don't already have errors set.
12904          */
12905         if (!STAILQ_EMPTY(&lun->error_list) &&
12906             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
12907             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
12908                 ctl_inject_error(lun, io);
12909
12910         /*
12911          * XXX KDM how do we treat commands that aren't completed
12912          * successfully?
12913          *
12914          * XXX KDM should we also track I/O latency?
12915          */
12916         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
12917             io->io_hdr.io_type == CTL_IO_SCSI) {
12918                 int type;
12919 #ifdef CTL_TIME_IO
12920                 struct bintime bt;
12921
12922                 getbinuptime(&bt);
12923                 bintime_sub(&bt, &io->io_hdr.start_bt);
12924 #endif
12925                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12926                     CTL_FLAG_DATA_IN)
12927                         type = CTL_STATS_READ;
12928                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12929                     CTL_FLAG_DATA_OUT)
12930                         type = CTL_STATS_WRITE;
12931                 else
12932                         type = CTL_STATS_NO_IO;
12933
12934                 lun->stats.bytes[type] += io->scsiio.kern_total_len;
12935                 lun->stats.operations[type] ++;
12936                 lun->stats.dmas[type] += io->io_hdr.num_dmas;
12937 #ifdef CTL_TIME_IO
12938                 bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt);
12939                 bintime_add(&lun->stats.time[type], &bt);
12940 #endif
12941
12942                 mtx_lock(&port->port_lock);
12943                 port->stats.bytes[type] += io->scsiio.kern_total_len;
12944                 port->stats.operations[type] ++;
12945                 port->stats.dmas[type] += io->io_hdr.num_dmas;
12946 #ifdef CTL_TIME_IO
12947                 bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt);
12948                 bintime_add(&port->stats.time[type], &bt);
12949 #endif
12950                 mtx_unlock(&port->port_lock);
12951         }
12952
12953         /*
12954          * Run through the blocked queue of this I/O and see if anything
12955          * can be unblocked, now that this I/O is done and will be removed.
12956          * We need to do it before removal to have OOA position to start.
12957          */
12958         ctl_try_unblock_others(lun, io, TRUE);
12959
12960         /*
12961          * Remove this from the OOA queue.
12962          */
12963         LIST_REMOVE(&io->io_hdr, ooa_links);
12964 #ifdef CTL_TIME_IO
12965         if (LIST_EMPTY(&lun->ooa_queue))
12966                 lun->last_busy = getsbinuptime();
12967 #endif
12968
12969         /*
12970          * If the LUN has been invalidated, free it if there is nothing
12971          * left on its OOA queue.
12972          */
12973         if ((lun->flags & CTL_LUN_INVALID)
12974          && LIST_EMPTY(&lun->ooa_queue)) {
12975                 mtx_unlock(&lun->lun_lock);
12976                 ctl_free_lun(lun);
12977         } else
12978                 mtx_unlock(&lun->lun_lock);
12979
12980 bailout:
12981
12982         /*
12983          * If this command has been aborted, make sure we set the status
12984          * properly.  The FETD is responsible for freeing the I/O and doing
12985          * whatever it needs to do to clean up its state.
12986          */
12987         if (io->io_hdr.flags & CTL_FLAG_ABORT)
12988                 ctl_set_task_aborted(&io->scsiio);
12989
12990         /*
12991          * If enabled, print command error status.
12992          */
12993         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
12994             (ctl_debug & CTL_DEBUG_INFO) != 0)
12995                 ctl_io_error_print(io, NULL);
12996
12997         /*
12998          * Tell the FETD or the other shelf controller we're done with this
12999          * command.  Note that only SCSI commands get to this point.  Task
13000          * management commands are completed above.
13001          */
13002         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13003             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13004                 memset(&msg, 0, sizeof(msg));
13005                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13006                 msg.hdr.serializing_sc = io->io_hdr.remote_io;
13007                 msg.hdr.nexus = io->io_hdr.nexus;
13008                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13009                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13010                     M_WAITOK);
13011         }
13012
13013         fe_done(io);
13014 }
13015
13016 /*
13017  * Front end should call this if it doesn't do autosense.  When the request
13018  * sense comes back in from the initiator, we'll dequeue this and send it.
13019  */
13020 int
13021 ctl_queue_sense(union ctl_io *io)
13022 {
13023         struct ctl_softc *softc = CTL_SOFTC(io);
13024         struct ctl_port *port = CTL_PORT(io);
13025         struct ctl_lun *lun;
13026         struct scsi_sense_data *ps;
13027         uint32_t initidx, p, targ_lun;
13028
13029         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13030
13031         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13032
13033         /*
13034          * LUN lookup will likely move to the ctl_work_thread() once we
13035          * have our new queueing infrastructure (that doesn't put things on
13036          * a per-LUN queue initially).  That is so that we can handle
13037          * things like an INQUIRY to a LUN that we don't have enabled.  We
13038          * can't deal with that right now.
13039          * If we don't have a LUN for this, just toss the sense information.
13040          */
13041         mtx_lock(&softc->ctl_lock);
13042         if (targ_lun >= ctl_max_luns ||
13043             (lun = softc->ctl_luns[targ_lun]) == NULL) {
13044                 mtx_unlock(&softc->ctl_lock);
13045                 goto bailout;
13046         }
13047         mtx_lock(&lun->lun_lock);
13048         mtx_unlock(&softc->ctl_lock);
13049
13050         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13051         p = initidx / CTL_MAX_INIT_PER_PORT;
13052         if (lun->pending_sense[p] == NULL) {
13053                 lun->pending_sense[p] = malloc(sizeof(*ps) * CTL_MAX_INIT_PER_PORT,
13054                     M_CTL, M_NOWAIT | M_ZERO);
13055         }
13056         if ((ps = lun->pending_sense[p]) != NULL) {
13057                 ps += initidx % CTL_MAX_INIT_PER_PORT;
13058                 memset(ps, 0, sizeof(*ps));
13059                 memcpy(ps, &io->scsiio.sense_data, io->scsiio.sense_len);
13060         }
13061         mtx_unlock(&lun->lun_lock);
13062
13063 bailout:
13064         ctl_free_io(io);
13065         return (CTL_RETVAL_COMPLETE);
13066 }
13067
13068 /*
13069  * Primary command inlet from frontend ports.  All SCSI and task I/O
13070  * requests must go through this function.
13071  */
13072 int
13073 ctl_queue(union ctl_io *io)
13074 {
13075         struct ctl_port *port = CTL_PORT(io);
13076
13077         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13078
13079 #ifdef CTL_TIME_IO
13080         io->io_hdr.start_time = time_uptime;
13081         getbinuptime(&io->io_hdr.start_bt);
13082 #endif /* CTL_TIME_IO */
13083
13084         /* Map FE-specific LUN ID into global one. */
13085         io->io_hdr.nexus.targ_mapped_lun =
13086             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13087
13088         switch (io->io_hdr.io_type) {
13089         case CTL_IO_SCSI:
13090         case CTL_IO_TASK:
13091                 if (ctl_debug & CTL_DEBUG_CDB)
13092                         ctl_io_print(io);
13093                 ctl_enqueue_incoming(io);
13094                 break;
13095         default:
13096                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13097                 return (EINVAL);
13098         }
13099
13100         return (CTL_RETVAL_COMPLETE);
13101 }
13102
13103 int
13104 ctl_run(union ctl_io *io)
13105 {
13106         struct ctl_port *port = CTL_PORT(io);
13107
13108         CTL_DEBUG_PRINT(("ctl_run cdb[0]=%02X\n", io->scsiio.cdb[0]));
13109
13110 #ifdef CTL_TIME_IO
13111         io->io_hdr.start_time = time_uptime;
13112         getbinuptime(&io->io_hdr.start_bt);
13113 #endif /* CTL_TIME_IO */
13114
13115         /* Map FE-specific LUN ID into global one. */
13116         io->io_hdr.nexus.targ_mapped_lun =
13117             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13118
13119         switch (io->io_hdr.io_type) {
13120         case CTL_IO_SCSI:
13121                 if (ctl_debug & CTL_DEBUG_CDB)
13122                         ctl_io_print(io);
13123                 ctl_scsiio_precheck(&io->scsiio);
13124                 break;
13125         case CTL_IO_TASK:
13126                 if (ctl_debug & CTL_DEBUG_CDB)
13127                         ctl_io_print(io);
13128                 ctl_run_task(io);
13129                 break;
13130         default:
13131                 printf("ctl_run: unknown I/O type %d\n", io->io_hdr.io_type);
13132                 return (EINVAL);
13133         }
13134
13135         return (CTL_RETVAL_COMPLETE);
13136 }
13137
13138 #ifdef CTL_IO_DELAY
13139 static void
13140 ctl_done_timer_wakeup(void *arg)
13141 {
13142         union ctl_io *io;
13143
13144         io = (union ctl_io *)arg;
13145         ctl_done(io);
13146 }
13147 #endif /* CTL_IO_DELAY */
13148
13149 void
13150 ctl_serseq_done(union ctl_io *io)
13151 {
13152         struct ctl_lun *lun = CTL_LUN(io);
13153
13154         /* This is racy, but should not be a problem. */
13155         if (!TAILQ_EMPTY(&io->io_hdr.blocked_queue)) {
13156                 mtx_lock(&lun->lun_lock);
13157                 io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13158                 ctl_try_unblock_others(lun, io, FALSE);
13159                 mtx_unlock(&lun->lun_lock);
13160         } else
13161                 io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13162 }
13163
13164 void
13165 ctl_done(union ctl_io *io)
13166 {
13167
13168         /*
13169          * Enable this to catch duplicate completion issues.
13170          */
13171 #if 0
13172         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13173                 printf("%s: type %d msg %d cdb %x iptl: "
13174                        "%u:%u:%u tag 0x%04x "
13175                        "flag %#x status %x\n",
13176                         __func__,
13177                         io->io_hdr.io_type,
13178                         io->io_hdr.msg_type,
13179                         io->scsiio.cdb[0],
13180                         io->io_hdr.nexus.initid,
13181                         io->io_hdr.nexus.targ_port,
13182                         io->io_hdr.nexus.targ_lun,
13183                         (io->io_hdr.io_type ==
13184                         CTL_IO_TASK) ?
13185                         io->taskio.tag_num :
13186                         io->scsiio.tag_num,
13187                         io->io_hdr.flags,
13188                         io->io_hdr.status);
13189         } else
13190                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13191 #endif
13192
13193         /*
13194          * This is an internal copy of an I/O, and should not go through
13195          * the normal done processing logic.
13196          */
13197         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13198                 return;
13199
13200 #ifdef CTL_IO_DELAY
13201         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13202                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13203         } else {
13204                 struct ctl_lun *lun = CTL_LUN(io);
13205
13206                 if ((lun != NULL)
13207                  && (lun->delay_info.done_delay > 0)) {
13208                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13209                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13210                         callout_reset(&io->io_hdr.delay_callout,
13211                                       lun->delay_info.done_delay * hz,
13212                                       ctl_done_timer_wakeup, io);
13213                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13214                                 lun->delay_info.done_delay = 0;
13215                         return;
13216                 }
13217         }
13218 #endif /* CTL_IO_DELAY */
13219
13220         ctl_enqueue_done(io);
13221 }
13222
13223 static void
13224 ctl_work_thread(void *arg)
13225 {
13226         struct ctl_thread *thr = (struct ctl_thread *)arg;
13227         struct ctl_softc *softc = thr->ctl_softc;
13228         union ctl_io *io;
13229         int retval;
13230
13231         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13232         thread_lock(curthread);
13233         sched_prio(curthread, PUSER - 1);
13234         thread_unlock(curthread);
13235
13236         while (!softc->shutdown) {
13237                 /*
13238                  * We handle the queues in this order:
13239                  * - ISC
13240                  * - done queue (to free up resources, unblock other commands)
13241                  * - incoming queue
13242                  * - RtR queue
13243                  *
13244                  * If those queues are empty, we break out of the loop and
13245                  * go to sleep.
13246                  */
13247                 mtx_lock(&thr->queue_lock);
13248                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13249                 if (io != NULL) {
13250                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13251                         mtx_unlock(&thr->queue_lock);
13252                         ctl_handle_isc(io);
13253                         continue;
13254                 }
13255                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13256                 if (io != NULL) {
13257                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13258                         /* clear any blocked commands, call fe_done */
13259                         mtx_unlock(&thr->queue_lock);
13260                         ctl_process_done(io);
13261                         continue;
13262                 }
13263                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13264                 if (io != NULL) {
13265                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13266                         mtx_unlock(&thr->queue_lock);
13267                         if (io->io_hdr.io_type == CTL_IO_TASK)
13268                                 ctl_run_task(io);
13269                         else
13270                                 ctl_scsiio_precheck(&io->scsiio);
13271                         continue;
13272                 }
13273                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13274                 if (io != NULL) {
13275                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13276                         mtx_unlock(&thr->queue_lock);
13277                         retval = ctl_scsiio(&io->scsiio);
13278                         if (retval != CTL_RETVAL_COMPLETE)
13279                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13280                         continue;
13281                 }
13282
13283                 /* Sleep until we have something to do. */
13284                 mtx_sleep(thr, &thr->queue_lock, PDROP, "-", 0);
13285         }
13286         thr->thread = NULL;
13287         kthread_exit();
13288 }
13289
13290 static void
13291 ctl_thresh_thread(void *arg)
13292 {
13293         struct ctl_softc *softc = (struct ctl_softc *)arg;
13294         struct ctl_lun *lun;
13295         struct ctl_logical_block_provisioning_page *page;
13296         const char *attr;
13297         union ctl_ha_msg msg;
13298         uint64_t thres, val;
13299         int i, e, set;
13300
13301         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13302         thread_lock(curthread);
13303         sched_prio(curthread, PUSER - 1);
13304         thread_unlock(curthread);
13305
13306         while (!softc->shutdown) {
13307                 mtx_lock(&softc->ctl_lock);
13308                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13309                         if ((lun->flags & CTL_LUN_DISABLED) ||
13310                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13311                             lun->backend->lun_attr == NULL)
13312                                 continue;
13313                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13314                             softc->ha_mode == CTL_HA_MODE_XFER)
13315                                 continue;
13316                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13317                                 continue;
13318                         e = 0;
13319                         page = &lun->MODE_LBP;
13320                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13321                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13322                                         continue;
13323                                 thres = scsi_4btoul(page->descr[i].count);
13324                                 thres <<= CTL_LBP_EXPONENT;
13325                                 switch (page->descr[i].resource) {
13326                                 case 0x01:
13327                                         attr = "blocksavail";
13328                                         break;
13329                                 case 0x02:
13330                                         attr = "blocksused";
13331                                         break;
13332                                 case 0xf1:
13333                                         attr = "poolblocksavail";
13334                                         break;
13335                                 case 0xf2:
13336                                         attr = "poolblocksused";
13337                                         break;
13338                                 default:
13339                                         continue;
13340                                 }
13341                                 mtx_unlock(&softc->ctl_lock); // XXX
13342                                 val = lun->backend->lun_attr(lun->be_lun, attr);
13343                                 mtx_lock(&softc->ctl_lock);
13344                                 if (val == UINT64_MAX)
13345                                         continue;
13346                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13347                                     == SLBPPD_ARMING_INC)
13348                                         e = (val >= thres);
13349                                 else
13350                                         e = (val <= thres);
13351                                 if (e)
13352                                         break;
13353                         }
13354                         mtx_lock(&lun->lun_lock);
13355                         if (e) {
13356                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13357                                     (uint8_t *)page, lun->ua_tpt_info);
13358                                 if (lun->lasttpt == 0 ||
13359                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13360                                         lun->lasttpt = time_uptime;
13361                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13362                                         set = 1;
13363                                 } else
13364                                         set = 0;
13365                         } else {
13366                                 lun->lasttpt = 0;
13367                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13368                                 set = -1;
13369                         }
13370                         mtx_unlock(&lun->lun_lock);
13371                         if (set != 0 &&
13372                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13373                                 /* Send msg to other side. */
13374                                 bzero(&msg.ua, sizeof(msg.ua));
13375                                 msg.hdr.msg_type = CTL_MSG_UA;
13376                                 msg.hdr.nexus.initid = -1;
13377                                 msg.hdr.nexus.targ_port = -1;
13378                                 msg.hdr.nexus.targ_lun = lun->lun;
13379                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13380                                 msg.ua.ua_all = 1;
13381                                 msg.ua.ua_set = (set > 0);
13382                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13383                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13384                                 mtx_unlock(&softc->ctl_lock); // XXX
13385                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13386                                     sizeof(msg.ua), M_WAITOK);
13387                                 mtx_lock(&softc->ctl_lock);
13388                         }
13389                 }
13390                 mtx_sleep(&softc->thresh_thread, &softc->ctl_lock,
13391                     PDROP, "-", CTL_LBP_PERIOD * hz);
13392         }
13393         softc->thresh_thread = NULL;
13394         kthread_exit();
13395 }
13396
13397 static void
13398 ctl_enqueue_incoming(union ctl_io *io)
13399 {
13400         struct ctl_softc *softc = CTL_SOFTC(io);
13401         struct ctl_thread *thr;
13402         u_int idx;
13403
13404         idx = (io->io_hdr.nexus.targ_port * 127 +
13405                io->io_hdr.nexus.initid) % worker_threads;
13406         thr = &softc->threads[idx];
13407         mtx_lock(&thr->queue_lock);
13408         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13409         mtx_unlock(&thr->queue_lock);
13410         wakeup(thr);
13411 }
13412
13413 static void
13414 ctl_enqueue_rtr(union ctl_io *io)
13415 {
13416         struct ctl_softc *softc = CTL_SOFTC(io);
13417         struct ctl_thread *thr;
13418
13419         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13420         mtx_lock(&thr->queue_lock);
13421         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13422         mtx_unlock(&thr->queue_lock);
13423         wakeup(thr);
13424 }
13425
13426 static void
13427 ctl_enqueue_done(union ctl_io *io)
13428 {
13429         struct ctl_softc *softc = CTL_SOFTC(io);
13430         struct ctl_thread *thr;
13431
13432         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13433         mtx_lock(&thr->queue_lock);
13434         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13435         mtx_unlock(&thr->queue_lock);
13436         wakeup(thr);
13437 }
13438
13439 static void
13440 ctl_enqueue_isc(union ctl_io *io)
13441 {
13442         struct ctl_softc *softc = CTL_SOFTC(io);
13443         struct ctl_thread *thr;
13444
13445         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13446         mtx_lock(&thr->queue_lock);
13447         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13448         mtx_unlock(&thr->queue_lock);
13449         wakeup(thr);
13450 }
13451
13452 /*
13453  *  vim: ts=8
13454  */