sys/cam/ctl/ctl.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   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/cdefs.h>
  48 __FBSDID("$FreeBSD$");
  49
  50 #include <sys/param.h>
  51 #include <sys/systm.h>
  52 #include <sys/ctype.h>
  53 #include <sys/kernel.h>
  54 #include <sys/types.h>
  55 #include <sys/kthread.h>
  56 #include <sys/bio.h>
  57 #include <sys/fcntl.h>
  58 #include <sys/lock.h>
  59 #include <sys/module.h>
  60 #include <sys/mutex.h>
  61 #include <sys/condvar.h>
  62 #include <sys/malloc.h>
  63 #include <sys/conf.h>
  64 #include <sys/ioccom.h>
  65 #include <sys/queue.h>
  66 #include <sys/sbuf.h>
  67 #include <sys/smp.h>
  68 #include <sys/endian.h>
  69 #include <sys/proc.h>
  70 #include <sys/sched.h>
  71 #include <sys/sysctl.h>
  72 #include <sys/nv.h>
  73 #include <sys/dnv.h>
  74 #include <vm/uma.h>
  75
  76 #include <cam/cam.h>
  77 #include <cam/scsi/scsi_all.h>
  78 #include <cam/scsi/scsi_cd.h>
  79 #include <cam/scsi/scsi_da.h>
  80 #include <cam/ctl/ctl_io.h>
  81 #include <cam/ctl/ctl.h>
  82 #include <cam/ctl/ctl_frontend.h>
  83 #include <cam/ctl/ctl_util.h>
  84 #include <cam/ctl/ctl_backend.h>
  85 #include <cam/ctl/ctl_ioctl.h>
  86 #include <cam/ctl/ctl_ha.h>
  87 #include <cam/ctl/ctl_private.h>
  88 #include <cam/ctl/ctl_debug.h>
  89 #include <cam/ctl/ctl_scsi_all.h>
  90 #include <cam/ctl/ctl_error.h>
  91
  92 struct ctl_softc *control_softc = NULL;
  93
  94 /*
  95  * Template mode pages.
  96  */
  97
  98 /*
  99  * Note that these are default values only.  The actual values will be
 100  * filled in when the user does a mode sense.
 101  */
 102 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
 103         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 104         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 105         /*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
 106         /*read_retry_count*/0,
 107         /*correction_span*/0,
 108         /*head_offset_count*/0,
 109         /*data_strobe_offset_cnt*/0,
 110         /*byte8*/SMS_RWER_LBPERE,
 111         /*write_retry_count*/0,
 112         /*reserved2*/0,
 113         /*recovery_time_limit*/{0, 0},
 114 };
 115
 116 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
 117         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 118         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 119         /*byte3*/SMS_RWER_PER,
 120         /*read_retry_count*/0,
 121         /*correction_span*/0,
 122         /*head_offset_count*/0,
 123         /*data_strobe_offset_cnt*/0,
 124         /*byte8*/SMS_RWER_LBPERE,
 125         /*write_retry_count*/0,
 126         /*reserved2*/0,
 127         /*recovery_time_limit*/{0, 0},
 128 };
 129
 130 const static struct scsi_format_page format_page_default = {
 131         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 132         /*page_length*/sizeof(struct scsi_format_page) - 2,
 133         /*tracks_per_zone*/ {0, 0},
 134         /*alt_sectors_per_zone*/ {0, 0},
 135         /*alt_tracks_per_zone*/ {0, 0},
 136         /*alt_tracks_per_lun*/ {0, 0},
 137         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
 138                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
 139         /*bytes_per_sector*/ {0, 0},
 140         /*interleave*/ {0, 0},
 141         /*track_skew*/ {0, 0},
 142         /*cylinder_skew*/ {0, 0},
 143         /*flags*/ SFP_HSEC,
 144         /*reserved*/ {0, 0, 0}
 145 };
 146
 147 const static struct scsi_format_page format_page_changeable = {
 148         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 149         /*page_length*/sizeof(struct scsi_format_page) - 2,
 150         /*tracks_per_zone*/ {0, 0},
 151         /*alt_sectors_per_zone*/ {0, 0},
 152         /*alt_tracks_per_zone*/ {0, 0},
 153         /*alt_tracks_per_lun*/ {0, 0},
 154         /*sectors_per_track*/ {0, 0},
 155         /*bytes_per_sector*/ {0, 0},
 156         /*interleave*/ {0, 0},
 157         /*track_skew*/ {0, 0},
 158         /*cylinder_skew*/ {0, 0},
 159         /*flags*/ 0,
 160         /*reserved*/ {0, 0, 0}
 161 };
 162
 163 const static struct scsi_rigid_disk_page rigid_disk_page_default = {
 164         /*page_code*/SMS_RIGID_DISK_PAGE,
 165         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 166         /*cylinders*/ {0, 0, 0},
 167         /*heads*/ CTL_DEFAULT_HEADS,
 168         /*start_write_precomp*/ {0, 0, 0},
 169         /*start_reduced_current*/ {0, 0, 0},
 170         /*step_rate*/ {0, 0},
 171         /*landing_zone_cylinder*/ {0, 0, 0},
 172         /*rpl*/ SRDP_RPL_DISABLED,
 173         /*rotational_offset*/ 0,
 174         /*reserved1*/ 0,
 175         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
 176                            CTL_DEFAULT_ROTATION_RATE & 0xff},
 177         /*reserved2*/ {0, 0}
 178 };
 179
 180 const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
 181         /*page_code*/SMS_RIGID_DISK_PAGE,
 182         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 183         /*cylinders*/ {0, 0, 0},
 184         /*heads*/ 0,
 185         /*start_write_precomp*/ {0, 0, 0},
 186         /*start_reduced_current*/ {0, 0, 0},
 187         /*step_rate*/ {0, 0},
 188         /*landing_zone_cylinder*/ {0, 0, 0},
 189         /*rpl*/ 0,
 190         /*rotational_offset*/ 0,
 191         /*reserved1*/ 0,
 192         /*rotation_rate*/ {0, 0},
 193         /*reserved2*/ {0, 0}
 194 };
 195
 196 const static struct scsi_da_verify_recovery_page verify_er_page_default = {
 197         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 198         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 199         /*byte3*/0,
 200         /*read_retry_count*/0,
 201         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 202         /*recovery_time_limit*/{0, 0},
 203 };
 204
 205 const static struct scsi_da_verify_recovery_page verify_er_page_changeable = {
 206         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 207         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 208         /*byte3*/SMS_VER_PER,
 209         /*read_retry_count*/0,
 210         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 211         /*recovery_time_limit*/{0, 0},
 212 };
 213
 214 const static struct scsi_caching_page caching_page_default = {
 215         /*page_code*/SMS_CACHING_PAGE,
 216         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 217         /*flags1*/ SCP_DISC | SCP_WCE,
 218         /*ret_priority*/ 0,
 219         /*disable_pf_transfer_len*/ {0xff, 0xff},
 220         /*min_prefetch*/ {0, 0},
 221         /*max_prefetch*/ {0xff, 0xff},
 222         /*max_pf_ceiling*/ {0xff, 0xff},
 223         /*flags2*/ 0,
 224         /*cache_segments*/ 0,
 225         /*cache_seg_size*/ {0, 0},
 226         /*reserved*/ 0,
 227         /*non_cache_seg_size*/ {0, 0, 0}
 228 };
 229
 230 const static struct scsi_caching_page caching_page_changeable = {
 231         /*page_code*/SMS_CACHING_PAGE,
 232         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 233         /*flags1*/ SCP_WCE | SCP_RCD,
 234         /*ret_priority*/ 0,
 235         /*disable_pf_transfer_len*/ {0, 0},
 236         /*min_prefetch*/ {0, 0},
 237         /*max_prefetch*/ {0, 0},
 238         /*max_pf_ceiling*/ {0, 0},
 239         /*flags2*/ 0,
 240         /*cache_segments*/ 0,
 241         /*cache_seg_size*/ {0, 0},
 242         /*reserved*/ 0,
 243         /*non_cache_seg_size*/ {0, 0, 0}
 244 };
 245
 246 const static struct scsi_control_page control_page_default = {
 247         /*page_code*/SMS_CONTROL_MODE_PAGE,
 248         /*page_length*/sizeof(struct scsi_control_page) - 2,
 249         /*rlec*/0,
 250         /*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
 251         /*eca_and_aen*/0,
 252         /*flags4*/SCP_TAS,
 253         /*aen_holdoff_period*/{0, 0},
 254         /*busy_timeout_period*/{0, 0},
 255         /*extended_selftest_completion_time*/{0, 0}
 256 };
 257
 258 const static struct scsi_control_page control_page_changeable = {
 259         /*page_code*/SMS_CONTROL_MODE_PAGE,
 260         /*page_length*/sizeof(struct scsi_control_page) - 2,
 261         /*rlec*/SCP_DSENSE,
 262         /*queue_flags*/SCP_QUEUE_ALG_MASK | SCP_NUAR,
 263         /*eca_and_aen*/SCP_SWP,
 264         /*flags4*/0,
 265         /*aen_holdoff_period*/{0, 0},
 266         /*busy_timeout_period*/{0, 0},
 267         /*extended_selftest_completion_time*/{0, 0}
 268 };
 269
 270 #define CTL_CEM_LEN     (sizeof(struct scsi_control_ext_page) - 4)
 271
 272 const static struct scsi_control_ext_page control_ext_page_default = {
 273         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 274         /*subpage_code*/0x01,
 275         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 276         /*flags*/0,
 277         /*prio*/0,
 278         /*max_sense*/0
 279 };
 280
 281 const static struct scsi_control_ext_page control_ext_page_changeable = {
 282         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 283         /*subpage_code*/0x01,
 284         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 285         /*flags*/0,
 286         /*prio*/0,
 287         /*max_sense*/0xff
 288 };
 289
 290 const static struct scsi_info_exceptions_page ie_page_default = {
 291         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 292         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 293         /*info_flags*/SIEP_FLAGS_EWASC,
 294         /*mrie*/SIEP_MRIE_NO,
 295         /*interval_timer*/{0, 0, 0, 0},
 296         /*report_count*/{0, 0, 0, 1}
 297 };
 298
 299 const static struct scsi_info_exceptions_page ie_page_changeable = {
 300         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 301         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 302         /*info_flags*/SIEP_FLAGS_EWASC | SIEP_FLAGS_DEXCPT | SIEP_FLAGS_TEST |
 303             SIEP_FLAGS_LOGERR,
 304         /*mrie*/0x0f,
 305         /*interval_timer*/{0xff, 0xff, 0xff, 0xff},
 306         /*report_count*/{0xff, 0xff, 0xff, 0xff}
 307 };
 308
 309 #define CTL_LBPM_LEN    (sizeof(struct ctl_logical_block_provisioning_page) - 4)
 310
 311 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
 312         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 313         /*subpage_code*/0x02,
 314         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 315         /*flags*/0,
 316         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 317         /*descr*/{}},
 318         {{/*flags*/0,
 319           /*resource*/0x01,
 320           /*reserved*/{0, 0},
 321           /*count*/{0, 0, 0, 0}},
 322          {/*flags*/0,
 323           /*resource*/0x02,
 324           /*reserved*/{0, 0},
 325           /*count*/{0, 0, 0, 0}},
 326          {/*flags*/0,
 327           /*resource*/0xf1,
 328           /*reserved*/{0, 0},
 329           /*count*/{0, 0, 0, 0}},
 330          {/*flags*/0,
 331           /*resource*/0xf2,
 332           /*reserved*/{0, 0},
 333           /*count*/{0, 0, 0, 0}}
 334         }
 335 };
 336
 337 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
 338         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 339         /*subpage_code*/0x02,
 340         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 341         /*flags*/SLBPP_SITUA,
 342         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 343         /*descr*/{}},
 344         {{/*flags*/0,
 345           /*resource*/0,
 346           /*reserved*/{0, 0},
 347           /*count*/{0, 0, 0, 0}},
 348          {/*flags*/0,
 349           /*resource*/0,
 350           /*reserved*/{0, 0},
 351           /*count*/{0, 0, 0, 0}},
 352          {/*flags*/0,
 353           /*resource*/0,
 354           /*reserved*/{0, 0},
 355           /*count*/{0, 0, 0, 0}},
 356          {/*flags*/0,
 357           /*resource*/0,
 358           /*reserved*/{0, 0},
 359           /*count*/{0, 0, 0, 0}}
 360         }
 361 };
 362
 363 const static struct scsi_cddvd_capabilities_page cddvd_page_default = {
 364         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 365         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 366         /*caps1*/0x3f,
 367         /*caps2*/0x00,
 368         /*caps3*/0xf0,
 369         /*caps4*/0x00,
 370         /*caps5*/0x29,
 371         /*caps6*/0x00,
 372         /*obsolete*/{0, 0},
 373         /*nvol_levels*/{0, 0},
 374         /*buffer_size*/{8, 0},
 375         /*obsolete2*/{0, 0},
 376         /*reserved*/0,
 377         /*digital*/0,
 378         /*obsolete3*/0,
 379         /*copy_management*/0,
 380         /*reserved2*/0,
 381         /*rotation_control*/0,
 382         /*cur_write_speed*/0,
 383         /*num_speed_descr*/0,
 384 };
 385
 386 const static struct scsi_cddvd_capabilities_page cddvd_page_changeable = {
 387         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 388         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 389         /*caps1*/0,
 390         /*caps2*/0,
 391         /*caps3*/0,
 392         /*caps4*/0,
 393         /*caps5*/0,
 394         /*caps6*/0,
 395         /*obsolete*/{0, 0},
 396         /*nvol_levels*/{0, 0},
 397         /*buffer_size*/{0, 0},
 398         /*obsolete2*/{0, 0},
 399         /*reserved*/0,
 400         /*digital*/0,
 401         /*obsolete3*/0,
 402         /*copy_management*/0,
 403         /*reserved2*/0,
 404         /*rotation_control*/0,
 405         /*cur_write_speed*/0,
 406         /*num_speed_descr*/0,
 407 };
 408
 409 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
 410     "CAM Target Layer");
 411 static int worker_threads = -1;
 412 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 413     &worker_threads, 1, "Number of worker threads");
 414 static int ctl_debug = CTL_DEBUG_NONE;
 415 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
 416     &ctl_debug, 0, "Enabled debug flags");
 417 static int ctl_lun_map_size = 1024;
 418 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, lun_map_size, CTLFLAG_RWTUN,
 419     &ctl_lun_map_size, 0, "Size of per-port LUN map (max LUN + 1)");
 420 #ifdef  CTL_TIME_IO
 421 static int ctl_time_io_secs = CTL_TIME_IO_DEFAULT_SECS;
 422 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, time_io_secs, CTLFLAG_RWTUN,
 423     &ctl_time_io_secs, 0, "Log requests taking more seconds");
 424 #endif
 425
 426 /*
 427  * Maximum number of LUNs we support.  MUST be a power of 2.
 428  */
 429 #define CTL_DEFAULT_MAX_LUNS    1024
 430 static int ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
 431 TUNABLE_INT("kern.cam.ctl.max_luns", &ctl_max_luns);
 432 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_luns, CTLFLAG_RDTUN,
 433     &ctl_max_luns, CTL_DEFAULT_MAX_LUNS, "Maximum number of LUNs");
 434
 435 /*
 436  * Maximum number of ports registered at one time.
 437  */
 438 #define CTL_DEFAULT_MAX_PORTS           256
 439 static int ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
 440 TUNABLE_INT("kern.cam.ctl.max_ports", &ctl_max_ports);
 441 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_ports, CTLFLAG_RDTUN,
 442     &ctl_max_ports, CTL_DEFAULT_MAX_LUNS, "Maximum number of ports");
 443
 444 /*
 445  * Maximum number of initiators we support.
 446  */
 447 #define CTL_MAX_INITIATORS      (CTL_MAX_INIT_PER_PORT * ctl_max_ports)
 448
 449 /*
 450  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 451  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 452  * SCSI Ports (0x88), Third-party Copy (0x8F), SCSI Feature Sets (0x92),
 453  * Block limits (0xB0), Block Device Characteristics (0xB1) and
 454  * Logical Block Provisioning (0xB2)
 455  */
 456 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   11
 457
 458 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 459                                   int param);
 460 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 461 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
 462 static int ctl_init(void);
 463 static int ctl_shutdown(void);
 464 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 465 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 466 static void ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 467 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 468                               struct ctl_ooa *ooa_hdr,
 469                               struct ctl_ooa_entry *kern_entries);
 470 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 471                      struct thread *td);
 472 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 473                          struct ctl_be_lun *be_lun);
 474 static int ctl_free_lun(struct ctl_lun *lun);
 475 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 476
 477 static int ctl_do_mode_select(union ctl_io *io);
 478 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 479                            uint64_t res_key, uint64_t sa_res_key,
 480                            uint8_t type, uint32_t residx,
 481                            struct ctl_scsiio *ctsio,
 482                            struct scsi_per_res_out *cdb,
 483                            struct scsi_per_res_out_parms* param);
 484 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 485                                   union ctl_ha_msg *msg);
 486 static void ctl_hndl_per_res_out_on_other_sc(union ctl_io *io);
 487 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 488 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 489 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 490 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 491 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 492 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 493                                          int alloc_len);
 494 static int ctl_inquiry_evpd_sfs(struct ctl_scsiio *ctsio, int alloc_len);
 495 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 496                                          int alloc_len);
 497 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 498 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 499 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 500 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 501 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 502 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
 503     bool seq);
 504 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
 505 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 506     union ctl_io *pending_io, union ctl_io *ooa_io);
 507 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 508                                 union ctl_io **starting_io);
 509 static void ctl_try_unblock_io(struct ctl_lun *lun, union ctl_io *io,
 510     bool skip);
 511 static void ctl_try_unblock_others(struct ctl_lun *lun, union ctl_io *io,
 512     bool skip);
 513 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
 514                                 const struct ctl_cmd_entry *entry,
 515                                 struct ctl_scsiio *ctsio);
 516 static void ctl_failover_lun(union ctl_io *io);
 517 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 518                                struct ctl_scsiio *ctsio);
 519 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 520
 521 static int ctl_target_reset(union ctl_io *io);
 522 static void ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx,
 523                          ctl_ua_type ua_type);
 524 static int ctl_lun_reset(union ctl_io *io);
 525 static int ctl_abort_task(union ctl_io *io);
 526 static int ctl_abort_task_set(union ctl_io *io);
 527 static int ctl_query_task(union ctl_io *io, int task_set);
 528 static void ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
 529                               ctl_ua_type ua_type);
 530 static int ctl_i_t_nexus_reset(union ctl_io *io);
 531 static int ctl_query_async_event(union ctl_io *io);
 532 static void ctl_run_task(union ctl_io *io);
 533 #ifdef CTL_IO_DELAY
 534 static void ctl_datamove_timer_wakeup(void *arg);
 535 static void ctl_done_timer_wakeup(void *arg);
 536 #endif /* CTL_IO_DELAY */
 537
 538 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 539 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 540 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 541 static void ctl_datamove_remote_write(union ctl_io *io);
 542 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 543 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 544 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 545 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 546                                     ctl_ha_dt_cb callback);
 547 static void ctl_datamove_remote_read(union ctl_io *io);
 548 static void ctl_datamove_remote(union ctl_io *io);
 549 static void ctl_process_done(union ctl_io *io);
 550 static void ctl_lun_thread(void *arg);
 551 static void ctl_thresh_thread(void *arg);
 552 static void ctl_work_thread(void *arg);
 553 static void ctl_enqueue_incoming(union ctl_io *io);
 554 static void ctl_enqueue_rtr(union ctl_io *io);
 555 static void ctl_enqueue_done(union ctl_io *io);
 556 static void ctl_enqueue_isc(union ctl_io *io);
 557 static const struct ctl_cmd_entry *
 558     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 559 static const struct ctl_cmd_entry *
 560     ctl_validate_command(struct ctl_scsiio *ctsio);
 561 static int ctl_cmd_applicable(uint8_t lun_type,
 562     const struct ctl_cmd_entry *entry);
 563 static int ctl_ha_init(void);
 564 static int ctl_ha_shutdown(void);
 565
 566 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
 567 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
 568 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
 569 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
 570
 571 /*
 572  * Load the serialization table.  This isn't very pretty, but is probably
 573  * the easiest way to do it.
 574  */
 575 #include "ctl_ser_table.c"
 576
 577 /*
 578  * We only need to define open, close and ioctl routines for this driver.
 579  */
 580 static struct cdevsw ctl_cdevsw = {
 581         .d_version =    D_VERSION,
 582         .d_flags =      0,
 583         .d_open =       ctl_open,
 584         .d_close =      ctl_close,
 585         .d_ioctl =      ctl_ioctl,
 586         .d_name =       "ctl",
 587 };
 588
 589
 590 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 591
 592 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 593
 594 static moduledata_t ctl_moduledata = {
 595         "ctl",
 596         ctl_module_event_handler,
 597         NULL
 598 };
 599
 600 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 601 MODULE_VERSION(ctl, 1);
 602
 603 static struct ctl_frontend ha_frontend =
 604 {
 605         .name = "ha",
 606         .init = ctl_ha_init,
 607         .shutdown = ctl_ha_shutdown,
 608 };
 609
 610 static int
 611 ctl_ha_init(void)
 612 {
 613         struct ctl_softc *softc = control_softc;
 614
 615         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
 616                             &softc->othersc_pool) != 0)
 617                 return (ENOMEM);
 618         if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
 619                 ctl_pool_free(softc->othersc_pool);
 620                 return (EIO);
 621         }
 622         if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
 623             != CTL_HA_STATUS_SUCCESS) {
 624                 ctl_ha_msg_destroy(softc);
 625                 ctl_pool_free(softc->othersc_pool);
 626                 return (EIO);
 627         }
 628         return (0);
 629 };
 630
 631 static int
 632 ctl_ha_shutdown(void)
 633 {
 634         struct ctl_softc *softc = control_softc;
 635         struct ctl_port *port;
 636
 637         ctl_ha_msg_shutdown(softc);
 638         if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL) != CTL_HA_STATUS_SUCCESS)
 639                 return (EIO);
 640         if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
 641                 return (EIO);
 642         ctl_pool_free(softc->othersc_pool);
 643         while ((port = STAILQ_FIRST(&ha_frontend.port_list)) != NULL) {
 644                 ctl_port_deregister(port);
 645                 free(port->port_name, M_CTL);
 646                 free(port, M_CTL);
 647         }
 648         return (0);
 649 };
 650
 651 static void
 652 ctl_ha_datamove(union ctl_io *io)
 653 {
 654         struct ctl_lun *lun = CTL_LUN(io);
 655         struct ctl_sg_entry *sgl;
 656         union ctl_ha_msg msg;
 657         uint32_t sg_entries_sent;
 658         int do_sg_copy, i, j;
 659
 660         memset(&msg.dt, 0, sizeof(msg.dt));
 661         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
 662         msg.hdr.original_sc = io->io_hdr.remote_io;
 663         msg.hdr.serializing_sc = io;
 664         msg.hdr.nexus = io->io_hdr.nexus;
 665         msg.hdr.status = io->io_hdr.status;
 666         msg.dt.flags = io->io_hdr.flags;
 667
 668         /*
 669          * We convert everything into a S/G list here.  We can't
 670          * pass by reference, only by value between controllers.
 671          * So we can't pass a pointer to the S/G list, only as many
 672          * S/G entries as we can fit in here.  If it's possible for
 673          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
 674          * then we need to break this up into multiple transfers.
 675          */
 676         if (io->scsiio.kern_sg_entries == 0) {
 677                 msg.dt.kern_sg_entries = 1;
 678 #if 0
 679                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 680                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 681                 } else {
 682                         /* XXX KDM use busdma here! */
 683                         msg.dt.sg_list[0].addr =
 684                             (void *)vtophys(io->scsiio.kern_data_ptr);
 685                 }
 686 #else
 687                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
 688                     ("HA does not support BUS_ADDR"));
 689                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 690 #endif
 691                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
 692                 do_sg_copy = 0;
 693         } else {
 694                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
 695                 do_sg_copy = 1;
 696         }
 697
 698         msg.dt.kern_data_len = io->scsiio.kern_data_len;
 699         msg.dt.kern_total_len = io->scsiio.kern_total_len;
 700         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
 701         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
 702         msg.dt.sg_sequence = 0;
 703
 704         /*
 705          * Loop until we've sent all of the S/G entries.  On the
 706          * other end, we'll recompose these S/G entries into one
 707          * contiguous list before processing.
 708          */
 709         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
 710             msg.dt.sg_sequence++) {
 711                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
 712                     sizeof(msg.dt.sg_list[0])),
 713                     msg.dt.kern_sg_entries - sg_entries_sent);
 714                 if (do_sg_copy != 0) {
 715                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
 716                         for (i = sg_entries_sent, j = 0;
 717                              i < msg.dt.cur_sg_entries; i++, j++) {
 718 #if 0
 719                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 720                                         msg.dt.sg_list[j].addr = sgl[i].addr;
 721                                 } else {
 722                                         /* XXX KDM use busdma here! */
 723                                         msg.dt.sg_list[j].addr =
 724                                             (void *)vtophys(sgl[i].addr);
 725                                 }
 726 #else
 727                                 KASSERT((io->io_hdr.flags &
 728                                     CTL_FLAG_BUS_ADDR) == 0,
 729                                     ("HA does not support BUS_ADDR"));
 730                                 msg.dt.sg_list[j].addr = sgl[i].addr;
 731 #endif
 732                                 msg.dt.sg_list[j].len = sgl[i].len;
 733                         }
 734                 }
 735
 736                 sg_entries_sent += msg.dt.cur_sg_entries;
 737                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
 738                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 739                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
 740                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
 741                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
 742                         io->io_hdr.port_status = 31341;
 743                         io->scsiio.be_move_done(io);
 744                         return;
 745                 }
 746                 msg.dt.sent_sg_entries = sg_entries_sent;
 747         }
 748
 749         /*
 750          * Officially handover the request from us to peer.
 751          * If failover has just happened, then we must return error.
 752          * If failover happen just after, then it is not our problem.
 753          */
 754         if (lun)
 755                 mtx_lock(&lun->lun_lock);
 756         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
 757                 if (lun)
 758                         mtx_unlock(&lun->lun_lock);
 759                 io->io_hdr.port_status = 31342;
 760                 io->scsiio.be_move_done(io);
 761                 return;
 762         }
 763         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
 764         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
 765         if (lun)
 766                 mtx_unlock(&lun->lun_lock);
 767 }
 768
 769 static void
 770 ctl_ha_done(union ctl_io *io)
 771 {
 772         union ctl_ha_msg msg;
 773
 774         if (io->io_hdr.io_type == CTL_IO_SCSI) {
 775                 memset(&msg, 0, sizeof(msg));
 776                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
 777                 msg.hdr.original_sc = io->io_hdr.remote_io;
 778                 msg.hdr.nexus = io->io_hdr.nexus;
 779                 msg.hdr.status = io->io_hdr.status;
 780                 msg.scsi.scsi_status = io->scsiio.scsi_status;
 781                 msg.scsi.tag_num = io->scsiio.tag_num;
 782                 msg.scsi.tag_type = io->scsiio.tag_type;
 783                 msg.scsi.sense_len = io->scsiio.sense_len;
 784                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
 785                     io->scsiio.sense_len);
 786                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 787                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
 788                     msg.scsi.sense_len, M_WAITOK);
 789         }
 790         ctl_free_io(io);
 791 }
 792
 793 static void
 794 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 795                             union ctl_ha_msg *msg_info)
 796 {
 797         struct ctl_scsiio *ctsio;
 798
 799         if (msg_info->hdr.original_sc == NULL) {
 800                 printf("%s: original_sc == NULL!\n", __func__);
 801                 /* XXX KDM now what? */
 802                 return;
 803         }
 804
 805         ctsio = &msg_info->hdr.original_sc->scsiio;
 806         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 807         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 808         ctsio->io_hdr.status = msg_info->hdr.status;
 809         ctsio->scsi_status = msg_info->scsi.scsi_status;
 810         ctsio->sense_len = msg_info->scsi.sense_len;
 811         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 812                msg_info->scsi.sense_len);
 813         ctl_enqueue_isc((union ctl_io *)ctsio);
 814 }
 815
 816 static void
 817 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 818                                 union ctl_ha_msg *msg_info)
 819 {
 820         struct ctl_scsiio *ctsio;
 821
 822         if (msg_info->hdr.serializing_sc == NULL) {
 823                 printf("%s: serializing_sc == NULL!\n", __func__);
 824                 /* XXX KDM now what? */
 825                 return;
 826         }
 827
 828         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 829         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 830         ctl_enqueue_isc((union ctl_io *)ctsio);
 831 }
 832
 833 void
 834 ctl_isc_announce_lun(struct ctl_lun *lun)
 835 {
 836         struct ctl_softc *softc = lun->ctl_softc;
 837         union ctl_ha_msg *msg;
 838         struct ctl_ha_msg_lun_pr_key pr_key;
 839         int i, k;
 840
 841         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 842                 return;
 843         mtx_lock(&lun->lun_lock);
 844         i = sizeof(msg->lun);
 845         if (lun->lun_devid)
 846                 i += lun->lun_devid->len;
 847         i += sizeof(pr_key) * lun->pr_key_count;
 848 alloc:
 849         mtx_unlock(&lun->lun_lock);
 850         msg = malloc(i, M_CTL, M_WAITOK);
 851         mtx_lock(&lun->lun_lock);
 852         k = sizeof(msg->lun);
 853         if (lun->lun_devid)
 854                 k += lun->lun_devid->len;
 855         k += sizeof(pr_key) * lun->pr_key_count;
 856         if (i < k) {
 857                 free(msg, M_CTL);
 858                 i = k;
 859                 goto alloc;
 860         }
 861         bzero(&msg->lun, sizeof(msg->lun));
 862         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 863         msg->hdr.nexus.targ_lun = lun->lun;
 864         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 865         msg->lun.flags = lun->flags;
 866         msg->lun.pr_generation = lun->pr_generation;
 867         msg->lun.pr_res_idx = lun->pr_res_idx;
 868         msg->lun.pr_res_type = lun->pr_res_type;
 869         msg->lun.pr_key_count = lun->pr_key_count;
 870         i = 0;
 871         if (lun->lun_devid) {
 872                 msg->lun.lun_devid_len = lun->lun_devid->len;
 873                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 874                     msg->lun.lun_devid_len);
 875                 i += msg->lun.lun_devid_len;
 876         }
 877         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 878                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 879                         continue;
 880                 pr_key.pr_iid = k;
 881                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 882                 i += sizeof(pr_key);
 883         }
 884         mtx_unlock(&lun->lun_lock);
 885         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 886             M_WAITOK);
 887         free(msg, M_CTL);
 888
 889         if (lun->flags & CTL_LUN_PRIMARY_SC) {
 890                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 891                         ctl_isc_announce_mode(lun, -1,
 892                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
 893                             lun->mode_pages.index[i].subpage);
 894                 }
 895         }
 896 }
 897
 898 void
 899 ctl_isc_announce_port(struct ctl_port *port)
 900 {
 901         struct ctl_softc *softc = port->ctl_softc;
 902         union ctl_ha_msg *msg;
 903         int i;
 904
 905         if (port->targ_port < softc->port_min ||
 906             port->targ_port >= softc->port_max ||
 907             softc->ha_link != CTL_HA_LINK_ONLINE)
 908                 return;
 909         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 910         if (port->lun_map)
 911                 i += port->lun_map_size * sizeof(uint32_t);
 912         if (port->port_devid)
 913                 i += port->port_devid->len;
 914         if (port->target_devid)
 915                 i += port->target_devid->len;
 916         if (port->init_devid)
 917                 i += port->init_devid->len;
 918         msg = malloc(i, M_CTL, M_WAITOK);
 919         bzero(&msg->port, sizeof(msg->port));
 920         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 921         msg->hdr.nexus.targ_port = port->targ_port;
 922         msg->port.port_type = port->port_type;
 923         msg->port.physical_port = port->physical_port;
 924         msg->port.virtual_port = port->virtual_port;
 925         msg->port.status = port->status;
 926         i = 0;
 927         msg->port.name_len = sprintf(&msg->port.data[i],
 928             "%d:%s", softc->ha_id, port->port_name) + 1;
 929         i += msg->port.name_len;
 930         if (port->lun_map) {
 931                 msg->port.lun_map_len = port->lun_map_size * sizeof(uint32_t);
 932                 memcpy(&msg->port.data[i], port->lun_map,
 933                     msg->port.lun_map_len);
 934                 i += msg->port.lun_map_len;
 935         }
 936         if (port->port_devid) {
 937                 msg->port.port_devid_len = port->port_devid->len;
 938                 memcpy(&msg->port.data[i], port->port_devid->data,
 939                     msg->port.port_devid_len);
 940                 i += msg->port.port_devid_len;
 941         }
 942         if (port->target_devid) {
 943                 msg->port.target_devid_len = port->target_devid->len;
 944                 memcpy(&msg->port.data[i], port->target_devid->data,
 945                     msg->port.target_devid_len);
 946                 i += msg->port.target_devid_len;
 947         }
 948         if (port->init_devid) {
 949                 msg->port.init_devid_len = port->init_devid->len;
 950                 memcpy(&msg->port.data[i], port->init_devid->data,
 951                     msg->port.init_devid_len);
 952                 i += msg->port.init_devid_len;
 953         }
 954         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 955             M_WAITOK);
 956         free(msg, M_CTL);
 957 }
 958
 959 void
 960 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 961 {
 962         struct ctl_softc *softc = port->ctl_softc;
 963         union ctl_ha_msg *msg;
 964         int i, l;
 965
 966         if (port->targ_port < softc->port_min ||
 967             port->targ_port >= softc->port_max ||
 968             softc->ha_link != CTL_HA_LINK_ONLINE)
 969                 return;
 970         mtx_lock(&softc->ctl_lock);
 971         i = sizeof(msg->iid);
 972         l = 0;
 973         if (port->wwpn_iid[iid].name)
 974                 l = strlen(port->wwpn_iid[iid].name) + 1;
 975         i += l;
 976         msg = malloc(i, M_CTL, M_NOWAIT);
 977         if (msg == NULL) {
 978                 mtx_unlock(&softc->ctl_lock);
 979                 return;
 980         }
 981         bzero(&msg->iid, sizeof(msg->iid));
 982         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 983         msg->hdr.nexus.targ_port = port->targ_port;
 984         msg->hdr.nexus.initid = iid;
 985         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 986         msg->iid.name_len = l;
 987         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 988         if (port->wwpn_iid[iid].name)
 989                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 990         mtx_unlock(&softc->ctl_lock);
 991         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 992         free(msg, M_CTL);
 993 }
 994
 995 void
 996 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
 997     uint8_t page, uint8_t subpage)
 998 {
 999         struct ctl_softc *softc = lun->ctl_softc;
1000         union ctl_ha_msg msg;
1001         u_int i;
1002
1003         if (softc->ha_link != CTL_HA_LINK_ONLINE)
1004                 return;
1005         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1006                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1007                     page && lun->mode_pages.index[i].subpage == subpage)
1008                         break;
1009         }
1010         if (i == CTL_NUM_MODE_PAGES)
1011                 return;
1012
1013         /* Don't try to replicate pages not present on this device. */
1014         if (lun->mode_pages.index[i].page_data == NULL)
1015                 return;
1016
1017         bzero(&msg.mode, sizeof(msg.mode));
1018         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
1019         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
1020         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
1021         msg.hdr.nexus.targ_lun = lun->lun;
1022         msg.hdr.nexus.targ_mapped_lun = lun->lun;
1023         msg.mode.page_code = page;
1024         msg.mode.subpage = subpage;
1025         msg.mode.page_len = lun->mode_pages.index[i].page_len;
1026         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
1027             msg.mode.page_len);
1028         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
1029             M_WAITOK);
1030 }
1031
1032 static void
1033 ctl_isc_ha_link_up(struct ctl_softc *softc)
1034 {
1035         struct ctl_port *port;
1036         struct ctl_lun *lun;
1037         union ctl_ha_msg msg;
1038         int i;
1039
1040         /* Announce this node parameters to peer for validation. */
1041         msg.login.msg_type = CTL_MSG_LOGIN;
1042         msg.login.version = CTL_HA_VERSION;
1043         msg.login.ha_mode = softc->ha_mode;
1044         msg.login.ha_id = softc->ha_id;
1045         msg.login.max_luns = ctl_max_luns;
1046         msg.login.max_ports = ctl_max_ports;
1047         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
1048         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
1049             M_WAITOK);
1050
1051         STAILQ_FOREACH(port, &softc->port_list, links) {
1052                 ctl_isc_announce_port(port);
1053                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1054                         if (port->wwpn_iid[i].in_use)
1055                                 ctl_isc_announce_iid(port, i);
1056                 }
1057         }
1058         STAILQ_FOREACH(lun, &softc->lun_list, links)
1059                 ctl_isc_announce_lun(lun);
1060 }
1061
1062 static void
1063 ctl_isc_ha_link_down(struct ctl_softc *softc)
1064 {
1065         struct ctl_port *port;
1066         struct ctl_lun *lun;
1067         union ctl_io *io;
1068         int i;
1069
1070         mtx_lock(&softc->ctl_lock);
1071         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1072                 mtx_lock(&lun->lun_lock);
1073                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
1074                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1075                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1076                 }
1077                 mtx_unlock(&lun->lun_lock);
1078
1079                 mtx_unlock(&softc->ctl_lock);
1080                 io = ctl_alloc_io(softc->othersc_pool);
1081                 mtx_lock(&softc->ctl_lock);
1082                 ctl_zero_io(io);
1083                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1084                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1085                 ctl_enqueue_isc(io);
1086         }
1087
1088         STAILQ_FOREACH(port, &softc->port_list, links) {
1089                 if (port->targ_port >= softc->port_min &&
1090                     port->targ_port < softc->port_max)
1091                         continue;
1092                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1093                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1094                         port->wwpn_iid[i].in_use = 0;
1095                         free(port->wwpn_iid[i].name, M_CTL);
1096                         port->wwpn_iid[i].name = NULL;
1097                 }
1098         }
1099         mtx_unlock(&softc->ctl_lock);
1100 }
1101
1102 static void
1103 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1104 {
1105         struct ctl_lun *lun;
1106         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
1107
1108         mtx_lock(&softc->ctl_lock);
1109         if (msg->hdr.nexus.targ_mapped_lun >= ctl_max_luns ||
1110             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) == NULL) {
1111                 mtx_unlock(&softc->ctl_lock);
1112                 return;
1113         }
1114         mtx_lock(&lun->lun_lock);
1115         mtx_unlock(&softc->ctl_lock);
1116         if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES && msg->ua.ua_set)
1117                 memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1118         if (msg->ua.ua_all) {
1119                 if (msg->ua.ua_set)
1120                         ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1121                 else
1122                         ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1123         } else {
1124                 if (msg->ua.ua_set)
1125                         ctl_est_ua(lun, iid, msg->ua.ua_type);
1126                 else
1127                         ctl_clr_ua(lun, iid, msg->ua.ua_type);
1128         }
1129         mtx_unlock(&lun->lun_lock);
1130 }
1131
1132 static void
1133 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1134 {
1135         struct ctl_lun *lun;
1136         struct ctl_ha_msg_lun_pr_key pr_key;
1137         int i, k;
1138         ctl_lun_flags oflags;
1139         uint32_t targ_lun;
1140
1141         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1142         mtx_lock(&softc->ctl_lock);
1143         if (targ_lun >= ctl_max_luns ||
1144             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1145                 mtx_unlock(&softc->ctl_lock);
1146                 return;
1147         }
1148         mtx_lock(&lun->lun_lock);
1149         mtx_unlock(&softc->ctl_lock);
1150         if (lun->flags & CTL_LUN_DISABLED) {
1151                 mtx_unlock(&lun->lun_lock);
1152                 return;
1153         }
1154         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1155         if (msg->lun.lun_devid_len != i || (i > 0 &&
1156             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1157                 mtx_unlock(&lun->lun_lock);
1158                 printf("%s: Received conflicting HA LUN %d\n",
1159                     __func__, targ_lun);
1160                 return;
1161         } else {
1162                 /* Record whether peer is primary. */
1163                 oflags = lun->flags;
1164                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1165                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1166                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1167                 else
1168                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1169                 if (oflags != lun->flags)
1170                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1171
1172                 /* If peer is primary and we are not -- use data */
1173                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1174                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1175                         lun->pr_generation = msg->lun.pr_generation;
1176                         lun->pr_res_idx = msg->lun.pr_res_idx;
1177                         lun->pr_res_type = msg->lun.pr_res_type;
1178                         lun->pr_key_count = msg->lun.pr_key_count;
1179                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1180                                 ctl_clr_prkey(lun, k);
1181                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1182                                 memcpy(&pr_key, &msg->lun.data[i],
1183                                     sizeof(pr_key));
1184                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1185                                 ctl_set_prkey(lun, pr_key.pr_iid,
1186                                     pr_key.pr_key);
1187                                 i += sizeof(pr_key);
1188                         }
1189                 }
1190
1191                 mtx_unlock(&lun->lun_lock);
1192                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1193                     __func__, targ_lun,
1194                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1195                     "primary" : "secondary"));
1196
1197                 /* If we are primary but peer doesn't know -- notify */
1198                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1199                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1200                         ctl_isc_announce_lun(lun);
1201         }
1202 }
1203
1204 static void
1205 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1206 {
1207         struct ctl_port *port;
1208         struct ctl_lun *lun;
1209         int i, new;
1210
1211         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1212         if (port == NULL) {
1213                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1214                     msg->hdr.nexus.targ_port));
1215                 new = 1;
1216                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1217                 port->frontend = &ha_frontend;
1218                 port->targ_port = msg->hdr.nexus.targ_port;
1219                 port->fe_datamove = ctl_ha_datamove;
1220                 port->fe_done = ctl_ha_done;
1221         } else if (port->frontend == &ha_frontend) {
1222                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1223                     msg->hdr.nexus.targ_port));
1224                 new = 0;
1225         } else {
1226                 printf("%s: Received conflicting HA port %d\n",
1227                     __func__, msg->hdr.nexus.targ_port);
1228                 return;
1229         }
1230         port->port_type = msg->port.port_type;
1231         port->physical_port = msg->port.physical_port;
1232         port->virtual_port = msg->port.virtual_port;
1233         port->status = msg->port.status;
1234         i = 0;
1235         free(port->port_name, M_CTL);
1236         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1237             M_CTL);
1238         i += msg->port.name_len;
1239         if (msg->port.lun_map_len != 0) {
1240                 if (port->lun_map == NULL ||
1241                     port->lun_map_size * sizeof(uint32_t) <
1242                     msg->port.lun_map_len) {
1243                         port->lun_map_size = 0;
1244                         free(port->lun_map, M_CTL);
1245                         port->lun_map = malloc(msg->port.lun_map_len,
1246                             M_CTL, M_WAITOK);
1247                 }
1248                 memcpy(port->lun_map, &msg->port.data[i], msg->port.lun_map_len);
1249                 port->lun_map_size = msg->port.lun_map_len / sizeof(uint32_t);
1250                 i += msg->port.lun_map_len;
1251         } else {
1252                 port->lun_map_size = 0;
1253                 free(port->lun_map, M_CTL);
1254                 port->lun_map = NULL;
1255         }
1256         if (msg->port.port_devid_len != 0) {
1257                 if (port->port_devid == NULL ||
1258                     port->port_devid->len < msg->port.port_devid_len) {
1259                         free(port->port_devid, M_CTL);
1260                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1261                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1262                 }
1263                 memcpy(port->port_devid->data, &msg->port.data[i],
1264                     msg->port.port_devid_len);
1265                 port->port_devid->len = msg->port.port_devid_len;
1266                 i += msg->port.port_devid_len;
1267         } else {
1268                 free(port->port_devid, M_CTL);
1269                 port->port_devid = NULL;
1270         }
1271         if (msg->port.target_devid_len != 0) {
1272                 if (port->target_devid == NULL ||
1273                     port->target_devid->len < msg->port.target_devid_len) {
1274                         free(port->target_devid, M_CTL);
1275                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1276                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1277                 }
1278                 memcpy(port->target_devid->data, &msg->port.data[i],
1279                     msg->port.target_devid_len);
1280                 port->target_devid->len = msg->port.target_devid_len;
1281                 i += msg->port.target_devid_len;
1282         } else {
1283                 free(port->target_devid, M_CTL);
1284                 port->target_devid = NULL;
1285         }
1286         if (msg->port.init_devid_len != 0) {
1287                 if (port->init_devid == NULL ||
1288                     port->init_devid->len < msg->port.init_devid_len) {
1289                         free(port->init_devid, M_CTL);
1290                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1291                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1292                 }
1293                 memcpy(port->init_devid->data, &msg->port.data[i],
1294                     msg->port.init_devid_len);
1295                 port->init_devid->len = msg->port.init_devid_len;
1296                 i += msg->port.init_devid_len;
1297         } else {
1298                 free(port->init_devid, M_CTL);
1299                 port->init_devid = NULL;
1300         }
1301         if (new) {
1302                 if (ctl_port_register(port) != 0) {
1303                         printf("%s: ctl_port_register() failed with error\n",
1304                             __func__);
1305                 }
1306         }
1307         mtx_lock(&softc->ctl_lock);
1308         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1309                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
1310                         continue;
1311                 mtx_lock(&lun->lun_lock);
1312                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1313                 mtx_unlock(&lun->lun_lock);
1314         }
1315         mtx_unlock(&softc->ctl_lock);
1316 }
1317
1318 static void
1319 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1320 {
1321         struct ctl_port *port;
1322         int iid;
1323
1324         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1325         if (port == NULL) {
1326                 printf("%s: Received IID for unknown port %d\n",
1327                     __func__, msg->hdr.nexus.targ_port);
1328                 return;
1329         }
1330         iid = msg->hdr.nexus.initid;
1331         if (port->wwpn_iid[iid].in_use != 0 &&
1332             msg->iid.in_use == 0)
1333                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
1334         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1335         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1336         free(port->wwpn_iid[iid].name, M_CTL);
1337         if (msg->iid.name_len) {
1338                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1339                     msg->iid.name_len, M_CTL);
1340         } else
1341                 port->wwpn_iid[iid].name = NULL;
1342 }
1343
1344 static void
1345 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1346 {
1347
1348         if (msg->login.version != CTL_HA_VERSION) {
1349                 printf("CTL HA peers have different versions %d != %d\n",
1350                     msg->login.version, CTL_HA_VERSION);
1351                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1352                 return;
1353         }
1354         if (msg->login.ha_mode != softc->ha_mode) {
1355                 printf("CTL HA peers have different ha_mode %d != %d\n",
1356                     msg->login.ha_mode, softc->ha_mode);
1357                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1358                 return;
1359         }
1360         if (msg->login.ha_id == softc->ha_id) {
1361                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1362                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1363                 return;
1364         }
1365         if (msg->login.max_luns != ctl_max_luns ||
1366             msg->login.max_ports != ctl_max_ports ||
1367             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1368                 printf("CTL HA peers have different limits\n");
1369                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1370                 return;
1371         }
1372 }
1373
1374 static void
1375 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1376 {
1377         struct ctl_lun *lun;
1378         u_int i;
1379         uint32_t initidx, targ_lun;
1380
1381         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1382         mtx_lock(&softc->ctl_lock);
1383         if (targ_lun >= ctl_max_luns ||
1384             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1385                 mtx_unlock(&softc->ctl_lock);
1386                 return;
1387         }
1388         mtx_lock(&lun->lun_lock);
1389         mtx_unlock(&softc->ctl_lock);
1390         if (lun->flags & CTL_LUN_DISABLED) {
1391                 mtx_unlock(&lun->lun_lock);
1392                 return;
1393         }
1394         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1395                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1396                     msg->mode.page_code &&
1397                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1398                         break;
1399         }
1400         if (i == CTL_NUM_MODE_PAGES) {
1401                 mtx_unlock(&lun->lun_lock);
1402                 return;
1403         }
1404         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1405             lun->mode_pages.index[i].page_len);
1406         initidx = ctl_get_initindex(&msg->hdr.nexus);
1407         if (initidx != -1)
1408                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1409         mtx_unlock(&lun->lun_lock);
1410 }
1411
1412 /*
1413  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1414  * subsystem come in here.
1415  */
1416 static void
1417 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1418 {
1419         struct ctl_softc *softc = control_softc;
1420         union ctl_io *io;
1421         struct ctl_prio *presio;
1422         ctl_ha_status isc_status;
1423
1424         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1425         if (event == CTL_HA_EVT_MSG_RECV) {
1426                 union ctl_ha_msg *msg, msgbuf;
1427
1428                 if (param > sizeof(msgbuf))
1429                         msg = malloc(param, M_CTL, M_WAITOK);
1430                 else
1431                         msg = &msgbuf;
1432                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1433                     M_WAITOK);
1434                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1435                         printf("%s: Error receiving message: %d\n",
1436                             __func__, isc_status);
1437                         if (msg != &msgbuf)
1438                                 free(msg, M_CTL);
1439                         return;
1440                 }
1441
1442                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->hdr.msg_type));
1443                 switch (msg->hdr.msg_type) {
1444                 case CTL_MSG_SERIALIZE:
1445                         io = ctl_alloc_io(softc->othersc_pool);
1446                         ctl_zero_io(io);
1447                         // populate ctsio from msg
1448                         io->io_hdr.io_type = CTL_IO_SCSI;
1449                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1450                         io->io_hdr.remote_io = msg->hdr.original_sc;
1451                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1452                                             CTL_FLAG_IO_ACTIVE;
1453                         /*
1454                          * If we're in serialization-only mode, we don't
1455                          * want to go through full done processing.  Thus
1456                          * the COPY flag.
1457                          *
1458                          * XXX KDM add another flag that is more specific.
1459                          */
1460                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1461                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1462                         io->io_hdr.nexus = msg->hdr.nexus;
1463                         io->scsiio.tag_num = msg->scsi.tag_num;
1464                         io->scsiio.tag_type = msg->scsi.tag_type;
1465 #ifdef CTL_TIME_IO
1466                         io->io_hdr.start_time = time_uptime;
1467                         getbinuptime(&io->io_hdr.start_bt);
1468 #endif /* CTL_TIME_IO */
1469                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1470                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1471                                CTL_MAX_CDBLEN);
1472                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1473                                 const struct ctl_cmd_entry *entry;
1474
1475                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1476                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1477                                 io->io_hdr.flags |=
1478                                         entry->flags & CTL_FLAG_DATA_MASK;
1479                         }
1480                         ctl_enqueue_isc(io);
1481                         break;
1482
1483                 /* Performed on the Originating SC, XFER mode only */
1484                 case CTL_MSG_DATAMOVE: {
1485                         struct ctl_sg_entry *sgl;
1486                         int i, j;
1487
1488                         io = msg->hdr.original_sc;
1489                         if (io == NULL) {
1490                                 printf("%s: original_sc == NULL!\n", __func__);
1491                                 /* XXX KDM do something here */
1492                                 break;
1493                         }
1494                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1495                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1496                         /*
1497                          * Keep track of this, we need to send it back over
1498                          * when the datamove is complete.
1499                          */
1500                         io->io_hdr.remote_io = msg->hdr.serializing_sc;
1501                         if (msg->hdr.status == CTL_SUCCESS)
1502                                 io->io_hdr.status = msg->hdr.status;
1503
1504                         if (msg->dt.sg_sequence == 0) {
1505 #ifdef CTL_TIME_IO
1506                                 getbinuptime(&io->io_hdr.dma_start_bt);
1507 #endif
1508                                 i = msg->dt.kern_sg_entries +
1509                                     msg->dt.kern_data_len /
1510                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1511                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1512                                     M_WAITOK | M_ZERO);
1513                                 CTL_RSGL(io) = sgl;
1514                                 CTL_LSGL(io) = &sgl[msg->dt.kern_sg_entries];
1515
1516                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1517
1518                                 io->scsiio.kern_sg_entries =
1519                                         msg->dt.kern_sg_entries;
1520                                 io->scsiio.rem_sg_entries =
1521                                         msg->dt.kern_sg_entries;
1522                                 io->scsiio.kern_data_len =
1523                                         msg->dt.kern_data_len;
1524                                 io->scsiio.kern_total_len =
1525                                         msg->dt.kern_total_len;
1526                                 io->scsiio.kern_data_resid =
1527                                         msg->dt.kern_data_resid;
1528                                 io->scsiio.kern_rel_offset =
1529                                         msg->dt.kern_rel_offset;
1530                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1531                                 io->io_hdr.flags |= msg->dt.flags &
1532                                     CTL_FLAG_BUS_ADDR;
1533                         } else
1534                                 sgl = (struct ctl_sg_entry *)
1535                                         io->scsiio.kern_data_ptr;
1536
1537                         for (i = msg->dt.sent_sg_entries, j = 0;
1538                              i < (msg->dt.sent_sg_entries +
1539                              msg->dt.cur_sg_entries); i++, j++) {
1540                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1541                                 sgl[i].len = msg->dt.sg_list[j].len;
1542                         }
1543
1544                         /*
1545                          * If this is the last piece of the I/O, we've got
1546                          * the full S/G list.  Queue processing in the thread.
1547                          * Otherwise wait for the next piece.
1548                          */
1549                         if (msg->dt.sg_last != 0)
1550                                 ctl_enqueue_isc(io);
1551                         break;
1552                 }
1553                 /* Performed on the Serializing (primary) SC, XFER mode only */
1554                 case CTL_MSG_DATAMOVE_DONE: {
1555                         if (msg->hdr.serializing_sc == NULL) {
1556                                 printf("%s: serializing_sc == NULL!\n",
1557                                        __func__);
1558                                 /* XXX KDM now what? */
1559                                 break;
1560                         }
1561                         /*
1562                          * We grab the sense information here in case
1563                          * there was a failure, so we can return status
1564                          * back to the initiator.
1565                          */
1566                         io = msg->hdr.serializing_sc;
1567                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1568                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1569                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1570                         io->io_hdr.port_status = msg->scsi.port_status;
1571                         io->scsiio.kern_data_resid = msg->scsi.kern_data_resid;
1572                         if (msg->hdr.status != CTL_STATUS_NONE) {
1573                                 io->io_hdr.status = msg->hdr.status;
1574                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1575                                 io->scsiio.sense_len = msg->scsi.sense_len;
1576                                 memcpy(&io->scsiio.sense_data,
1577                                     &msg->scsi.sense_data,
1578                                     msg->scsi.sense_len);
1579                                 if (msg->hdr.status == CTL_SUCCESS)
1580                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1581                         }
1582                         ctl_enqueue_isc(io);
1583                         break;
1584                 }
1585
1586                 /* Preformed on Originating SC, SER_ONLY mode */
1587                 case CTL_MSG_R2R:
1588                         io = msg->hdr.original_sc;
1589                         if (io == NULL) {
1590                                 printf("%s: original_sc == NULL!\n",
1591                                     __func__);
1592                                 break;
1593                         }
1594                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1595                         io->io_hdr.msg_type = CTL_MSG_R2R;
1596                         io->io_hdr.remote_io = msg->hdr.serializing_sc;
1597                         ctl_enqueue_isc(io);
1598                         break;
1599
1600                 /*
1601                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1602                  * mode.
1603                  * Performed on the Originating (i.e. secondary) SC in XFER
1604                  * mode
1605                  */
1606                 case CTL_MSG_FINISH_IO:
1607                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1608                                 ctl_isc_handler_finish_xfer(softc, msg);
1609                         else
1610                                 ctl_isc_handler_finish_ser_only(softc, msg);
1611                         break;
1612
1613                 /* Preformed on Originating SC */
1614                 case CTL_MSG_BAD_JUJU:
1615                         io = msg->hdr.original_sc;
1616                         if (io == NULL) {
1617                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1618                                        __func__);
1619                                 break;
1620                         }
1621                         ctl_copy_sense_data(msg, io);
1622                         /*
1623                          * IO should have already been cleaned up on other
1624                          * SC so clear this flag so we won't send a message
1625                          * back to finish the IO there.
1626                          */
1627                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1628                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1629
1630                         /* io = msg->hdr.serializing_sc; */
1631                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1632                         ctl_enqueue_isc(io);
1633                         break;
1634
1635                 /* Handle resets sent from the other side */
1636                 case CTL_MSG_MANAGE_TASKS: {
1637                         struct ctl_taskio *taskio;
1638                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1639                             softc->othersc_pool);
1640                         ctl_zero_io((union ctl_io *)taskio);
1641                         taskio->io_hdr.io_type = CTL_IO_TASK;
1642                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1643                         taskio->io_hdr.nexus = msg->hdr.nexus;
1644                         taskio->task_action = msg->task.task_action;
1645                         taskio->tag_num = msg->task.tag_num;
1646                         taskio->tag_type = msg->task.tag_type;
1647 #ifdef CTL_TIME_IO
1648                         taskio->io_hdr.start_time = time_uptime;
1649                         getbinuptime(&taskio->io_hdr.start_bt);
1650 #endif /* CTL_TIME_IO */
1651                         ctl_run_task((union ctl_io *)taskio);
1652                         break;
1653                 }
1654                 /* Persistent Reserve action which needs attention */
1655                 case CTL_MSG_PERS_ACTION:
1656                         presio = (struct ctl_prio *)ctl_alloc_io(
1657                             softc->othersc_pool);
1658                         ctl_zero_io((union ctl_io *)presio);
1659                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1660                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1661                         presio->io_hdr.nexus = msg->hdr.nexus;
1662                         presio->pr_msg = msg->pr;
1663                         ctl_enqueue_isc((union ctl_io *)presio);
1664                         break;
1665                 case CTL_MSG_UA:
1666                         ctl_isc_ua(softc, msg, param);
1667                         break;
1668                 case CTL_MSG_PORT_SYNC:
1669                         ctl_isc_port_sync(softc, msg, param);
1670                         break;
1671                 case CTL_MSG_LUN_SYNC:
1672                         ctl_isc_lun_sync(softc, msg, param);
1673                         break;
1674                 case CTL_MSG_IID_SYNC:
1675                         ctl_isc_iid_sync(softc, msg, param);
1676                         break;
1677                 case CTL_MSG_LOGIN:
1678                         ctl_isc_login(softc, msg, param);
1679                         break;
1680                 case CTL_MSG_MODE_SYNC:
1681                         ctl_isc_mode_sync(softc, msg, param);
1682                         break;
1683                 default:
1684                         printf("Received HA message of unknown type %d\n",
1685                             msg->hdr.msg_type);
1686                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1687                         break;
1688                 }
1689                 if (msg != &msgbuf)
1690                         free(msg, M_CTL);
1691         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1692                 printf("CTL: HA link status changed from %d to %d\n",
1693                     softc->ha_link, param);
1694                 if (param == softc->ha_link)
1695                         return;
1696                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1697                         softc->ha_link = param;
1698                         ctl_isc_ha_link_down(softc);
1699                 } else {
1700                         softc->ha_link = param;
1701                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1702                                 ctl_isc_ha_link_up(softc);
1703                 }
1704                 return;
1705         } else {
1706                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1707                 return;
1708         }
1709 }
1710
1711 static void
1712 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1713 {
1714
1715         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1716             src->scsi.sense_len);
1717         dest->scsiio.scsi_status = src->scsi.scsi_status;
1718         dest->scsiio.sense_len = src->scsi.sense_len;
1719         dest->io_hdr.status = src->hdr.status;
1720 }
1721
1722 static void
1723 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1724 {
1725
1726         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1727             src->scsiio.sense_len);
1728         dest->scsi.scsi_status = src->scsiio.scsi_status;
1729         dest->scsi.sense_len = src->scsiio.sense_len;
1730         dest->hdr.status = src->io_hdr.status;
1731 }
1732
1733 void
1734 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1735 {
1736         struct ctl_softc *softc = lun->ctl_softc;
1737         ctl_ua_type *pu;
1738
1739         if (initidx < softc->init_min || initidx >= softc->init_max)
1740                 return;
1741         mtx_assert(&lun->lun_lock, MA_OWNED);
1742         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1743         if (pu == NULL)
1744                 return;
1745         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1746 }
1747
1748 void
1749 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1750 {
1751         int i;
1752
1753         mtx_assert(&lun->lun_lock, MA_OWNED);
1754         if (lun->pending_ua[port] == NULL)
1755                 return;
1756         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1757                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1758                         continue;
1759                 lun->pending_ua[port][i] |= ua;
1760         }
1761 }
1762
1763 void
1764 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1765 {
1766         struct ctl_softc *softc = lun->ctl_softc;
1767         int i;
1768
1769         mtx_assert(&lun->lun_lock, MA_OWNED);
1770         for (i = softc->port_min; i < softc->port_max; i++)
1771                 ctl_est_ua_port(lun, i, except, ua);
1772 }
1773
1774 void
1775 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1776 {
1777         struct ctl_softc *softc = lun->ctl_softc;
1778         ctl_ua_type *pu;
1779
1780         if (initidx < softc->init_min || initidx >= softc->init_max)
1781                 return;
1782         mtx_assert(&lun->lun_lock, MA_OWNED);
1783         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1784         if (pu == NULL)
1785                 return;
1786         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1787 }
1788
1789 void
1790 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1791 {
1792         struct ctl_softc *softc = lun->ctl_softc;
1793         int i, j;
1794
1795         mtx_assert(&lun->lun_lock, MA_OWNED);
1796         for (i = softc->port_min; i < softc->port_max; i++) {
1797                 if (lun->pending_ua[i] == NULL)
1798                         continue;
1799                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1800                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1801                                 continue;
1802                         lun->pending_ua[i][j] &= ~ua;
1803                 }
1804         }
1805 }
1806
1807 void
1808 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1809     ctl_ua_type ua_type)
1810 {
1811         struct ctl_lun *lun;
1812
1813         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1814         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1815                 mtx_lock(&lun->lun_lock);
1816                 ctl_clr_ua(lun, initidx, ua_type);
1817                 mtx_unlock(&lun->lun_lock);
1818         }
1819 }
1820
1821 static int
1822 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1823 {
1824         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1825         struct ctl_lun *lun;
1826         struct ctl_lun_req ireq;
1827         int error, value;
1828
1829         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1830         error = sysctl_handle_int(oidp, &value, 0, req);
1831         if ((error != 0) || (req->newptr == NULL))
1832                 return (error);
1833
1834         mtx_lock(&softc->ctl_lock);
1835         if (value == 0)
1836                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1837         else
1838                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1839         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1840                 mtx_unlock(&softc->ctl_lock);
1841                 bzero(&ireq, sizeof(ireq));
1842                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1843                 ireq.reqdata.modify.lun_id = lun->lun;
1844                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1845                     curthread);
1846                 if (ireq.status != CTL_LUN_OK) {
1847                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1848                             __func__, ireq.status, ireq.error_str);
1849                 }
1850                 mtx_lock(&softc->ctl_lock);
1851         }
1852         mtx_unlock(&softc->ctl_lock);
1853         return (0);
1854 }
1855
1856 static int
1857 ctl_init(void)
1858 {
1859         struct make_dev_args args;
1860         struct ctl_softc *softc;
1861         int i, error;
1862
1863         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1864                                M_WAITOK | M_ZERO);
1865
1866         make_dev_args_init(&args);
1867         args.mda_devsw = &ctl_cdevsw;
1868         args.mda_uid = UID_ROOT;
1869         args.mda_gid = GID_OPERATOR;
1870         args.mda_mode = 0600;
1871         args.mda_si_drv1 = softc;
1872         args.mda_si_drv2 = NULL;
1873         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1874         if (error != 0) {
1875                 free(softc, M_DEVBUF);
1876                 control_softc = NULL;
1877                 return (error);
1878         }
1879
1880         sysctl_ctx_init(&softc->sysctl_ctx);
1881         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1882                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1883                 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "CAM Target Layer");
1884
1885         if (softc->sysctl_tree == NULL) {
1886                 printf("%s: unable to allocate sysctl tree\n", __func__);
1887                 destroy_dev(softc->dev);
1888                 free(softc, M_DEVBUF);
1889                 control_softc = NULL;
1890                 return (ENOMEM);
1891         }
1892
1893         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1894         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1895             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1896         softc->flags = 0;
1897
1898         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1899             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1900             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1901
1902         if (ctl_max_luns <= 0 || powerof2(ctl_max_luns) == 0) {
1903                 printf("Bad value %d for kern.cam.ctl.max_luns, must be a power of two, using %d\n",
1904                     ctl_max_luns, CTL_DEFAULT_MAX_LUNS);
1905                 ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
1906         }
1907         softc->ctl_luns = malloc(sizeof(struct ctl_lun *) * ctl_max_luns,
1908             M_DEVBUF, M_WAITOK | M_ZERO);
1909         softc->ctl_lun_mask = malloc(sizeof(uint32_t) *
1910             ((ctl_max_luns + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1911         if (ctl_max_ports <= 0 || powerof2(ctl_max_ports) == 0) {
1912                 printf("Bad value %d for kern.cam.ctl.max_ports, must be a power of two, using %d\n",
1913                     ctl_max_ports, CTL_DEFAULT_MAX_PORTS);
1914                 ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
1915         }
1916         softc->ctl_port_mask = malloc(sizeof(uint32_t) *
1917           ((ctl_max_ports + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1918         softc->ctl_ports = malloc(sizeof(struct ctl_port *) * ctl_max_ports,
1919              M_DEVBUF, M_WAITOK | M_ZERO);
1920
1921
1922         /*
1923          * In Copan's HA scheme, the "master" and "slave" roles are
1924          * figured out through the slot the controller is in.  Although it
1925          * is an active/active system, someone has to be in charge.
1926          */
1927         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1928             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1929             "HA head ID (0 - no HA)");
1930         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1931                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1932                 softc->is_single = 1;
1933                 softc->port_cnt = ctl_max_ports;
1934                 softc->port_min = 0;
1935         } else {
1936                 softc->port_cnt = ctl_max_ports / NUM_HA_SHELVES;
1937                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1938         }
1939         softc->port_max = softc->port_min + softc->port_cnt;
1940         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1941         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1942
1943         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1944             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1945             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1946
1947         STAILQ_INIT(&softc->lun_list);
1948         STAILQ_INIT(&softc->pending_lun_queue);
1949         STAILQ_INIT(&softc->fe_list);
1950         STAILQ_INIT(&softc->port_list);
1951         STAILQ_INIT(&softc->be_list);
1952         ctl_tpc_init(softc);
1953
1954         if (worker_threads <= 0)
1955                 worker_threads = max(1, mp_ncpus / 4);
1956         if (worker_threads > CTL_MAX_THREADS)
1957                 worker_threads = CTL_MAX_THREADS;
1958
1959         for (i = 0; i < worker_threads; i++) {
1960                 struct ctl_thread *thr = &softc->threads[i];
1961
1962                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1963                 thr->ctl_softc = softc;
1964                 STAILQ_INIT(&thr->incoming_queue);
1965                 STAILQ_INIT(&thr->rtr_queue);
1966                 STAILQ_INIT(&thr->done_queue);
1967                 STAILQ_INIT(&thr->isc_queue);
1968
1969                 error = kproc_kthread_add(ctl_work_thread, thr,
1970                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1971                 if (error != 0) {
1972                         printf("error creating CTL work thread!\n");
1973                         return (error);
1974                 }
1975         }
1976         error = kproc_kthread_add(ctl_lun_thread, softc,
1977             &softc->ctl_proc, &softc->lun_thread, 0, 0, "ctl", "lun");
1978         if (error != 0) {
1979                 printf("error creating CTL lun thread!\n");
1980                 return (error);
1981         }
1982         error = kproc_kthread_add(ctl_thresh_thread, softc,
1983             &softc->ctl_proc, &softc->thresh_thread, 0, 0, "ctl", "thresh");
1984         if (error != 0) {
1985                 printf("error creating CTL threshold thread!\n");
1986                 return (error);
1987         }
1988
1989         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1990             OID_AUTO, "ha_role",
1991             CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT,
1992             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1993
1994         if (softc->is_single == 0) {
1995                 if (ctl_frontend_register(&ha_frontend) != 0)
1996                         softc->is_single = 1;
1997         }
1998         return (0);
1999 }
2000
2001 static int
2002 ctl_shutdown(void)
2003 {
2004         struct ctl_softc *softc = control_softc;
2005         int i;
2006
2007         if (softc->is_single == 0)
2008                 ctl_frontend_deregister(&ha_frontend);
2009
2010         destroy_dev(softc->dev);
2011
2012         /* Shutdown CTL threads. */
2013         softc->shutdown = 1;
2014         for (i = 0; i < worker_threads; i++) {
2015                 struct ctl_thread *thr = &softc->threads[i];
2016                 while (thr->thread != NULL) {
2017                         wakeup(thr);
2018                         if (thr->thread != NULL)
2019                                 pause("CTL thr shutdown", 1);
2020                 }
2021                 mtx_destroy(&thr->queue_lock);
2022         }
2023         while (softc->lun_thread != NULL) {
2024                 wakeup(&softc->pending_lun_queue);
2025                 if (softc->lun_thread != NULL)
2026                         pause("CTL thr shutdown", 1);
2027         }
2028         while (softc->thresh_thread != NULL) {
2029                 wakeup(softc->thresh_thread);
2030                 if (softc->thresh_thread != NULL)
2031                         pause("CTL thr shutdown", 1);
2032         }
2033
2034         ctl_tpc_shutdown(softc);
2035         uma_zdestroy(softc->io_zone);
2036         mtx_destroy(&softc->ctl_lock);
2037
2038         free(softc->ctl_luns, M_DEVBUF);
2039         free(softc->ctl_lun_mask, M_DEVBUF);
2040         free(softc->ctl_port_mask, M_DEVBUF);
2041         free(softc->ctl_ports, M_DEVBUF);
2042
2043         sysctl_ctx_free(&softc->sysctl_ctx);
2044
2045         free(softc, M_DEVBUF);
2046         control_softc = NULL;
2047         return (0);
2048 }
2049
2050 static int
2051 ctl_module_event_handler(module_t mod, int what, void *arg)
2052 {
2053
2054         switch (what) {
2055         case MOD_LOAD:
2056                 return (ctl_init());
2057         case MOD_UNLOAD:
2058                 return (ctl_shutdown());
2059         default:
2060                 return (EOPNOTSUPP);
2061         }
2062 }
2063
2064 /*
2065  * XXX KDM should we do some access checks here?  Bump a reference count to
2066  * prevent a CTL module from being unloaded while someone has it open?
2067  */
2068 static int
2069 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
2070 {
2071         return (0);
2072 }
2073
2074 static int
2075 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2076 {
2077         return (0);
2078 }
2079
2080 /*
2081  * Remove an initiator by port number and initiator ID.
2082  * Returns 0 for success, -1 for failure.
2083  */
2084 int
2085 ctl_remove_initiator(struct ctl_port *port, int iid)
2086 {
2087         struct ctl_softc *softc = port->ctl_softc;
2088         int last;
2089
2090         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2091
2092         if (iid > CTL_MAX_INIT_PER_PORT) {
2093                 printf("%s: initiator ID %u > maximun %u!\n",
2094                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2095                 return (-1);
2096         }
2097
2098         mtx_lock(&softc->ctl_lock);
2099         last = (--port->wwpn_iid[iid].in_use == 0);
2100         port->wwpn_iid[iid].last_use = time_uptime;
2101         mtx_unlock(&softc->ctl_lock);
2102         if (last)
2103                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
2104         ctl_isc_announce_iid(port, iid);
2105
2106         return (0);
2107 }
2108
2109 /*
2110  * Add an initiator to the initiator map.
2111  * Returns iid for success, < 0 for failure.
2112  */
2113 int
2114 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2115 {
2116         struct ctl_softc *softc = port->ctl_softc;
2117         time_t best_time;
2118         int i, best;
2119
2120         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2121
2122         if (iid >= CTL_MAX_INIT_PER_PORT) {
2123                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2124                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2125                 free(name, M_CTL);
2126                 return (-1);
2127         }
2128
2129         mtx_lock(&softc->ctl_lock);
2130
2131         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2132                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2133                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2134                                 iid = i;
2135                                 break;
2136                         }
2137                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2138                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2139                                 iid = i;
2140                                 break;
2141                         }
2142                 }
2143         }
2144
2145         if (iid < 0) {
2146                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2147                         if (port->wwpn_iid[i].in_use == 0 &&
2148                             port->wwpn_iid[i].wwpn == 0 &&
2149                             port->wwpn_iid[i].name == NULL) {
2150                                 iid = i;
2151                                 break;
2152                         }
2153                 }
2154         }
2155
2156         if (iid < 0) {
2157                 best = -1;
2158                 best_time = INT32_MAX;
2159                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2160                         if (port->wwpn_iid[i].in_use == 0) {
2161                                 if (port->wwpn_iid[i].last_use < best_time) {
2162                                         best = i;
2163                                         best_time = port->wwpn_iid[i].last_use;
2164                                 }
2165                         }
2166                 }
2167                 iid = best;
2168         }
2169
2170         if (iid < 0) {
2171                 mtx_unlock(&softc->ctl_lock);
2172                 free(name, M_CTL);
2173                 return (-2);
2174         }
2175
2176         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2177                 /*
2178                  * This is not an error yet.
2179                  */
2180                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2181 #if 0
2182                         printf("%s: port %d iid %u WWPN %#jx arrived"
2183                             " again\n", __func__, port->targ_port,
2184                             iid, (uintmax_t)wwpn);
2185 #endif
2186                         goto take;
2187                 }
2188                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2189                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2190 #if 0
2191                         printf("%s: port %d iid %u name '%s' arrived"
2192                             " again\n", __func__, port->targ_port,
2193                             iid, name);
2194 #endif
2195                         goto take;
2196                 }
2197
2198                 /*
2199                  * This is an error, but what do we do about it?  The
2200                  * driver is telling us we have a new WWPN for this
2201                  * initiator ID, so we pretty much need to use it.
2202                  */
2203                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2204                     " but WWPN %#jx '%s' is still at that address\n",
2205                     __func__, port->targ_port, iid, wwpn, name,
2206                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2207                     port->wwpn_iid[iid].name);
2208         }
2209 take:
2210         free(port->wwpn_iid[iid].name, M_CTL);
2211         port->wwpn_iid[iid].name = name;
2212         port->wwpn_iid[iid].wwpn = wwpn;
2213         port->wwpn_iid[iid].in_use++;
2214         mtx_unlock(&softc->ctl_lock);
2215         ctl_isc_announce_iid(port, iid);
2216
2217         return (iid);
2218 }
2219
2220 static int
2221 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2222 {
2223         int len;
2224
2225         switch (port->port_type) {
2226         case CTL_PORT_FC:
2227         {
2228                 struct scsi_transportid_fcp *id =
2229                     (struct scsi_transportid_fcp *)buf;
2230                 if (port->wwpn_iid[iid].wwpn == 0)
2231                         return (0);
2232                 memset(id, 0, sizeof(*id));
2233                 id->format_protocol = SCSI_PROTO_FC;
2234                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2235                 return (sizeof(*id));
2236         }
2237         case CTL_PORT_ISCSI:
2238         {
2239                 struct scsi_transportid_iscsi_port *id =
2240                     (struct scsi_transportid_iscsi_port *)buf;
2241                 if (port->wwpn_iid[iid].name == NULL)
2242                         return (0);
2243                 memset(id, 0, 256);
2244                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2245                     SCSI_PROTO_ISCSI;
2246                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2247                 len = roundup2(min(len, 252), 4);
2248                 scsi_ulto2b(len, id->additional_length);
2249                 return (sizeof(*id) + len);
2250         }
2251         case CTL_PORT_SAS:
2252         {
2253                 struct scsi_transportid_sas *id =
2254                     (struct scsi_transportid_sas *)buf;
2255                 if (port->wwpn_iid[iid].wwpn == 0)
2256                         return (0);
2257                 memset(id, 0, sizeof(*id));
2258                 id->format_protocol = SCSI_PROTO_SAS;
2259                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2260                 return (sizeof(*id));
2261         }
2262         default:
2263         {
2264                 struct scsi_transportid_spi *id =
2265                     (struct scsi_transportid_spi *)buf;
2266                 memset(id, 0, sizeof(*id));
2267                 id->format_protocol = SCSI_PROTO_SPI;
2268                 scsi_ulto2b(iid, id->scsi_addr);
2269                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2270                 return (sizeof(*id));
2271         }
2272         }
2273 }
2274
2275 /*
2276  * Serialize a command that went down the "wrong" side, and so was sent to
2277  * this controller for execution.  The logic is a little different than the
2278  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2279  * sent back to the other side, but in the success case, we execute the
2280  * command on this side (XFER mode) or tell the other side to execute it
2281  * (SER_ONLY mode).
2282  */
2283 static void
2284 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2285 {
2286         struct ctl_softc *softc = CTL_SOFTC(ctsio);
2287         struct ctl_port *port = CTL_PORT(ctsio);
2288         union ctl_ha_msg msg_info;
2289         struct ctl_lun *lun;
2290         const struct ctl_cmd_entry *entry;
2291         union ctl_io *bio;
2292         uint32_t targ_lun;
2293
2294         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2295
2296         /* Make sure that we know about this port. */
2297         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2298                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2299                                          /*retry_count*/ 1);
2300                 goto badjuju;
2301         }
2302
2303         /* Make sure that we know about this LUN. */
2304         mtx_lock(&softc->ctl_lock);
2305         if (targ_lun >= ctl_max_luns ||
2306             (lun = softc->ctl_luns[targ_lun]) == NULL) {
2307                 mtx_unlock(&softc->ctl_lock);
2308
2309                 /*
2310                  * The other node would not send this request to us unless
2311                  * received announce that we are primary node for this LUN.
2312                  * If this LUN does not exist now, it is probably result of
2313                  * a race, so respond to initiator in the most opaque way.
2314                  */
2315                 ctl_set_busy(ctsio);
2316                 goto badjuju;
2317         }
2318         mtx_lock(&lun->lun_lock);
2319         mtx_unlock(&softc->ctl_lock);
2320
2321         /*
2322          * If the LUN is invalid, pretend that it doesn't exist.
2323          * It will go away as soon as all pending I/Os completed.
2324          */
2325         if (lun->flags & CTL_LUN_DISABLED) {
2326                 mtx_unlock(&lun->lun_lock);
2327                 ctl_set_busy(ctsio);
2328                 goto badjuju;
2329         }
2330
2331         entry = ctl_get_cmd_entry(ctsio, NULL);
2332         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2333                 mtx_unlock(&lun->lun_lock);
2334                 goto badjuju;
2335         }
2336
2337         CTL_LUN(ctsio) = lun;
2338         CTL_BACKEND_LUN(ctsio) = lun->be_lun;
2339
2340         /*
2341          * Every I/O goes into the OOA queue for a
2342          * particular LUN, and stays there until completion.
2343          */
2344 #ifdef CTL_TIME_IO
2345         if (TAILQ_EMPTY(&lun->ooa_queue))
2346                 lun->idle_time += getsbinuptime() - lun->last_busy;
2347 #endif
2348         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2349
2350         bio = (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq, ooa_links);
2351         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio, &bio)) {
2352         case CTL_ACTION_BLOCK:
2353                 ctsio->io_hdr.blocker = bio;
2354                 TAILQ_INSERT_TAIL(&bio->io_hdr.blocked_queue, &ctsio->io_hdr,
2355                                   blocked_links);
2356                 mtx_unlock(&lun->lun_lock);
2357                 break;
2358         case CTL_ACTION_PASS:
2359         case CTL_ACTION_SKIP:
2360                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2361                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2362                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2363                         mtx_unlock(&lun->lun_lock);
2364                 } else {
2365                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2366                         mtx_unlock(&lun->lun_lock);
2367
2368                         /* send msg back to other side */
2369                         msg_info.hdr.original_sc = ctsio->io_hdr.remote_io;
2370                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2371                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2372                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2373                             sizeof(msg_info.hdr), M_WAITOK);
2374                 }
2375                 break;
2376         case CTL_ACTION_OVERLAP:
2377                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2378                 mtx_unlock(&lun->lun_lock);
2379                 ctl_set_overlapped_cmd(ctsio);
2380                 goto badjuju;
2381         case CTL_ACTION_OVERLAP_TAG:
2382                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2383                 mtx_unlock(&lun->lun_lock);
2384                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2385                 goto badjuju;
2386         case CTL_ACTION_ERROR:
2387         default:
2388                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2389                 mtx_unlock(&lun->lun_lock);
2390
2391                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2392                                          /*retry_count*/ 0);
2393 badjuju:
2394                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2395                 msg_info.hdr.original_sc = ctsio->io_hdr.remote_io;
2396                 msg_info.hdr.serializing_sc = NULL;
2397                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2398                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2399                     sizeof(msg_info.scsi), M_WAITOK);
2400                 ctl_free_io((union ctl_io *)ctsio);
2401                 break;
2402         }
2403 }
2404
2405 /*
2406  * Returns 0 for success, errno for failure.
2407  */
2408 static void
2409 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2410                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2411 {
2412         union ctl_io *io;
2413
2414         mtx_lock(&lun->lun_lock);
2415         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2416              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2417              ooa_links)) {
2418                 struct ctl_ooa_entry *entry;
2419
2420                 /*
2421                  * If we've got more than we can fit, just count the
2422                  * remaining entries.
2423                  */
2424                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2425                         continue;
2426
2427                 entry = &kern_entries[*cur_fill_num];
2428
2429                 entry->tag_num = io->scsiio.tag_num;
2430                 entry->lun_num = lun->lun;
2431 #ifdef CTL_TIME_IO
2432                 entry->start_bt = io->io_hdr.start_bt;
2433 #endif
2434                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2435                 entry->cdb_len = io->scsiio.cdb_len;
2436                 if (io->io_hdr.blocker != NULL)
2437                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2438
2439                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2440                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2441
2442                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2443                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2444
2445                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2446                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2447
2448                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2449                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2450         }
2451         mtx_unlock(&lun->lun_lock);
2452 }
2453
2454 /*
2455  * Escape characters that are illegal or not recommended in XML.
2456  */
2457 int
2458 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2459 {
2460         char *end = str + size;
2461         int retval;
2462
2463         retval = 0;
2464
2465         for (; *str && str < end; str++) {
2466                 switch (*str) {
2467                 case '&':
2468                         retval = sbuf_printf(sb, "&amp;");
2469                         break;
2470                 case '>':
2471                         retval = sbuf_printf(sb, "&gt;");
2472                         break;
2473                 case '<':
2474                         retval = sbuf_printf(sb, "&lt;");
2475                         break;
2476                 default:
2477                         retval = sbuf_putc(sb, *str);
2478                         break;
2479                 }
2480
2481                 if (retval != 0)
2482                         break;
2483
2484         }
2485
2486         return (retval);
2487 }
2488
2489 static void
2490 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2491 {
2492         struct scsi_vpd_id_descriptor *desc;
2493         int i;
2494
2495         if (id == NULL || id->len < 4)
2496                 return;
2497         desc = (struct scsi_vpd_id_descriptor *)id->data;
2498         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2499         case SVPD_ID_TYPE_T10:
2500                 sbuf_printf(sb, "t10.");
2501                 break;
2502         case SVPD_ID_TYPE_EUI64:
2503                 sbuf_printf(sb, "eui.");
2504                 break;
2505         case SVPD_ID_TYPE_NAA:
2506                 sbuf_printf(sb, "naa.");
2507                 break;
2508         case SVPD_ID_TYPE_SCSI_NAME:
2509                 break;
2510         }
2511         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2512         case SVPD_ID_CODESET_BINARY:
2513                 for (i = 0; i < desc->length; i++)
2514                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2515                 break;
2516         case SVPD_ID_CODESET_ASCII:
2517                 sbuf_printf(sb, "%.*s", (int)desc->length,
2518                     (char *)desc->identifier);
2519                 break;
2520         case SVPD_ID_CODESET_UTF8:
2521                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2522                 break;
2523         }
2524 }
2525
2526 static int
2527 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2528           struct thread *td)
2529 {
2530         struct ctl_softc *softc = dev->si_drv1;
2531         struct ctl_port *port;
2532         struct ctl_lun *lun;
2533         int retval;
2534
2535         retval = 0;
2536
2537         switch (cmd) {
2538         case CTL_IO:
2539                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2540                 break;
2541         case CTL_ENABLE_PORT:
2542         case CTL_DISABLE_PORT:
2543         case CTL_SET_PORT_WWNS: {
2544                 struct ctl_port *port;
2545                 struct ctl_port_entry *entry;
2546
2547                 entry = (struct ctl_port_entry *)addr;
2548
2549                 mtx_lock(&softc->ctl_lock);
2550                 STAILQ_FOREACH(port, &softc->port_list, links) {
2551                         int action, done;
2552
2553                         if (port->targ_port < softc->port_min ||
2554                             port->targ_port >= softc->port_max)
2555                                 continue;
2556
2557                         action = 0;
2558                         done = 0;
2559                         if ((entry->port_type == CTL_PORT_NONE)
2560                          && (entry->targ_port == port->targ_port)) {
2561                                 /*
2562                                  * If the user only wants to enable or
2563                                  * disable or set WWNs on a specific port,
2564                                  * do the operation and we're done.
2565                                  */
2566                                 action = 1;
2567                                 done = 1;
2568                         } else if (entry->port_type & port->port_type) {
2569                                 /*
2570                                  * Compare the user's type mask with the
2571                                  * particular frontend type to see if we
2572                                  * have a match.
2573                                  */
2574                                 action = 1;
2575                                 done = 0;
2576
2577                                 /*
2578                                  * Make sure the user isn't trying to set
2579                                  * WWNs on multiple ports at the same time.
2580                                  */
2581                                 if (cmd == CTL_SET_PORT_WWNS) {
2582                                         printf("%s: Can't set WWNs on "
2583                                                "multiple ports\n", __func__);
2584                                         retval = EINVAL;
2585                                         break;
2586                                 }
2587                         }
2588                         if (action == 0)
2589                                 continue;
2590
2591                         /*
2592                          * XXX KDM we have to drop the lock here, because
2593                          * the online/offline operations can potentially
2594                          * block.  We need to reference count the frontends
2595                          * so they can't go away,
2596                          */
2597                         if (cmd == CTL_ENABLE_PORT) {
2598                                 mtx_unlock(&softc->ctl_lock);
2599                                 ctl_port_online(port);
2600                                 mtx_lock(&softc->ctl_lock);
2601                         } else if (cmd == CTL_DISABLE_PORT) {
2602                                 mtx_unlock(&softc->ctl_lock);
2603                                 ctl_port_offline(port);
2604                                 mtx_lock(&softc->ctl_lock);
2605                         } else if (cmd == CTL_SET_PORT_WWNS) {
2606                                 ctl_port_set_wwns(port,
2607                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2608                                     1 : 0, entry->wwnn,
2609                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2610                                     1 : 0, entry->wwpn);
2611                         }
2612                         if (done != 0)
2613                                 break;
2614                 }
2615                 mtx_unlock(&softc->ctl_lock);
2616                 break;
2617         }
2618         case CTL_GET_OOA: {
2619                 struct ctl_ooa *ooa_hdr;
2620                 struct ctl_ooa_entry *entries;
2621                 uint32_t cur_fill_num;
2622
2623                 ooa_hdr = (struct ctl_ooa *)addr;
2624
2625                 if ((ooa_hdr->alloc_len == 0)
2626                  || (ooa_hdr->alloc_num == 0)) {
2627                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2628                                "must be non-zero\n", __func__,
2629                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2630                         retval = EINVAL;
2631                         break;
2632                 }
2633
2634                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2635                     sizeof(struct ctl_ooa_entry))) {
2636                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2637                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2638                                __func__, ooa_hdr->alloc_len,
2639                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2640                         retval = EINVAL;
2641                         break;
2642                 }
2643
2644                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2645                 if (entries == NULL) {
2646                         printf("%s: could not allocate %d bytes for OOA "
2647                                "dump\n", __func__, ooa_hdr->alloc_len);
2648                         retval = ENOMEM;
2649                         break;
2650                 }
2651
2652                 mtx_lock(&softc->ctl_lock);
2653                 if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 &&
2654                     (ooa_hdr->lun_num >= ctl_max_luns ||
2655                      softc->ctl_luns[ooa_hdr->lun_num] == NULL)) {
2656                         mtx_unlock(&softc->ctl_lock);
2657                         free(entries, M_CTL);
2658                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2659                                __func__, (uintmax_t)ooa_hdr->lun_num);
2660                         retval = EINVAL;
2661                         break;
2662                 }
2663
2664                 cur_fill_num = 0;
2665
2666                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2667                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2668                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2669                                     ooa_hdr, entries);
2670                         }
2671                 } else {
2672                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2673                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2674                             entries);
2675                 }
2676                 mtx_unlock(&softc->ctl_lock);
2677
2678                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2679                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2680                         sizeof(struct ctl_ooa_entry);
2681                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2682                 if (retval != 0) {
2683                         printf("%s: error copying out %d bytes for OOA dump\n",
2684                                __func__, ooa_hdr->fill_len);
2685                 }
2686
2687                 getbinuptime(&ooa_hdr->cur_bt);
2688
2689                 if (cur_fill_num > ooa_hdr->alloc_num) {
2690                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2691                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2692                 } else {
2693                         ooa_hdr->dropped_num = 0;
2694                         ooa_hdr->status = CTL_OOA_OK;
2695                 }
2696
2697                 free(entries, M_CTL);
2698                 break;
2699         }
2700         case CTL_DELAY_IO: {
2701                 struct ctl_io_delay_info *delay_info;
2702
2703                 delay_info = (struct ctl_io_delay_info *)addr;
2704
2705 #ifdef CTL_IO_DELAY
2706                 mtx_lock(&softc->ctl_lock);
2707                 if (delay_info->lun_id >= ctl_max_luns ||
2708                     (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) {
2709                         mtx_unlock(&softc->ctl_lock);
2710                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2711                         break;
2712                 }
2713                 mtx_lock(&lun->lun_lock);
2714                 mtx_unlock(&softc->ctl_lock);
2715                 delay_info->status = CTL_DELAY_STATUS_OK;
2716                 switch (delay_info->delay_type) {
2717                 case CTL_DELAY_TYPE_CONT:
2718                 case CTL_DELAY_TYPE_ONESHOT:
2719                         break;
2720                 default:
2721                         delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE;
2722                         break;
2723                 }
2724                 switch (delay_info->delay_loc) {
2725                 case CTL_DELAY_LOC_DATAMOVE:
2726                         lun->delay_info.datamove_type = delay_info->delay_type;
2727                         lun->delay_info.datamove_delay = delay_info->delay_secs;
2728                         break;
2729                 case CTL_DELAY_LOC_DONE:
2730                         lun->delay_info.done_type = delay_info->delay_type;
2731                         lun->delay_info.done_delay = delay_info->delay_secs;
2732                         break;
2733                 default:
2734                         delay_info->status = CTL_DELAY_STATUS_INVALID_LOC;
2735                         break;
2736                 }
2737                 mtx_unlock(&lun->lun_lock);
2738 #else
2739                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2740 #endif /* CTL_IO_DELAY */
2741                 break;
2742         }
2743         case CTL_ERROR_INJECT: {
2744                 struct ctl_error_desc *err_desc, *new_err_desc;
2745
2746                 err_desc = (struct ctl_error_desc *)addr;
2747
2748                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2749                                       M_WAITOK | M_ZERO);
2750                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2751
2752                 mtx_lock(&softc->ctl_lock);
2753                 if (err_desc->lun_id >= ctl_max_luns ||
2754                     (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) {
2755                         mtx_unlock(&softc->ctl_lock);
2756                         free(new_err_desc, M_CTL);
2757                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2758                                __func__, (uintmax_t)err_desc->lun_id);
2759                         retval = EINVAL;
2760                         break;
2761                 }
2762                 mtx_lock(&lun->lun_lock);
2763                 mtx_unlock(&softc->ctl_lock);
2764
2765                 /*
2766                  * We could do some checking here to verify the validity
2767                  * of the request, but given the complexity of error
2768                  * injection requests, the checking logic would be fairly
2769                  * complex.
2770                  *
2771                  * For now, if the request is invalid, it just won't get
2772                  * executed and might get deleted.
2773                  */
2774                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2775
2776                 /*
2777                  * XXX KDM check to make sure the serial number is unique,
2778                  * in case we somehow manage to wrap.  That shouldn't
2779                  * happen for a very long time, but it's the right thing to
2780                  * do.
2781                  */
2782                 new_err_desc->serial = lun->error_serial;
2783                 err_desc->serial = lun->error_serial;
2784                 lun->error_serial++;
2785
2786                 mtx_unlock(&lun->lun_lock);
2787                 break;
2788         }
2789         case CTL_ERROR_INJECT_DELETE: {
2790                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2791                 int delete_done;
2792
2793                 delete_desc = (struct ctl_error_desc *)addr;
2794                 delete_done = 0;
2795
2796                 mtx_lock(&softc->ctl_lock);
2797                 if (delete_desc->lun_id >= ctl_max_luns ||
2798                     (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) {
2799                         mtx_unlock(&softc->ctl_lock);
2800                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2801                                __func__, (uintmax_t)delete_desc->lun_id);
2802                         retval = EINVAL;
2803                         break;
2804                 }
2805                 mtx_lock(&lun->lun_lock);
2806                 mtx_unlock(&softc->ctl_lock);
2807                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2808                         if (desc->serial != delete_desc->serial)
2809                                 continue;
2810
2811                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2812                                       links);
2813                         free(desc, M_CTL);
2814                         delete_done = 1;
2815                 }
2816                 mtx_unlock(&lun->lun_lock);
2817                 if (delete_done == 0) {
2818                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2819                                "error serial %ju on LUN %u\n", __func__,
2820                                delete_desc->serial, delete_desc->lun_id);
2821                         retval = EINVAL;
2822                         break;
2823                 }
2824                 break;
2825         }
2826         case CTL_DUMP_STRUCTS: {
2827                 int j, k;
2828                 struct ctl_port *port;
2829                 struct ctl_frontend *fe;
2830
2831                 mtx_lock(&softc->ctl_lock);
2832                 printf("CTL Persistent Reservation information start:\n");
2833                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2834                         mtx_lock(&lun->lun_lock);
2835                         if ((lun->flags & CTL_LUN_DISABLED) != 0) {
2836                                 mtx_unlock(&lun->lun_lock);
2837                                 continue;
2838                         }
2839
2840                         for (j = 0; j < ctl_max_ports; j++) {
2841                                 if (lun->pr_keys[j] == NULL)
2842                                         continue;
2843                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2844                                         if (lun->pr_keys[j][k] == 0)
2845                                                 continue;
2846                                         printf("  LUN %ju port %d iid %d key "
2847                                                "%#jx\n", lun->lun, j, k,
2848                                                (uintmax_t)lun->pr_keys[j][k]);
2849                                 }
2850                         }
2851                         mtx_unlock(&lun->lun_lock);
2852                 }
2853                 printf("CTL Persistent Reservation information end\n");
2854                 printf("CTL Ports:\n");
2855                 STAILQ_FOREACH(port, &softc->port_list, links) {
2856                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2857                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2858                                port->frontend->name, port->port_type,
2859                                port->physical_port, port->virtual_port,
2860                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2861                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2862                                 if (port->wwpn_iid[j].in_use == 0 &&
2863                                     port->wwpn_iid[j].wwpn == 0 &&
2864                                     port->wwpn_iid[j].name == NULL)
2865                                         continue;
2866
2867                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2868                                     j, port->wwpn_iid[j].in_use,
2869                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2870                                     port->wwpn_iid[j].name);
2871                         }
2872                 }
2873                 printf("CTL Port information end\n");
2874                 mtx_unlock(&softc->ctl_lock);
2875                 /*
2876                  * XXX KDM calling this without a lock.  We'd likely want
2877                  * to drop the lock before calling the frontend's dump
2878                  * routine anyway.
2879                  */
2880                 printf("CTL Frontends:\n");
2881                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2882                         printf("  Frontend '%s'\n", fe->name);
2883                         if (fe->fe_dump != NULL)
2884                                 fe->fe_dump();
2885                 }
2886                 printf("CTL Frontend information end\n");
2887                 break;
2888         }
2889         case CTL_LUN_REQ: {
2890                 struct ctl_lun_req *lun_req;
2891                 struct ctl_backend_driver *backend;
2892                 void *packed;
2893                 nvlist_t *tmp_args_nvl;
2894                 size_t packed_len;
2895
2896                 lun_req = (struct ctl_lun_req *)addr;
2897                 tmp_args_nvl = lun_req->args_nvl;
2898
2899                 backend = ctl_backend_find(lun_req->backend);
2900                 if (backend == NULL) {
2901                         lun_req->status = CTL_LUN_ERROR;
2902                         snprintf(lun_req->error_str,
2903                                  sizeof(lun_req->error_str),
2904                                  "Backend \"%s\" not found.",
2905                                  lun_req->backend);
2906                         break;
2907                 }
2908
2909                 if (lun_req->args != NULL) {
2910                         packed = malloc(lun_req->args_len, M_CTL, M_WAITOK);
2911                         if (copyin(lun_req->args, packed, lun_req->args_len) != 0) {
2912                                 free(packed, M_CTL);
2913                                 lun_req->status = CTL_LUN_ERROR;
2914                                 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2915                                     "Cannot copyin args.");
2916                                 break;
2917                         }
2918                         lun_req->args_nvl = nvlist_unpack(packed,
2919                             lun_req->args_len, 0);
2920                         free(packed, M_CTL);
2921
2922                         if (lun_req->args_nvl == NULL) {
2923                                 lun_req->status = CTL_LUN_ERROR;
2924                                 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2925                                     "Cannot unpack args nvlist.");
2926                                 break;
2927                         }
2928                 } else
2929                         lun_req->args_nvl = nvlist_create(0);
2930
2931                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2932                 nvlist_destroy(lun_req->args_nvl);
2933                 lun_req->args_nvl = tmp_args_nvl;
2934
2935                 if (lun_req->result_nvl != NULL) {
2936                         if (lun_req->result != NULL) {
2937                                 packed = nvlist_pack(lun_req->result_nvl,
2938                                     &packed_len);
2939                                 if (packed == NULL) {
2940                                         lun_req->status = CTL_LUN_ERROR;
2941                                         snprintf(lun_req->error_str,
2942                                             sizeof(lun_req->error_str),
2943                                             "Cannot pack result nvlist.");
2944                                         break;
2945                                 }
2946
2947                                 if (packed_len > lun_req->result_len) {
2948                                         lun_req->status = CTL_LUN_ERROR;
2949                                         snprintf(lun_req->error_str,
2950                                             sizeof(lun_req->error_str),
2951                                             "Result nvlist too large.");
2952                                         free(packed, M_NVLIST);
2953                                         break;
2954                                 }
2955
2956                                 if (copyout(packed, lun_req->result, packed_len)) {
2957                                         lun_req->status = CTL_LUN_ERROR;
2958                                         snprintf(lun_req->error_str,
2959                                             sizeof(lun_req->error_str),
2960                                             "Cannot copyout() the result.");
2961                                         free(packed, M_NVLIST);
2962                                         break;
2963                                 }
2964
2965                                 lun_req->result_len = packed_len;
2966                                 free(packed, M_NVLIST);
2967                         }
2968
2969                         nvlist_destroy(lun_req->result_nvl);
2970                 }
2971                 break;
2972         }
2973         case CTL_LUN_LIST: {
2974                 struct sbuf *sb;
2975                 struct ctl_lun_list *list;
2976                 const char *name, *value;
2977                 void *cookie;
2978                 int type;
2979
2980                 list = (struct ctl_lun_list *)addr;
2981
2982                 /*
2983                  * Allocate a fixed length sbuf here, based on the length
2984                  * of the user's buffer.  We could allocate an auto-extending
2985                  * buffer, and then tell the user how much larger our
2986                  * amount of data is than his buffer, but that presents
2987                  * some problems:
2988                  *
2989                  * 1.  The sbuf(9) routines use a blocking malloc, and so
2990                  *     we can't hold a lock while calling them with an
2991                  *     auto-extending buffer.
2992                  *
2993                  * 2.  There is not currently a LUN reference counting
2994                  *     mechanism, outside of outstanding transactions on
2995                  *     the LUN's OOA queue.  So a LUN could go away on us
2996                  *     while we're getting the LUN number, backend-specific
2997                  *     information, etc.  Thus, given the way things
2998                  *     currently work, we need to hold the CTL lock while
2999                  *     grabbing LUN information.
3000                  *
3001                  * So, from the user's standpoint, the best thing to do is
3002                  * allocate what he thinks is a reasonable buffer length,
3003                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3004                  * double the buffer length and try again.  (And repeat
3005                  * that until he succeeds.)
3006                  */
3007                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3008                 if (sb == NULL) {
3009                         list->status = CTL_LUN_LIST_ERROR;
3010                         snprintf(list->error_str, sizeof(list->error_str),
3011                                  "Unable to allocate %d bytes for LUN list",
3012                                  list->alloc_len);
3013                         break;
3014                 }
3015
3016                 sbuf_printf(sb, "<ctllunlist>\n");
3017
3018                 mtx_lock(&softc->ctl_lock);
3019                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3020                         mtx_lock(&lun->lun_lock);
3021                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3022                                              (uintmax_t)lun->lun);
3023
3024                         /*
3025                          * Bail out as soon as we see that we've overfilled
3026                          * the buffer.
3027                          */
3028                         if (retval != 0)
3029                                 break;
3030
3031                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3032                                              "</backend_type>\n",
3033                                              (lun->backend == NULL) ?  "none" :
3034                                              lun->backend->name);
3035
3036                         if (retval != 0)
3037                                 break;
3038
3039                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3040                                              lun->be_lun->lun_type);
3041
3042                         if (retval != 0)
3043                                 break;
3044
3045                         if (lun->backend == NULL) {
3046                                 retval = sbuf_printf(sb, "</lun>\n");
3047                                 if (retval != 0)
3048                                         break;
3049                                 continue;
3050                         }
3051
3052                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3053                                              (lun->be_lun->maxlba > 0) ?
3054                                              lun->be_lun->maxlba + 1 : 0);
3055
3056                         if (retval != 0)
3057                                 break;
3058
3059                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3060                                              lun->be_lun->blocksize);
3061
3062                         if (retval != 0)
3063                                 break;
3064
3065                         retval = sbuf_printf(sb, "\t<serial_number>");
3066
3067                         if (retval != 0)
3068                                 break;
3069
3070                         retval = ctl_sbuf_printf_esc(sb,
3071                             lun->be_lun->serial_num,
3072                             sizeof(lun->be_lun->serial_num));
3073
3074                         if (retval != 0)
3075                                 break;
3076
3077                         retval = sbuf_printf(sb, "</serial_number>\n");
3078
3079                         if (retval != 0)
3080                                 break;
3081
3082                         retval = sbuf_printf(sb, "\t<device_id>");
3083
3084                         if (retval != 0)
3085                                 break;
3086
3087                         retval = ctl_sbuf_printf_esc(sb,
3088                             lun->be_lun->device_id,
3089                             sizeof(lun->be_lun->device_id));
3090
3091                         if (retval != 0)
3092                                 break;
3093
3094                         retval = sbuf_printf(sb, "</device_id>\n");
3095
3096                         if (retval != 0)
3097                                 break;
3098
3099                         if (lun->backend->lun_info != NULL) {
3100                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3101                                 if (retval != 0)
3102                                         break;
3103                         }
3104
3105                         cookie = NULL;
3106                         while ((name = nvlist_next(lun->be_lun->options, &type,
3107                             &cookie)) != NULL) {
3108                                 sbuf_printf(sb, "\t<%s>", name);
3109
3110                                 if (type == NV_TYPE_STRING) {
3111                                         value = dnvlist_get_string(
3112                                             lun->be_lun->options, name, NULL);
3113                                         if (value != NULL)
3114                                                 sbuf_printf(sb, "%s", value);
3115                                 }
3116
3117                                 sbuf_printf(sb, "</%s>\n", name);
3118                         }
3119
3120                         retval = sbuf_printf(sb, "</lun>\n");
3121
3122                         if (retval != 0)
3123                                 break;
3124                         mtx_unlock(&lun->lun_lock);
3125                 }
3126                 if (lun != NULL)
3127                         mtx_unlock(&lun->lun_lock);
3128                 mtx_unlock(&softc->ctl_lock);
3129
3130                 if ((retval != 0)
3131                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3132                         retval = 0;
3133                         sbuf_delete(sb);
3134                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3135                         snprintf(list->error_str, sizeof(list->error_str),
3136                                  "Out of space, %d bytes is too small",
3137                                  list->alloc_len);
3138                         break;
3139                 }
3140
3141                 sbuf_finish(sb);
3142
3143                 retval = copyout(sbuf_data(sb), list->lun_xml,
3144                                  sbuf_len(sb) + 1);
3145
3146                 list->fill_len = sbuf_len(sb) + 1;
3147                 list->status = CTL_LUN_LIST_OK;
3148                 sbuf_delete(sb);
3149                 break;
3150         }
3151         case CTL_ISCSI: {
3152                 struct ctl_iscsi *ci;
3153                 struct ctl_frontend *fe;
3154
3155                 ci = (struct ctl_iscsi *)addr;
3156
3157                 fe = ctl_frontend_find("iscsi");
3158                 if (fe == NULL) {
3159                         ci->status = CTL_ISCSI_ERROR;
3160                         snprintf(ci->error_str, sizeof(ci->error_str),
3161                             "Frontend \"iscsi\" not found.");
3162                         break;
3163                 }
3164
3165                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3166                 break;
3167         }
3168         case CTL_PORT_REQ: {
3169                 struct ctl_req *req;
3170                 struct ctl_frontend *fe;
3171                 void *packed;
3172                 nvlist_t *tmp_args_nvl;
3173                 size_t packed_len;
3174
3175                 req = (struct ctl_req *)addr;
3176                 tmp_args_nvl = req->args_nvl;
3177
3178                 fe = ctl_frontend_find(req->driver);
3179                 if (fe == NULL) {
3180                         req->status = CTL_LUN_ERROR;
3181                         snprintf(req->error_str, sizeof(req->error_str),
3182                             "Frontend \"%s\" not found.", req->driver);
3183                         break;
3184                 }
3185
3186                 if (req->args != NULL) {
3187                         packed = malloc(req->args_len, M_CTL, M_WAITOK);
3188                         if (copyin(req->args, packed, req->args_len) != 0) {
3189                                 free(packed, M_CTL);
3190                                 req->status = CTL_LUN_ERROR;
3191                                 snprintf(req->error_str, sizeof(req->error_str),
3192                                     "Cannot copyin args.");
3193                                 break;
3194                         }
3195                         req->args_nvl = nvlist_unpack(packed,
3196                             req->args_len, 0);
3197                         free(packed, M_CTL);
3198
3199                         if (req->args_nvl == NULL) {
3200                                 req->status = CTL_LUN_ERROR;
3201                                 snprintf(req->error_str, sizeof(req->error_str),
3202                                     "Cannot unpack args nvlist.");
3203                                 break;
3204                         }
3205                 } else
3206                         req->args_nvl = nvlist_create(0);
3207
3208                 if (fe->ioctl)
3209                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3210                 else
3211                         retval = ENODEV;
3212
3213                 nvlist_destroy(req->args_nvl);
3214                 req->args_nvl = tmp_args_nvl;
3215
3216                 if (req->result_nvl != NULL) {
3217                         if (req->result != NULL) {
3218                                 packed = nvlist_pack(req->result_nvl,
3219                                     &packed_len);
3220                                 if (packed == NULL) {
3221                                         req->status = CTL_LUN_ERROR;
3222                                         snprintf(req->error_str,
3223                                             sizeof(req->error_str),
3224                                             "Cannot pack result nvlist.");
3225                                         break;
3226                                 }
3227
3228                                 if (packed_len > req->result_len) {
3229                                         req->status = CTL_LUN_ERROR;
3230                                         snprintf(req->error_str,
3231                                             sizeof(req->error_str),
3232                                             "Result nvlist too large.");
3233                                         free(packed, M_NVLIST);
3234                                         break;
3235                                 }
3236
3237                                 if (copyout(packed, req->result, packed_len)) {
3238                                         req->status = CTL_LUN_ERROR;
3239                                         snprintf(req->error_str,
3240                                             sizeof(req->error_str),
3241                                             "Cannot copyout() the result.");
3242                                         free(packed, M_NVLIST);
3243                                         break;
3244                                 }
3245
3246                                 req->result_len = packed_len;
3247                                 free(packed, M_NVLIST);
3248                         }
3249
3250                         nvlist_destroy(req->result_nvl);
3251                 }
3252                 break;
3253         }
3254         case CTL_PORT_LIST: {
3255                 struct sbuf *sb;
3256                 struct ctl_port *port;
3257                 struct ctl_lun_list *list;
3258                 const char *name, *value;
3259                 void *cookie;
3260                 int j, type;
3261                 uint32_t plun;
3262
3263                 list = (struct ctl_lun_list *)addr;
3264
3265                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3266                 if (sb == NULL) {
3267                         list->status = CTL_LUN_LIST_ERROR;
3268                         snprintf(list->error_str, sizeof(list->error_str),
3269                                  "Unable to allocate %d bytes for LUN list",
3270                                  list->alloc_len);
3271                         break;
3272                 }
3273
3274                 sbuf_printf(sb, "<ctlportlist>\n");
3275
3276                 mtx_lock(&softc->ctl_lock);
3277                 STAILQ_FOREACH(port, &softc->port_list, links) {
3278                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3279                                              (uintmax_t)port->targ_port);
3280
3281                         /*
3282                          * Bail out as soon as we see that we've overfilled
3283                          * the buffer.
3284                          */
3285                         if (retval != 0)
3286                                 break;
3287
3288                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3289                             "</frontend_type>\n", port->frontend->name);
3290                         if (retval != 0)
3291                                 break;
3292
3293                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3294                                              port->port_type);
3295                         if (retval != 0)
3296                                 break;
3297
3298                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3299                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3300                         if (retval != 0)
3301                                 break;
3302
3303                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3304                             port->port_name);
3305                         if (retval != 0)
3306                                 break;
3307
3308                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3309                             port->physical_port);
3310                         if (retval != 0)
3311                                 break;
3312
3313                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3314                             port->virtual_port);
3315                         if (retval != 0)
3316                                 break;
3317
3318                         if (port->target_devid != NULL) {
3319                                 sbuf_printf(sb, "\t<target>");
3320                                 ctl_id_sbuf(port->target_devid, sb);
3321                                 sbuf_printf(sb, "</target>\n");
3322                         }
3323
3324                         if (port->port_devid != NULL) {
3325                                 sbuf_printf(sb, "\t<port>");
3326                                 ctl_id_sbuf(port->port_devid, sb);
3327                                 sbuf_printf(sb, "</port>\n");
3328                         }
3329
3330                         if (port->port_info != NULL) {
3331                                 retval = port->port_info(port->onoff_arg, sb);
3332                                 if (retval != 0)
3333                                         break;
3334                         }
3335
3336                         cookie = NULL;
3337                         while ((name = nvlist_next(port->options, &type,
3338                             &cookie)) != NULL) {
3339                                 sbuf_printf(sb, "\t<%s>", name);
3340
3341                                 if (type == NV_TYPE_STRING) {
3342                                         value = dnvlist_get_string(port->options,
3343                                             name, NULL);
3344                                         if (value != NULL)
3345                                                 sbuf_printf(sb, "%s", value);
3346                                 }
3347
3348                                 sbuf_printf(sb, "</%s>\n", name);
3349                         }
3350
3351                         if (port->lun_map != NULL) {
3352                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3353                                 for (j = 0; j < port->lun_map_size; j++) {
3354                                         plun = ctl_lun_map_from_port(port, j);
3355                                         if (plun == UINT32_MAX)
3356                                                 continue;
3357                                         sbuf_printf(sb,
3358                                             "\t<lun id=\"%u\">%u</lun>\n",
3359                                             j, plun);
3360                                 }
3361                         }
3362
3363                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3364                                 if (port->wwpn_iid[j].in_use == 0 ||
3365                                     (port->wwpn_iid[j].wwpn == 0 &&
3366                                      port->wwpn_iid[j].name == NULL))
3367                                         continue;
3368
3369                                 if (port->wwpn_iid[j].name != NULL)
3370                                         retval = sbuf_printf(sb,
3371                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3372                                             j, port->wwpn_iid[j].name);
3373                                 else
3374                                         retval = sbuf_printf(sb,
3375                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3376                                             j, port->wwpn_iid[j].wwpn);
3377                                 if (retval != 0)
3378                                         break;
3379                         }
3380                         if (retval != 0)
3381                                 break;
3382
3383                         retval = sbuf_printf(sb, "</targ_port>\n");
3384                         if (retval != 0)
3385                                 break;
3386                 }
3387                 mtx_unlock(&softc->ctl_lock);
3388
3389                 if ((retval != 0)
3390                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3391                         retval = 0;
3392                         sbuf_delete(sb);
3393                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3394                         snprintf(list->error_str, sizeof(list->error_str),
3395                                  "Out of space, %d bytes is too small",
3396                                  list->alloc_len);
3397                         break;
3398                 }
3399
3400                 sbuf_finish(sb);
3401
3402                 retval = copyout(sbuf_data(sb), list->lun_xml,
3403                                  sbuf_len(sb) + 1);
3404
3405                 list->fill_len = sbuf_len(sb) + 1;
3406                 list->status = CTL_LUN_LIST_OK;
3407                 sbuf_delete(sb);
3408                 break;
3409         }
3410         case CTL_LUN_MAP: {
3411                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3412                 struct ctl_port *port;
3413
3414                 mtx_lock(&softc->ctl_lock);
3415                 if (lm->port < softc->port_min ||
3416                     lm->port >= softc->port_max ||
3417                     (port = softc->ctl_ports[lm->port]) == NULL) {
3418                         mtx_unlock(&softc->ctl_lock);
3419                         return (ENXIO);
3420                 }
3421                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3422                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3423                                 if (ctl_lun_map_to_port(port, lun->lun) ==
3424                                     UINT32_MAX)
3425                                         continue;
3426                                 mtx_lock(&lun->lun_lock);
3427                                 ctl_est_ua_port(lun, lm->port, -1,
3428                                     CTL_UA_LUN_CHANGE);
3429                                 mtx_unlock(&lun->lun_lock);
3430                         }
3431                 }
3432                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3433                 if (lm->plun != UINT32_MAX) {
3434                         if (lm->lun == UINT32_MAX)
3435                                 retval = ctl_lun_map_unset(port, lm->plun);
3436                         else if (lm->lun < ctl_max_luns &&
3437                             softc->ctl_luns[lm->lun] != NULL)
3438                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3439                         else
3440                                 return (ENXIO);
3441                 } else {
3442                         if (lm->lun == UINT32_MAX)
3443                                 retval = ctl_lun_map_deinit(port);
3444                         else
3445                                 retval = ctl_lun_map_init(port);
3446                 }
3447                 if (port->status & CTL_PORT_STATUS_ONLINE)
3448                         ctl_isc_announce_port(port);
3449                 break;
3450         }
3451         case CTL_GET_LUN_STATS: {
3452                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3453                 int i;
3454
3455                 /*
3456                  * XXX KDM no locking here.  If the LUN list changes,
3457                  * things can blow up.
3458                  */
3459                 i = 0;
3460                 stats->status = CTL_SS_OK;
3461                 stats->fill_len = 0;
3462                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3463                         if (lun->lun < stats->first_item)
3464                                 continue;
3465                         if (stats->fill_len + sizeof(lun->stats) >
3466                             stats->alloc_len) {
3467                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3468                                 break;
3469                         }
3470                         retval = copyout(&lun->stats, &stats->stats[i++],
3471                                          sizeof(lun->stats));
3472                         if (retval != 0)
3473                                 break;
3474                         stats->fill_len += sizeof(lun->stats);
3475                 }
3476                 stats->num_items = softc->num_luns;
3477                 stats->flags = CTL_STATS_FLAG_NONE;
3478 #ifdef CTL_TIME_IO
3479                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3480 #endif
3481                 getnanouptime(&stats->timestamp);
3482                 break;
3483         }
3484         case CTL_GET_PORT_STATS: {
3485                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3486                 int i;
3487
3488                 /*
3489                  * XXX KDM no locking here.  If the LUN list changes,
3490                  * things can blow up.
3491                  */
3492                 i = 0;
3493                 stats->status = CTL_SS_OK;
3494                 stats->fill_len = 0;
3495                 STAILQ_FOREACH(port, &softc->port_list, links) {
3496                         if (port->targ_port < stats->first_item)
3497                                 continue;
3498                         if (stats->fill_len + sizeof(port->stats) >
3499                             stats->alloc_len) {
3500                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3501                                 break;
3502                         }
3503                         retval = copyout(&port->stats, &stats->stats[i++],
3504                                          sizeof(port->stats));
3505                         if (retval != 0)
3506                                 break;
3507                         stats->fill_len += sizeof(port->stats);
3508                 }
3509                 stats->num_items = softc->num_ports;
3510                 stats->flags = CTL_STATS_FLAG_NONE;
3511 #ifdef CTL_TIME_IO
3512                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3513 #endif
3514                 getnanouptime(&stats->timestamp);
3515                 break;
3516         }
3517         default: {
3518                 /* XXX KDM should we fix this? */
3519 #if 0
3520                 struct ctl_backend_driver *backend;
3521                 unsigned int type;
3522                 int found;
3523
3524                 found = 0;
3525
3526                 /*
3527                  * We encode the backend type as the ioctl type for backend
3528                  * ioctls.  So parse it out here, and then search for a
3529                  * backend of this type.
3530                  */
3531                 type = _IOC_TYPE(cmd);
3532
3533                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3534                         if (backend->type == type) {
3535                                 found = 1;
3536                                 break;
3537                         }
3538                 }
3539                 if (found == 0) {
3540                         printf("ctl: unknown ioctl command %#lx or backend "
3541                                "%d\n", cmd, type);
3542                         retval = EINVAL;
3543                         break;
3544                 }
3545                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3546 #endif
3547                 retval = ENOTTY;
3548                 break;
3549         }
3550         }
3551         return (retval);
3552 }
3553
3554 uint32_t
3555 ctl_get_initindex(struct ctl_nexus *nexus)
3556 {
3557         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3558 }
3559
3560 int
3561 ctl_lun_map_init(struct ctl_port *port)
3562 {
3563         struct ctl_softc *softc = port->ctl_softc;
3564         struct ctl_lun *lun;
3565         int size = ctl_lun_map_size;
3566         uint32_t i;
3567
3568         if (port->lun_map == NULL || port->lun_map_size < size) {
3569                 port->lun_map_size = 0;
3570                 free(port->lun_map, M_CTL);
3571                 port->lun_map = malloc(size * sizeof(uint32_t),
3572                     M_CTL, M_NOWAIT);
3573         }
3574         if (port->lun_map == NULL)
3575                 return (ENOMEM);
3576         for (i = 0; i < size; i++)
3577                 port->lun_map[i] = UINT32_MAX;
3578         port->lun_map_size = size;
3579         if (port->status & CTL_PORT_STATUS_ONLINE) {
3580                 if (port->lun_disable != NULL) {
3581                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3582                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3583                 }
3584                 ctl_isc_announce_port(port);
3585         }
3586         return (0);
3587 }
3588
3589 int
3590 ctl_lun_map_deinit(struct ctl_port *port)
3591 {
3592         struct ctl_softc *softc = port->ctl_softc;
3593         struct ctl_lun *lun;
3594
3595         if (port->lun_map == NULL)
3596                 return (0);
3597         port->lun_map_size = 0;
3598         free(port->lun_map, M_CTL);
3599         port->lun_map = NULL;
3600         if (port->status & CTL_PORT_STATUS_ONLINE) {
3601                 if (port->lun_enable != NULL) {
3602                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3603                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3604                 }
3605                 ctl_isc_announce_port(port);
3606         }
3607         return (0);
3608 }
3609
3610 int
3611 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3612 {
3613         int status;
3614         uint32_t old;
3615
3616         if (port->lun_map == NULL) {
3617                 status = ctl_lun_map_init(port);
3618                 if (status != 0)
3619                         return (status);
3620         }
3621         if (plun >= port->lun_map_size)
3622                 return (EINVAL);
3623         old = port->lun_map[plun];
3624         port->lun_map[plun] = glun;
3625         if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) {
3626                 if (port->lun_enable != NULL)
3627                         port->lun_enable(port->targ_lun_arg, plun);
3628                 ctl_isc_announce_port(port);
3629         }
3630         return (0);
3631 }
3632
3633 int
3634 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3635 {
3636         uint32_t old;
3637
3638         if (port->lun_map == NULL || plun >= port->lun_map_size)
3639                 return (0);
3640         old = port->lun_map[plun];
3641         port->lun_map[plun] = UINT32_MAX;
3642         if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) {
3643                 if (port->lun_disable != NULL)
3644                         port->lun_disable(port->targ_lun_arg, plun);
3645                 ctl_isc_announce_port(port);
3646         }
3647         return (0);
3648 }
3649
3650 uint32_t
3651 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3652 {
3653
3654         if (port == NULL)
3655                 return (UINT32_MAX);
3656         if (port->lun_map == NULL)
3657                 return (lun_id);
3658         if (lun_id > port->lun_map_size)
3659                 return (UINT32_MAX);
3660         return (port->lun_map[lun_id]);
3661 }
3662
3663 uint32_t
3664 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3665 {
3666         uint32_t i;
3667
3668         if (port == NULL)
3669                 return (UINT32_MAX);
3670         if (port->lun_map == NULL)
3671                 return (lun_id);
3672         for (i = 0; i < port->lun_map_size; i++) {
3673                 if (port->lun_map[i] == lun_id)
3674                         return (i);
3675         }
3676         return (UINT32_MAX);
3677 }
3678
3679 uint32_t
3680 ctl_decode_lun(uint64_t encoded)
3681 {
3682         uint8_t lun[8];
3683         uint32_t result = 0xffffffff;
3684
3685         be64enc(lun, encoded);
3686         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3687         case RPL_LUNDATA_ATYP_PERIPH:
3688                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3689                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3690                         result = lun[1];
3691                 break;
3692         case RPL_LUNDATA_ATYP_FLAT:
3693                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3694                     lun[6] == 0 && lun[7] == 0)
3695                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3696                 break;
3697         case RPL_LUNDATA_ATYP_EXTLUN:
3698                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3699                 case 0x02:
3700                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3701                         case 0x00:
3702                                 result = lun[1];
3703                                 break;
3704                         case 0x10:
3705                                 result = (lun[1] << 16) + (lun[2] << 8) +
3706                                     lun[3];
3707                                 break;
3708                         case 0x20:
3709                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3710                                         result = (lun[2] << 24) +
3711                                             (lun[3] << 16) + (lun[4] << 8) +
3712                                             lun[5];
3713                                 break;
3714                         }
3715                         break;
3716                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3717                         result = 0xffffffff;
3718                         break;
3719                 }
3720                 break;
3721         }
3722         return (result);
3723 }
3724
3725 uint64_t
3726 ctl_encode_lun(uint32_t decoded)
3727 {
3728         uint64_t l = decoded;
3729
3730         if (l <= 0xff)
3731                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3732         if (l <= 0x3fff)
3733                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3734         if (l <= 0xffffff)
3735                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3736                     (l << 32));
3737         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3738 }
3739
3740 int
3741 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3742 {
3743         int i;
3744
3745         for (i = first; i < last; i++) {
3746                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3747                         return (i);
3748         }
3749         return (-1);
3750 }
3751
3752 int
3753 ctl_set_mask(uint32_t *mask, uint32_t bit)
3754 {
3755         uint32_t chunk, piece;
3756
3757         chunk = bit >> 5;
3758         piece = bit % (sizeof(uint32_t) * 8);
3759
3760         if ((mask[chunk] & (1 << piece)) != 0)
3761                 return (-1);
3762         else
3763                 mask[chunk] |= (1 << piece);
3764
3765         return (0);
3766 }
3767
3768 int
3769 ctl_clear_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_is_set(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 (0);
3794         else
3795                 return (1);
3796 }
3797
3798 static uint64_t
3799 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3800 {
3801         uint64_t *t;
3802
3803         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3804         if (t == NULL)
3805                 return (0);
3806         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3807 }
3808
3809 static void
3810 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3811 {
3812         uint64_t *t;
3813
3814         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3815         if (t == NULL)
3816                 return;
3817         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3818 }
3819
3820 static void
3821 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3822 {
3823         uint64_t *p;
3824         u_int i;
3825
3826         i = residx/CTL_MAX_INIT_PER_PORT;
3827         if (lun->pr_keys[i] != NULL)
3828                 return;
3829         mtx_unlock(&lun->lun_lock);
3830         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3831             M_WAITOK | M_ZERO);
3832         mtx_lock(&lun->lun_lock);
3833         if (lun->pr_keys[i] == NULL)
3834                 lun->pr_keys[i] = p;
3835         else
3836                 free(p, M_CTL);
3837 }
3838
3839 static void
3840 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3841 {
3842         uint64_t *t;
3843
3844         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3845         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3846         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3847 }
3848
3849 /*
3850  * ctl_softc, pool_name, total_ctl_io are passed in.
3851  * npool is passed out.
3852  */
3853 int
3854 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3855                 uint32_t total_ctl_io, void **npool)
3856 {
3857         struct ctl_io_pool *pool;
3858
3859         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3860                                             M_NOWAIT | M_ZERO);
3861         if (pool == NULL)
3862                 return (ENOMEM);
3863
3864         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3865         pool->ctl_softc = ctl_softc;
3866 #ifdef IO_POOLS
3867         pool->zone = uma_zsecond_create(pool->name, NULL,
3868             NULL, NULL, NULL, ctl_softc->io_zone);
3869         /* uma_prealloc(pool->zone, total_ctl_io); */
3870 #else
3871         pool->zone = ctl_softc->io_zone;
3872 #endif
3873
3874         *npool = pool;
3875         return (0);
3876 }
3877
3878 void
3879 ctl_pool_free(struct ctl_io_pool *pool)
3880 {
3881
3882         if (pool == NULL)
3883                 return;
3884
3885 #ifdef IO_POOLS
3886         uma_zdestroy(pool->zone);
3887 #endif
3888         free(pool, M_CTL);
3889 }
3890
3891 union ctl_io *
3892 ctl_alloc_io(void *pool_ref)
3893 {
3894         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3895         union ctl_io *io;
3896
3897         io = uma_zalloc(pool->zone, M_WAITOK);
3898         if (io != NULL) {
3899                 io->io_hdr.pool = pool_ref;
3900                 CTL_SOFTC(io) = pool->ctl_softc;
3901                 TAILQ_INIT(&io->io_hdr.blocked_queue);
3902         }
3903         return (io);
3904 }
3905
3906 union ctl_io *
3907 ctl_alloc_io_nowait(void *pool_ref)
3908 {
3909         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3910         union ctl_io *io;
3911
3912         io = uma_zalloc(pool->zone, M_NOWAIT);
3913         if (io != NULL) {
3914                 io->io_hdr.pool = pool_ref;
3915                 CTL_SOFTC(io) = pool->ctl_softc;
3916                 TAILQ_INIT(&io->io_hdr.blocked_queue);
3917         }
3918         return (io);
3919 }
3920
3921 void
3922 ctl_free_io(union ctl_io *io)
3923 {
3924         struct ctl_io_pool *pool;
3925
3926         if (io == NULL)
3927                 return;
3928
3929         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3930         uma_zfree(pool->zone, io);
3931 }
3932
3933 void
3934 ctl_zero_io(union ctl_io *io)
3935 {
3936         struct ctl_io_pool *pool;
3937
3938         if (io == NULL)
3939                 return;
3940
3941         /*
3942          * May need to preserve linked list pointers at some point too.
3943          */
3944         pool = io->io_hdr.pool;
3945         memset(io, 0, sizeof(*io));
3946         io->io_hdr.pool = pool;
3947         CTL_SOFTC(io) = pool->ctl_softc;
3948         TAILQ_INIT(&io->io_hdr.blocked_queue);
3949 }
3950
3951 int
3952 ctl_expand_number(const char *buf, uint64_t *num)
3953 {
3954         char *endptr;
3955         uint64_t number;
3956         unsigned shift;
3957
3958         number = strtoq(buf, &endptr, 0);
3959
3960         switch (tolower((unsigned char)*endptr)) {
3961         case 'e':
3962                 shift = 60;
3963                 break;
3964         case 'p':
3965                 shift = 50;
3966                 break;
3967         case 't':
3968                 shift = 40;
3969                 break;
3970         case 'g':
3971                 shift = 30;
3972                 break;
3973         case 'm':
3974                 shift = 20;
3975                 break;
3976         case 'k':
3977                 shift = 10;
3978                 break;
3979         case 'b':
3980         case '\0': /* No unit. */
3981                 *num = number;
3982                 return (0);
3983         default:
3984                 /* Unrecognized unit. */
3985                 return (-1);
3986         }
3987
3988         if ((number << shift) >> shift != number) {
3989                 /* Overflow */
3990                 return (-1);
3991         }
3992         *num = number << shift;
3993         return (0);
3994 }
3995
3996
3997 /*
3998  * This routine could be used in the future to load default and/or saved
3999  * mode page parameters for a particuar lun.
4000  */
4001 static int
4002 ctl_init_page_index(struct ctl_lun *lun)
4003 {
4004         int i, page_code;
4005         struct ctl_page_index *page_index;
4006         const char *value;
4007         uint64_t ival;
4008
4009         memcpy(&lun->mode_pages.index, page_index_template,
4010                sizeof(page_index_template));
4011
4012         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4013
4014                 page_index = &lun->mode_pages.index[i];
4015                 if (lun->be_lun->lun_type == T_DIRECT &&
4016                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4017                         continue;
4018                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4019                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4020                         continue;
4021                 if (lun->be_lun->lun_type == T_CDROM &&
4022                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4023                         continue;
4024
4025                 page_code = page_index->page_code & SMPH_PC_MASK;
4026                 switch (page_code) {
4027                 case SMS_RW_ERROR_RECOVERY_PAGE: {
4028                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4029                             ("subpage %#x for page %#x is incorrect!",
4030                             page_index->subpage, page_code));
4031                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4032                                &rw_er_page_default,
4033                                sizeof(rw_er_page_default));
4034                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4035                                &rw_er_page_changeable,
4036                                sizeof(rw_er_page_changeable));
4037                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4038                                &rw_er_page_default,
4039                                sizeof(rw_er_page_default));
4040                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4041                                &rw_er_page_default,
4042                                sizeof(rw_er_page_default));
4043                         page_index->page_data =
4044                                 (uint8_t *)lun->mode_pages.rw_er_page;
4045                         break;
4046                 }
4047                 case SMS_FORMAT_DEVICE_PAGE: {
4048                         struct scsi_format_page *format_page;
4049
4050                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4051                             ("subpage %#x for page %#x is incorrect!",
4052                             page_index->subpage, page_code));
4053
4054                         /*
4055                          * Sectors per track are set above.  Bytes per
4056                          * sector need to be set here on a per-LUN basis.
4057                          */
4058                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4059                                &format_page_default,
4060                                sizeof(format_page_default));
4061                         memcpy(&lun->mode_pages.format_page[
4062                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4063                                sizeof(format_page_changeable));
4064                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4065                                &format_page_default,
4066                                sizeof(format_page_default));
4067                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4068                                &format_page_default,
4069                                sizeof(format_page_default));
4070
4071                         format_page = &lun->mode_pages.format_page[
4072                                 CTL_PAGE_CURRENT];
4073                         scsi_ulto2b(lun->be_lun->blocksize,
4074                                     format_page->bytes_per_sector);
4075
4076                         format_page = &lun->mode_pages.format_page[
4077                                 CTL_PAGE_DEFAULT];
4078                         scsi_ulto2b(lun->be_lun->blocksize,
4079                                     format_page->bytes_per_sector);
4080
4081                         format_page = &lun->mode_pages.format_page[
4082                                 CTL_PAGE_SAVED];
4083                         scsi_ulto2b(lun->be_lun->blocksize,
4084                                     format_page->bytes_per_sector);
4085
4086                         page_index->page_data =
4087                                 (uint8_t *)lun->mode_pages.format_page;
4088                         break;
4089                 }
4090                 case SMS_RIGID_DISK_PAGE: {
4091                         struct scsi_rigid_disk_page *rigid_disk_page;
4092                         uint32_t sectors_per_cylinder;
4093                         uint64_t cylinders;
4094 #ifndef __XSCALE__
4095                         int shift;
4096 #endif /* !__XSCALE__ */
4097
4098                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4099                             ("subpage %#x for page %#x is incorrect!",
4100                             page_index->subpage, page_code));
4101
4102                         /*
4103                          * Rotation rate and sectors per track are set
4104                          * above.  We calculate the cylinders here based on
4105                          * capacity.  Due to the number of heads and
4106                          * sectors per track we're using, smaller arrays
4107                          * may turn out to have 0 cylinders.  Linux and
4108                          * FreeBSD don't pay attention to these mode pages
4109                          * to figure out capacity, but Solaris does.  It
4110                          * seems to deal with 0 cylinders just fine, and
4111                          * works out a fake geometry based on the capacity.
4112                          */
4113                         memcpy(&lun->mode_pages.rigid_disk_page[
4114                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4115                                sizeof(rigid_disk_page_default));
4116                         memcpy(&lun->mode_pages.rigid_disk_page[
4117                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4118                                sizeof(rigid_disk_page_changeable));
4119
4120                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4121                                 CTL_DEFAULT_HEADS;
4122
4123                         /*
4124                          * The divide method here will be more accurate,
4125                          * probably, but results in floating point being
4126                          * used in the kernel on i386 (__udivdi3()).  On the
4127                          * XScale, though, __udivdi3() is implemented in
4128                          * software.
4129                          *
4130                          * The shift method for cylinder calculation is
4131                          * accurate if sectors_per_cylinder is a power of
4132                          * 2.  Otherwise it might be slightly off -- you
4133                          * might have a bit of a truncation problem.
4134                          */
4135 #ifdef  __XSCALE__
4136                         cylinders = (lun->be_lun->maxlba + 1) /
4137                                 sectors_per_cylinder;
4138 #else
4139                         for (shift = 31; shift > 0; shift--) {
4140                                 if (sectors_per_cylinder & (1 << shift))
4141                                         break;
4142                         }
4143                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4144 #endif
4145
4146                         /*
4147                          * We've basically got 3 bytes, or 24 bits for the
4148                          * cylinder size in the mode page.  If we're over,
4149                          * just round down to 2^24.
4150                          */
4151                         if (cylinders > 0xffffff)
4152                                 cylinders = 0xffffff;
4153
4154                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4155                                 CTL_PAGE_DEFAULT];
4156                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4157
4158                         if ((value = dnvlist_get_string(lun->be_lun->options,
4159                             "rpm", NULL)) != NULL) {
4160                                 scsi_ulto2b(strtol(value, NULL, 0),
4161                                      rigid_disk_page->rotation_rate);
4162                         }
4163
4164                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4165                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4166                                sizeof(rigid_disk_page_default));
4167                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4168                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4169                                sizeof(rigid_disk_page_default));
4170
4171                         page_index->page_data =
4172                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4173                         break;
4174                 }
4175                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4176                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4177                             ("subpage %#x for page %#x is incorrect!",
4178                             page_index->subpage, page_code));
4179                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4180                                &verify_er_page_default,
4181                                sizeof(verify_er_page_default));
4182                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4183                                &verify_er_page_changeable,
4184                                sizeof(verify_er_page_changeable));
4185                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4186                                &verify_er_page_default,
4187                                sizeof(verify_er_page_default));
4188                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4189                                &verify_er_page_default,
4190                                sizeof(verify_er_page_default));
4191                         page_index->page_data =
4192                                 (uint8_t *)lun->mode_pages.verify_er_page;
4193                         break;
4194                 }
4195                 case SMS_CACHING_PAGE: {
4196                         struct scsi_caching_page *caching_page;
4197
4198                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4199                             ("subpage %#x for page %#x is incorrect!",
4200                             page_index->subpage, page_code));
4201                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4202                                &caching_page_default,
4203                                sizeof(caching_page_default));
4204                         memcpy(&lun->mode_pages.caching_page[
4205                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4206                                sizeof(caching_page_changeable));
4207                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4208                                &caching_page_default,
4209                                sizeof(caching_page_default));
4210                         caching_page = &lun->mode_pages.caching_page[
4211                             CTL_PAGE_SAVED];
4212                         value = dnvlist_get_string(lun->be_lun->options,
4213                             "writecache", NULL);
4214                         if (value != NULL && strcmp(value, "off") == 0)
4215                                 caching_page->flags1 &= ~SCP_WCE;
4216                         value = dnvlist_get_string(lun->be_lun->options,
4217                             "readcache", NULL);
4218                         if (value != NULL && strcmp(value, "off") == 0)
4219                                 caching_page->flags1 |= SCP_RCD;
4220                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4221                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4222                                sizeof(caching_page_default));
4223                         page_index->page_data =
4224                                 (uint8_t *)lun->mode_pages.caching_page;
4225                         break;
4226                 }
4227                 case SMS_CONTROL_MODE_PAGE: {
4228                         switch (page_index->subpage) {
4229                         case SMS_SUBPAGE_PAGE_0: {
4230                                 struct scsi_control_page *control_page;
4231
4232                                 memcpy(&lun->mode_pages.control_page[
4233                                     CTL_PAGE_DEFAULT],
4234                                        &control_page_default,
4235                                        sizeof(control_page_default));
4236                                 memcpy(&lun->mode_pages.control_page[
4237                                     CTL_PAGE_CHANGEABLE],
4238                                        &control_page_changeable,
4239                                        sizeof(control_page_changeable));
4240                                 memcpy(&lun->mode_pages.control_page[
4241                                     CTL_PAGE_SAVED],
4242                                        &control_page_default,
4243                                        sizeof(control_page_default));
4244                                 control_page = &lun->mode_pages.control_page[
4245                                     CTL_PAGE_SAVED];
4246                                 value = dnvlist_get_string(lun->be_lun->options,
4247                                     "reordering", NULL);
4248                                 if (value != NULL &&
4249                                     strcmp(value, "unrestricted") == 0) {
4250                                         control_page->queue_flags &=
4251                                             ~SCP_QUEUE_ALG_MASK;
4252                                         control_page->queue_flags |=
4253                                             SCP_QUEUE_ALG_UNRESTRICTED;
4254                                 }
4255                                 memcpy(&lun->mode_pages.control_page[
4256                                     CTL_PAGE_CURRENT],
4257                                        &lun->mode_pages.control_page[
4258                                     CTL_PAGE_SAVED],
4259                                        sizeof(control_page_default));
4260                                 page_index->page_data =
4261                                     (uint8_t *)lun->mode_pages.control_page;
4262                                 break;
4263                         }
4264                         case 0x01:
4265                                 memcpy(&lun->mode_pages.control_ext_page[
4266                                     CTL_PAGE_DEFAULT],
4267                                        &control_ext_page_default,
4268                                        sizeof(control_ext_page_default));
4269                                 memcpy(&lun->mode_pages.control_ext_page[
4270                                     CTL_PAGE_CHANGEABLE],
4271                                        &control_ext_page_changeable,
4272                                        sizeof(control_ext_page_changeable));
4273                                 memcpy(&lun->mode_pages.control_ext_page[
4274                                     CTL_PAGE_SAVED],
4275                                        &control_ext_page_default,
4276                                        sizeof(control_ext_page_default));
4277                                 memcpy(&lun->mode_pages.control_ext_page[
4278                                     CTL_PAGE_CURRENT],
4279                                        &lun->mode_pages.control_ext_page[
4280                                     CTL_PAGE_SAVED],
4281                                        sizeof(control_ext_page_default));
4282                                 page_index->page_data =
4283                                     (uint8_t *)lun->mode_pages.control_ext_page;
4284                                 break;
4285                         default:
4286                                 panic("subpage %#x for page %#x is incorrect!",
4287                                       page_index->subpage, page_code);
4288                         }
4289                         break;
4290                 }
4291                 case SMS_INFO_EXCEPTIONS_PAGE: {
4292                         switch (page_index->subpage) {
4293                         case SMS_SUBPAGE_PAGE_0:
4294                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4295                                        &ie_page_default,
4296                                        sizeof(ie_page_default));
4297                                 memcpy(&lun->mode_pages.ie_page[
4298                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4299                                        sizeof(ie_page_changeable));
4300                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4301                                        &ie_page_default,
4302                                        sizeof(ie_page_default));
4303                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4304                                        &ie_page_default,
4305                                        sizeof(ie_page_default));
4306                                 page_index->page_data =
4307                                         (uint8_t *)lun->mode_pages.ie_page;
4308                                 break;
4309                         case 0x02: {
4310                                 struct ctl_logical_block_provisioning_page *page;
4311
4312                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4313                                        &lbp_page_default,
4314                                        sizeof(lbp_page_default));
4315                                 memcpy(&lun->mode_pages.lbp_page[
4316                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4317                                        sizeof(lbp_page_changeable));
4318                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4319                                        &lbp_page_default,
4320                                        sizeof(lbp_page_default));
4321                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4322                                 value = dnvlist_get_string(lun->be_lun->options,
4323                                     "avail-threshold", NULL);
4324                                 if (value != NULL &&
4325                                     ctl_expand_number(value, &ival) == 0) {
4326                                         page->descr[0].flags |= SLBPPD_ENABLED |
4327                                             SLBPPD_ARMING_DEC;
4328                                         if (lun->be_lun->blocksize)
4329                                                 ival /= lun->be_lun->blocksize;
4330                                         else
4331                                                 ival /= 512;
4332                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4333                                             page->descr[0].count);
4334                                 }
4335                                 value = dnvlist_get_string(lun->be_lun->options,
4336                                     "used-threshold", NULL);
4337                                 if (value != NULL &&
4338                                     ctl_expand_number(value, &ival) == 0) {
4339                                         page->descr[1].flags |= SLBPPD_ENABLED |
4340                                             SLBPPD_ARMING_INC;
4341                                         if (lun->be_lun->blocksize)
4342                                                 ival /= lun->be_lun->blocksize;
4343                                         else
4344                                                 ival /= 512;
4345                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4346                                             page->descr[1].count);
4347                                 }
4348                                 value = dnvlist_get_string(lun->be_lun->options,
4349                                     "pool-avail-threshold", NULL);
4350                                 if (value != NULL &&
4351                                     ctl_expand_number(value, &ival) == 0) {
4352                                         page->descr[2].flags |= SLBPPD_ENABLED |
4353                                             SLBPPD_ARMING_DEC;
4354                                         if (lun->be_lun->blocksize)
4355                                                 ival /= lun->be_lun->blocksize;
4356                                         else
4357                                                 ival /= 512;
4358                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4359                                             page->descr[2].count);
4360                                 }
4361                                 value = dnvlist_get_string(lun->be_lun->options,
4362                                     "pool-used-threshold", NULL);
4363                                 if (value != NULL &&
4364                                     ctl_expand_number(value, &ival) == 0) {
4365                                         page->descr[3].flags |= SLBPPD_ENABLED |
4366                                             SLBPPD_ARMING_INC;
4367                                         if (lun->be_lun->blocksize)
4368                                                 ival /= lun->be_lun->blocksize;
4369                                         else
4370                                                 ival /= 512;
4371                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4372                                             page->descr[3].count);
4373                                 }
4374                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4375                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4376                                        sizeof(lbp_page_default));
4377                                 page_index->page_data =
4378                                         (uint8_t *)lun->mode_pages.lbp_page;
4379                                 break;
4380                         }
4381                         default:
4382                                 panic("subpage %#x for page %#x is incorrect!",
4383                                       page_index->subpage, page_code);
4384                         }
4385                         break;
4386                 }
4387                 case SMS_CDDVD_CAPS_PAGE:{
4388                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4389                             ("subpage %#x for page %#x is incorrect!",
4390                             page_index->subpage, page_code));
4391                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4392                                &cddvd_page_default,
4393                                sizeof(cddvd_page_default));
4394                         memcpy(&lun->mode_pages.cddvd_page[
4395                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4396                                sizeof(cddvd_page_changeable));
4397                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4398                                &cddvd_page_default,
4399                                sizeof(cddvd_page_default));
4400                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4401                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4402                                sizeof(cddvd_page_default));
4403                         page_index->page_data =
4404                                 (uint8_t *)lun->mode_pages.cddvd_page;
4405                         break;
4406                 }
4407                 default:
4408                         panic("invalid page code value %#x", page_code);
4409                 }
4410         }
4411
4412         return (CTL_RETVAL_COMPLETE);
4413 }
4414
4415 static int
4416 ctl_init_log_page_index(struct ctl_lun *lun)
4417 {
4418         struct ctl_page_index *page_index;
4419         int i, j, k, prev;
4420
4421         memcpy(&lun->log_pages.index, log_page_index_template,
4422                sizeof(log_page_index_template));
4423
4424         prev = -1;
4425         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4426
4427                 page_index = &lun->log_pages.index[i];
4428                 if (lun->be_lun->lun_type == T_DIRECT &&
4429                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4430                         continue;
4431                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4432                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4433                         continue;
4434                 if (lun->be_lun->lun_type == T_CDROM &&
4435                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4436                         continue;
4437
4438                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4439                     lun->backend->lun_attr == NULL)
4440                         continue;
4441
4442                 if (page_index->page_code != prev) {
4443                         lun->log_pages.pages_page[j] = page_index->page_code;
4444                         prev = page_index->page_code;
4445                         j++;
4446                 }
4447                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4448                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4449                 k++;
4450         }
4451         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4452         lun->log_pages.index[0].page_len = j;
4453         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4454         lun->log_pages.index[1].page_len = k * 2;
4455         lun->log_pages.index[2].page_data = (uint8_t *)&lun->log_pages.temp_page;
4456         lun->log_pages.index[2].page_len = sizeof(lun->log_pages.temp_page);
4457         lun->log_pages.index[3].page_data = &lun->log_pages.lbp_page[0];
4458         lun->log_pages.index[3].page_len = 12*CTL_NUM_LBP_PARAMS;
4459         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.stat_page;
4460         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.stat_page);
4461         lun->log_pages.index[5].page_data = (uint8_t *)&lun->log_pages.ie_page;
4462         lun->log_pages.index[5].page_len = sizeof(lun->log_pages.ie_page);
4463
4464         return (CTL_RETVAL_COMPLETE);
4465 }
4466
4467 static int
4468 hex2bin(const char *str, uint8_t *buf, int buf_size)
4469 {
4470         int i;
4471         u_char c;
4472
4473         memset(buf, 0, buf_size);
4474         while (isspace(str[0]))
4475                 str++;
4476         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4477                 str += 2;
4478         buf_size *= 2;
4479         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4480                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4481                         str++;
4482                 c = str[i];
4483                 if (isdigit(c))
4484                         c -= '0';
4485                 else if (isalpha(c))
4486                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4487                 else
4488                         break;
4489                 if (c >= 16)
4490                         break;
4491                 if ((i & 1) == 0)
4492                         buf[i / 2] |= (c << 4);
4493                 else
4494                         buf[i / 2] |= c;
4495         }
4496         return ((i + 1) / 2);
4497 }
4498
4499 /*
4500  * LUN allocation.
4501  *
4502  * Requirements:
4503  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4504  *   wants us to allocate the LUN and he can block.
4505  * - ctl_softc is always set
4506  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4507  *
4508  * Returns 0 for success, non-zero (errno) for failure.
4509  */
4510 static int
4511 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4512               struct ctl_be_lun *const be_lun)
4513 {
4514         struct ctl_lun *nlun, *lun;
4515         struct scsi_vpd_id_descriptor *desc;
4516         struct scsi_vpd_id_t10 *t10id;
4517         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4518         int lun_number, lun_malloced;
4519         int devidlen, idlen1, idlen2 = 0, len;
4520
4521         if (be_lun == NULL)
4522                 return (EINVAL);
4523
4524         /*
4525          * We currently only support Direct Access or Processor LUN types.
4526          */
4527         switch (be_lun->lun_type) {
4528         case T_DIRECT:
4529         case T_PROCESSOR:
4530         case T_CDROM:
4531                 break;
4532         case T_SEQUENTIAL:
4533         case T_CHANGER:
4534         default:
4535                 be_lun->lun_config_status(be_lun->be_lun,
4536                                           CTL_LUN_CONFIG_FAILURE);
4537                 break;
4538         }
4539         if (ctl_lun == NULL) {
4540                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4541                 lun_malloced = 1;
4542         } else {
4543                 lun_malloced = 0;
4544                 lun = ctl_lun;
4545         }
4546
4547         memset(lun, 0, sizeof(*lun));
4548         if (lun_malloced)
4549                 lun->flags = CTL_LUN_MALLOCED;
4550
4551         lun->pending_sense = malloc(sizeof(struct scsi_sense_data *) *
4552             ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO);
4553         lun->pending_ua = malloc(sizeof(ctl_ua_type *) * ctl_max_ports,
4554             M_DEVBUF, M_WAITOK | M_ZERO);
4555         lun->pr_keys = malloc(sizeof(uint64_t *) * ctl_max_ports,
4556             M_DEVBUF, M_WAITOK | M_ZERO);
4557
4558         /* Generate LUN ID. */
4559         devidlen = max(CTL_DEVID_MIN_LEN,
4560             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4561         idlen1 = sizeof(*t10id) + devidlen;
4562         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4563         scsiname = dnvlist_get_string(be_lun->options, "scsiname", NULL);
4564         if (scsiname != NULL) {
4565                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4566                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4567         }
4568         eui = dnvlist_get_string(be_lun->options, "eui", NULL);
4569         if (eui != NULL) {
4570                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4571         }
4572         naa = dnvlist_get_string(be_lun->options, "naa", NULL);
4573         if (naa != NULL) {
4574                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4575         }
4576         uuid = dnvlist_get_string(be_lun->options, "uuid", NULL);
4577         if (uuid != NULL) {
4578                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4579         }
4580         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4581             M_CTL, M_WAITOK | M_ZERO);
4582         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4583         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4584         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4585         desc->length = idlen1;
4586         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4587         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4588         if ((vendor = dnvlist_get_string(be_lun->options, "vendor", NULL)) == NULL) {
4589                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4590         } else {
4591                 strncpy(t10id->vendor, vendor,
4592                     min(sizeof(t10id->vendor), strlen(vendor)));
4593         }
4594         strncpy((char *)t10id->vendor_spec_id,
4595             (char *)be_lun->device_id, devidlen);
4596         if (scsiname != NULL) {
4597                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4598                     desc->length);
4599                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4600                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4601                     SVPD_ID_TYPE_SCSI_NAME;
4602                 desc->length = idlen2;
4603                 strlcpy(desc->identifier, scsiname, idlen2);
4604         }
4605         if (eui != NULL) {
4606                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4607                     desc->length);
4608                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4609                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4610                     SVPD_ID_TYPE_EUI64;
4611                 desc->length = hex2bin(eui, desc->identifier, 16);
4612                 desc->length = desc->length > 12 ? 16 :
4613                     (desc->length > 8 ? 12 : 8);
4614                 len -= 16 - desc->length;
4615         }
4616         if (naa != NULL) {
4617                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4618                     desc->length);
4619                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4620                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4621                     SVPD_ID_TYPE_NAA;
4622                 desc->length = hex2bin(naa, desc->identifier, 16);
4623                 desc->length = desc->length > 8 ? 16 : 8;
4624                 len -= 16 - desc->length;
4625         }
4626         if (uuid != NULL) {
4627                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4628                     desc->length);
4629                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4630                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4631                     SVPD_ID_TYPE_UUID;
4632                 desc->identifier[0] = 0x10;
4633                 hex2bin(uuid, &desc->identifier[2], 16);
4634                 desc->length = 18;
4635         }
4636         lun->lun_devid->len = len;
4637
4638         mtx_lock(&ctl_softc->ctl_lock);
4639         /*
4640          * See if the caller requested a particular LUN number.  If so, see
4641          * if it is available.  Otherwise, allocate the first available LUN.
4642          */
4643         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4644                 if ((be_lun->req_lun_id > (ctl_max_luns - 1))
4645                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4646                         mtx_unlock(&ctl_softc->ctl_lock);
4647                         if (be_lun->req_lun_id > (ctl_max_luns - 1)) {
4648                                 printf("ctl: requested LUN ID %d is higher "
4649                                        "than ctl_max_luns - 1 (%d)\n",
4650                                        be_lun->req_lun_id, ctl_max_luns - 1);
4651                         } else {
4652                                 /*
4653                                  * XXX KDM return an error, or just assign
4654                                  * another LUN ID in this case??
4655                                  */
4656                                 printf("ctl: requested LUN ID %d is already "
4657                                        "in use\n", be_lun->req_lun_id);
4658                         }
4659 fail:
4660                         free(lun->lun_devid, M_CTL);
4661                         if (lun->flags & CTL_LUN_MALLOCED)
4662                                 free(lun, M_CTL);
4663                         be_lun->lun_config_status(be_lun->be_lun,
4664                                                   CTL_LUN_CONFIG_FAILURE);
4665                         return (ENOSPC);
4666                 }
4667                 lun_number = be_lun->req_lun_id;
4668         } else {
4669                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, ctl_max_luns);
4670                 if (lun_number == -1) {
4671                         mtx_unlock(&ctl_softc->ctl_lock);
4672                         printf("ctl: can't allocate LUN, out of LUNs\n");
4673                         goto fail;
4674                 }
4675         }
4676         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4677         mtx_unlock(&ctl_softc->ctl_lock);
4678
4679         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4680         lun->lun = lun_number;
4681         lun->be_lun = be_lun;
4682         /*
4683          * The processor LUN is always enabled.  Disk LUNs come on line
4684          * disabled, and must be enabled by the backend.
4685          */
4686         lun->flags |= CTL_LUN_DISABLED;
4687         lun->backend = be_lun->be;
4688         be_lun->ctl_lun = lun;
4689         be_lun->lun_id = lun_number;
4690         atomic_add_int(&be_lun->be->num_luns, 1);
4691         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4692                 lun->flags |= CTL_LUN_EJECTED;
4693         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4694                 lun->flags |= CTL_LUN_NO_MEDIA;
4695         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4696                 lun->flags |= CTL_LUN_STOPPED;
4697
4698         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4699                 lun->flags |= CTL_LUN_PRIMARY_SC;
4700
4701         value = dnvlist_get_string(be_lun->options, "removable", NULL);
4702         if (value != NULL) {
4703                 if (strcmp(value, "on") == 0)
4704                         lun->flags |= CTL_LUN_REMOVABLE;
4705         } else if (be_lun->lun_type == T_CDROM)
4706                 lun->flags |= CTL_LUN_REMOVABLE;
4707
4708         lun->ctl_softc = ctl_softc;
4709 #ifdef CTL_TIME_IO
4710         lun->last_busy = getsbinuptime();
4711 #endif
4712         TAILQ_INIT(&lun->ooa_queue);
4713         STAILQ_INIT(&lun->error_list);
4714         lun->ie_reported = 1;
4715         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4716         ctl_tpc_lun_init(lun);
4717         if (lun->flags & CTL_LUN_REMOVABLE) {
4718                 lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4,
4719                     M_CTL, M_WAITOK);
4720         }
4721
4722         /*
4723          * Initialize the mode and log page index.
4724          */
4725         ctl_init_page_index(lun);
4726         ctl_init_log_page_index(lun);
4727
4728         /* Setup statistics gathering */
4729         lun->stats.item = lun_number;
4730
4731         /*
4732          * Now, before we insert this lun on the lun list, set the lun
4733          * inventory changed UA for all other luns.
4734          */
4735         mtx_lock(&ctl_softc->ctl_lock);
4736         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4737                 mtx_lock(&nlun->lun_lock);
4738                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4739                 mtx_unlock(&nlun->lun_lock);
4740         }
4741         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4742         ctl_softc->ctl_luns[lun_number] = lun;
4743         ctl_softc->num_luns++;
4744         mtx_unlock(&ctl_softc->ctl_lock);
4745
4746         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4747         return (0);
4748 }
4749
4750 /*
4751  * Delete a LUN.
4752  * Assumptions:
4753  * - LUN has already been marked invalid and any pending I/O has been taken
4754  *   care of.
4755  */
4756 static int
4757 ctl_free_lun(struct ctl_lun *lun)
4758 {
4759         struct ctl_softc *softc = lun->ctl_softc;
4760         struct ctl_lun *nlun;
4761         int i;
4762
4763         KASSERT(TAILQ_EMPTY(&lun->ooa_queue),
4764             ("Freeing a LUN %p with outstanding I/O!\n", lun));
4765
4766         mtx_lock(&softc->ctl_lock);
4767         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4768         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4769         softc->ctl_luns[lun->lun] = NULL;
4770         softc->num_luns--;
4771         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4772                 mtx_lock(&nlun->lun_lock);
4773                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4774                 mtx_unlock(&nlun->lun_lock);
4775         }
4776         mtx_unlock(&softc->ctl_lock);
4777
4778         /*
4779          * Tell the backend to free resources, if this LUN has a backend.
4780          */
4781         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4782         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4783
4784         lun->ie_reportcnt = UINT32_MAX;
4785         callout_drain(&lun->ie_callout);
4786         ctl_tpc_lun_shutdown(lun);
4787         mtx_destroy(&lun->lun_lock);
4788         free(lun->lun_devid, M_CTL);
4789         for (i = 0; i < ctl_max_ports; i++)
4790                 free(lun->pending_ua[i], M_CTL);
4791         free(lun->pending_ua, M_DEVBUF);
4792         for (i = 0; i < ctl_max_ports; i++)
4793                 free(lun->pr_keys[i], M_CTL);
4794         free(lun->pr_keys, M_DEVBUF);
4795         free(lun->write_buffer, M_CTL);
4796         free(lun->prevent, M_CTL);
4797         if (lun->flags & CTL_LUN_MALLOCED)
4798                 free(lun, M_CTL);
4799
4800         return (0);
4801 }
4802
4803 static void
4804 ctl_create_lun(struct ctl_be_lun *be_lun)
4805 {
4806
4807         /*
4808          * ctl_alloc_lun() should handle all potential failure cases.
4809          */
4810         ctl_alloc_lun(control_softc, NULL, be_lun);
4811 }
4812
4813 int
4814 ctl_add_lun(struct ctl_be_lun *be_lun)
4815 {
4816         struct ctl_softc *softc = control_softc;
4817
4818         mtx_lock(&softc->ctl_lock);
4819         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4820         mtx_unlock(&softc->ctl_lock);
4821         wakeup(&softc->pending_lun_queue);
4822
4823         return (0);
4824 }
4825
4826 int
4827 ctl_enable_lun(struct ctl_be_lun *be_lun)
4828 {
4829         struct ctl_softc *softc;
4830         struct ctl_port *port, *nport;
4831         struct ctl_lun *lun;
4832         int retval;
4833
4834         lun = (struct ctl_lun *)be_lun->ctl_lun;
4835         softc = lun->ctl_softc;
4836
4837         mtx_lock(&softc->ctl_lock);
4838         mtx_lock(&lun->lun_lock);
4839         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4840                 /*
4841                  * eh?  Why did we get called if the LUN is already
4842                  * enabled?
4843                  */
4844                 mtx_unlock(&lun->lun_lock);
4845                 mtx_unlock(&softc->ctl_lock);
4846                 return (0);
4847         }
4848         lun->flags &= ~CTL_LUN_DISABLED;
4849         mtx_unlock(&lun->lun_lock);
4850
4851         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4852                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4853                     port->lun_map != NULL || port->lun_enable == NULL)
4854                         continue;
4855
4856                 /*
4857                  * Drop the lock while we call the FETD's enable routine.
4858                  * This can lead to a callback into CTL (at least in the
4859                  * case of the internal initiator frontend.
4860                  */
4861                 mtx_unlock(&softc->ctl_lock);
4862                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4863                 mtx_lock(&softc->ctl_lock);
4864                 if (retval != 0) {
4865                         printf("%s: FETD %s port %d returned error "
4866                                "%d for lun_enable on lun %jd\n",
4867                                __func__, port->port_name, port->targ_port,
4868                                retval, (intmax_t)lun->lun);
4869                 }
4870         }
4871
4872         mtx_unlock(&softc->ctl_lock);
4873         ctl_isc_announce_lun(lun);
4874
4875         return (0);
4876 }
4877
4878 int
4879 ctl_disable_lun(struct ctl_be_lun *be_lun)
4880 {
4881         struct ctl_softc *softc;
4882         struct ctl_port *port;
4883         struct ctl_lun *lun;
4884         int retval;
4885
4886         lun = (struct ctl_lun *)be_lun->ctl_lun;
4887         softc = lun->ctl_softc;
4888
4889         mtx_lock(&softc->ctl_lock);
4890         mtx_lock(&lun->lun_lock);
4891         if (lun->flags & CTL_LUN_DISABLED) {
4892                 mtx_unlock(&lun->lun_lock);
4893                 mtx_unlock(&softc->ctl_lock);
4894                 return (0);
4895         }
4896         lun->flags |= CTL_LUN_DISABLED;
4897         mtx_unlock(&lun->lun_lock);
4898
4899         STAILQ_FOREACH(port, &softc->port_list, links) {
4900                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4901                     port->lun_map != NULL || port->lun_disable == NULL)
4902                         continue;
4903
4904                 /*
4905                  * Drop the lock before we call the frontend's disable
4906                  * routine, to avoid lock order reversals.
4907                  *
4908                  * XXX KDM what happens if the frontend list changes while
4909                  * we're traversing it?  It's unlikely, but should be handled.
4910                  */
4911                 mtx_unlock(&softc->ctl_lock);
4912                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4913                 mtx_lock(&softc->ctl_lock);
4914                 if (retval != 0) {
4915                         printf("%s: FETD %s port %d returned error "
4916                                "%d for lun_disable on lun %jd\n",
4917                                __func__, port->port_name, port->targ_port,
4918                                retval, (intmax_t)lun->lun);
4919                 }
4920         }
4921
4922         mtx_unlock(&softc->ctl_lock);
4923         ctl_isc_announce_lun(lun);
4924
4925         return (0);
4926 }
4927
4928 int
4929 ctl_start_lun(struct ctl_be_lun *be_lun)
4930 {
4931         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4932
4933         mtx_lock(&lun->lun_lock);
4934         lun->flags &= ~CTL_LUN_STOPPED;
4935         mtx_unlock(&lun->lun_lock);
4936         return (0);
4937 }
4938
4939 int
4940 ctl_stop_lun(struct ctl_be_lun *be_lun)
4941 {
4942         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4943
4944         mtx_lock(&lun->lun_lock);
4945         lun->flags |= CTL_LUN_STOPPED;
4946         mtx_unlock(&lun->lun_lock);
4947         return (0);
4948 }
4949
4950 int
4951 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4952 {
4953         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4954
4955         mtx_lock(&lun->lun_lock);
4956         lun->flags |= CTL_LUN_NO_MEDIA;
4957         mtx_unlock(&lun->lun_lock);
4958         return (0);
4959 }
4960
4961 int
4962 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4963 {
4964         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4965         union ctl_ha_msg msg;
4966
4967         mtx_lock(&lun->lun_lock);
4968         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4969         if (lun->flags & CTL_LUN_REMOVABLE)
4970                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4971         mtx_unlock(&lun->lun_lock);
4972         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4973             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4974                 bzero(&msg.ua, sizeof(msg.ua));
4975                 msg.hdr.msg_type = CTL_MSG_UA;
4976                 msg.hdr.nexus.initid = -1;
4977                 msg.hdr.nexus.targ_port = -1;
4978                 msg.hdr.nexus.targ_lun = lun->lun;
4979                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4980                 msg.ua.ua_all = 1;
4981                 msg.ua.ua_set = 1;
4982                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4983                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4984                     M_WAITOK);
4985         }
4986         return (0);
4987 }
4988
4989 int
4990 ctl_lun_ejected(struct ctl_be_lun *be_lun)
4991 {
4992         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4993
4994         mtx_lock(&lun->lun_lock);
4995         lun->flags |= CTL_LUN_EJECTED;
4996         mtx_unlock(&lun->lun_lock);
4997         return (0);
4998 }
4999
5000 int
5001 ctl_lun_primary(struct ctl_be_lun *be_lun)
5002 {
5003         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5004
5005         mtx_lock(&lun->lun_lock);
5006         lun->flags |= CTL_LUN_PRIMARY_SC;
5007         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5008         mtx_unlock(&lun->lun_lock);
5009         ctl_isc_announce_lun(lun);
5010         return (0);
5011 }
5012
5013 int
5014 ctl_lun_secondary(struct ctl_be_lun *be_lun)
5015 {
5016         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5017
5018         mtx_lock(&lun->lun_lock);
5019         lun->flags &= ~CTL_LUN_PRIMARY_SC;
5020         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5021         mtx_unlock(&lun->lun_lock);
5022         ctl_isc_announce_lun(lun);
5023         return (0);
5024 }
5025
5026 int
5027 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
5028 {
5029         struct ctl_lun *lun;
5030
5031         lun = (struct ctl_lun *)be_lun->ctl_lun;
5032
5033         mtx_lock(&lun->lun_lock);
5034
5035         /*
5036          * The LUN needs to be disabled before it can be marked invalid.
5037          */
5038         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
5039                 mtx_unlock(&lun->lun_lock);
5040                 return (-1);
5041         }
5042         /*
5043          * Mark the LUN invalid.
5044          */
5045         lun->flags |= CTL_LUN_INVALID;
5046
5047         /*
5048          * If there is nothing in the OOA queue, go ahead and free the LUN.
5049          * If we have something in the OOA queue, we'll free it when the
5050          * last I/O completes.
5051          */
5052         if (TAILQ_EMPTY(&lun->ooa_queue)) {
5053                 mtx_unlock(&lun->lun_lock);
5054                 ctl_free_lun(lun);
5055         } else
5056                 mtx_unlock(&lun->lun_lock);
5057
5058         return (0);
5059 }
5060
5061 void
5062 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5063 {
5064         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5065         union ctl_ha_msg msg;
5066
5067         mtx_lock(&lun->lun_lock);
5068         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
5069         mtx_unlock(&lun->lun_lock);
5070         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5071                 /* Send msg to other side. */
5072                 bzero(&msg.ua, sizeof(msg.ua));
5073                 msg.hdr.msg_type = CTL_MSG_UA;
5074                 msg.hdr.nexus.initid = -1;
5075                 msg.hdr.nexus.targ_port = -1;
5076                 msg.hdr.nexus.targ_lun = lun->lun;
5077                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5078                 msg.ua.ua_all = 1;
5079                 msg.ua.ua_set = 1;
5080                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
5081                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5082                     M_WAITOK);
5083         }
5084 }
5085
5086 /*
5087  * Backend "memory move is complete" callback for requests that never
5088  * make it down to say RAIDCore's configuration code.
5089  */
5090 int
5091 ctl_config_move_done(union ctl_io *io)
5092 {
5093         int retval;
5094
5095         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5096         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5097             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5098
5099         if ((io->io_hdr.port_status != 0) &&
5100             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5101              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5102                 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
5103                     /*retry_count*/ io->io_hdr.port_status);
5104         } else if (io->scsiio.kern_data_resid != 0 &&
5105             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
5106             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5107              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5108                 ctl_set_invalid_field_ciu(&io->scsiio);
5109         }
5110
5111         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5112                 ctl_data_print(io);
5113         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5114             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5115              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5116             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5117                 /*
5118                  * XXX KDM just assuming a single pointer here, and not a
5119                  * S/G list.  If we start using S/G lists for config data,
5120                  * we'll need to know how to clean them up here as well.
5121                  */
5122                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5123                         free(io->scsiio.kern_data_ptr, M_CTL);
5124                 ctl_done(io);
5125                 retval = CTL_RETVAL_COMPLETE;
5126         } else {
5127                 /*
5128                  * XXX KDM now we need to continue data movement.  Some
5129                  * options:
5130                  * - call ctl_scsiio() again?  We don't do this for data
5131                  *   writes, because for those at least we know ahead of
5132                  *   time where the write will go and how long it is.  For
5133                  *   config writes, though, that information is largely
5134                  *   contained within the write itself, thus we need to
5135                  *   parse out the data again.
5136                  *
5137                  * - Call some other function once the data is in?
5138                  */
5139
5140                 /*
5141                  * XXX KDM call ctl_scsiio() again for now, and check flag
5142                  * bits to see whether we're allocated or not.
5143                  */
5144                 retval = ctl_scsiio(&io->scsiio);
5145         }
5146         return (retval);
5147 }
5148
5149 /*
5150  * This gets called by a backend driver when it is done with a
5151  * data_submit method.
5152  */
5153 void
5154 ctl_data_submit_done(union ctl_io *io)
5155 {
5156         /*
5157          * If the IO_CONT flag is set, we need to call the supplied
5158          * function to continue processing the I/O, instead of completing
5159          * the I/O just yet.
5160          *
5161          * If there is an error, though, we don't want to keep processing.
5162          * Instead, just send status back to the initiator.
5163          */
5164         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5165             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5166             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5167              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5168                 io->scsiio.io_cont(io);
5169                 return;
5170         }
5171         ctl_done(io);
5172 }
5173
5174 /*
5175  * This gets called by a backend driver when it is done with a
5176  * configuration write.
5177  */
5178 void
5179 ctl_config_write_done(union ctl_io *io)
5180 {
5181         uint8_t *buf;
5182
5183         /*
5184          * If the IO_CONT flag is set, we need to call the supplied
5185          * function to continue processing the I/O, instead of completing
5186          * the I/O just yet.
5187          *
5188          * If there is an error, though, we don't want to keep processing.
5189          * Instead, just send status back to the initiator.
5190          */
5191         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5192             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5193             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5194              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5195                 io->scsiio.io_cont(io);
5196                 return;
5197         }
5198         /*
5199          * Since a configuration write can be done for commands that actually
5200          * have data allocated, like write buffer, and commands that have
5201          * no data, like start/stop unit, we need to check here.
5202          */
5203         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5204                 buf = io->scsiio.kern_data_ptr;
5205         else
5206                 buf = NULL;
5207         ctl_done(io);
5208         if (buf)
5209                 free(buf, M_CTL);
5210 }
5211
5212 void
5213 ctl_config_read_done(union ctl_io *io)
5214 {
5215         uint8_t *buf;
5216
5217         /*
5218          * If there is some error -- we are done, skip data transfer.
5219          */
5220         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5221             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5222              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5223                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5224                         buf = io->scsiio.kern_data_ptr;
5225                 else
5226                         buf = NULL;
5227                 ctl_done(io);
5228                 if (buf)
5229                         free(buf, M_CTL);
5230                 return;
5231         }
5232
5233         /*
5234          * If the IO_CONT flag is set, we need to call the supplied
5235          * function to continue processing the I/O, instead of completing
5236          * the I/O just yet.
5237          */
5238         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5239                 io->scsiio.io_cont(io);
5240                 return;
5241         }
5242
5243         ctl_datamove(io);
5244 }
5245
5246 /*
5247  * SCSI release command.
5248  */
5249 int
5250 ctl_scsi_release(struct ctl_scsiio *ctsio)
5251 {
5252         struct ctl_lun *lun = CTL_LUN(ctsio);
5253         uint32_t residx;
5254
5255         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5256
5257         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5258
5259         /*
5260          * XXX KDM right now, we only support LUN reservation.  We don't
5261          * support 3rd party reservations, or extent reservations, which
5262          * might actually need the parameter list.  If we've gotten this
5263          * far, we've got a LUN reservation.  Anything else got kicked out
5264          * above.  So, according to SPC, ignore the length.
5265          */
5266
5267         mtx_lock(&lun->lun_lock);
5268
5269         /*
5270          * According to SPC, it is not an error for an intiator to attempt
5271          * to release a reservation on a LUN that isn't reserved, or that
5272          * is reserved by another initiator.  The reservation can only be
5273          * released, though, by the initiator who made it or by one of
5274          * several reset type events.
5275          */
5276         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5277                         lun->flags &= ~CTL_LUN_RESERVED;
5278
5279         mtx_unlock(&lun->lun_lock);
5280
5281         ctl_set_success(ctsio);
5282         ctl_done((union ctl_io *)ctsio);
5283         return (CTL_RETVAL_COMPLETE);
5284 }
5285
5286 int
5287 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5288 {
5289         struct ctl_lun *lun = CTL_LUN(ctsio);
5290         uint32_t residx;
5291
5292         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5293
5294         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5295
5296         /*
5297          * XXX KDM right now, we only support LUN reservation.  We don't
5298          * support 3rd party reservations, or extent reservations, which
5299          * might actually need the parameter list.  If we've gotten this
5300          * far, we've got a LUN reservation.  Anything else got kicked out
5301          * above.  So, according to SPC, ignore the length.
5302          */
5303
5304         mtx_lock(&lun->lun_lock);
5305         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5306                 ctl_set_reservation_conflict(ctsio);
5307                 goto bailout;
5308         }
5309
5310         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5311         if (lun->flags & CTL_LUN_PR_RESERVED) {
5312                 ctl_set_success(ctsio);
5313                 goto bailout;
5314         }
5315
5316         lun->flags |= CTL_LUN_RESERVED;
5317         lun->res_idx = residx;
5318         ctl_set_success(ctsio);
5319
5320 bailout:
5321         mtx_unlock(&lun->lun_lock);
5322         ctl_done((union ctl_io *)ctsio);
5323         return (CTL_RETVAL_COMPLETE);
5324 }
5325
5326 int
5327 ctl_start_stop(struct ctl_scsiio *ctsio)
5328 {
5329         struct ctl_lun *lun = CTL_LUN(ctsio);
5330         struct scsi_start_stop_unit *cdb;
5331         int retval;
5332
5333         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5334
5335         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5336
5337         if ((cdb->how & SSS_PC_MASK) == 0) {
5338                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5339                     (cdb->how & SSS_START) == 0) {
5340                         uint32_t residx;
5341
5342                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5343                         if (ctl_get_prkey(lun, residx) == 0 ||
5344                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5345
5346                                 ctl_set_reservation_conflict(ctsio);
5347                                 ctl_done((union ctl_io *)ctsio);
5348                                 return (CTL_RETVAL_COMPLETE);
5349                         }
5350                 }
5351
5352                 if ((cdb->how & SSS_LOEJ) &&
5353                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5354                         ctl_set_invalid_field(ctsio,
5355                                               /*sks_valid*/ 1,
5356                                               /*command*/ 1,
5357                                               /*field*/ 4,
5358                                               /*bit_valid*/ 1,
5359                                               /*bit*/ 1);
5360                         ctl_done((union ctl_io *)ctsio);
5361                         return (CTL_RETVAL_COMPLETE);
5362                 }
5363
5364                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5365                     lun->prevent_count > 0) {
5366                         /* "Medium removal prevented" */
5367                         ctl_set_sense(ctsio, /*current_error*/ 1,
5368                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5369                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5370                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5371                         ctl_done((union ctl_io *)ctsio);
5372                         return (CTL_RETVAL_COMPLETE);
5373                 }
5374         }
5375
5376         retval = lun->backend->config_write((union ctl_io *)ctsio);
5377         return (retval);
5378 }
5379
5380 int
5381 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5382 {
5383         struct ctl_lun *lun = CTL_LUN(ctsio);
5384         struct scsi_prevent *cdb;
5385         int retval;
5386         uint32_t initidx;
5387
5388         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5389
5390         cdb = (struct scsi_prevent *)ctsio->cdb;
5391
5392         if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) {
5393                 ctl_set_invalid_opcode(ctsio);
5394                 ctl_done((union ctl_io *)ctsio);
5395                 return (CTL_RETVAL_COMPLETE);
5396         }
5397
5398         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5399         mtx_lock(&lun->lun_lock);
5400         if ((cdb->how & PR_PREVENT) &&
5401             ctl_is_set(lun->prevent, initidx) == 0) {
5402                 ctl_set_mask(lun->prevent, initidx);
5403                 lun->prevent_count++;
5404         } else if ((cdb->how & PR_PREVENT) == 0 &&
5405             ctl_is_set(lun->prevent, initidx)) {
5406                 ctl_clear_mask(lun->prevent, initidx);
5407                 lun->prevent_count--;
5408         }
5409         mtx_unlock(&lun->lun_lock);
5410         retval = lun->backend->config_write((union ctl_io *)ctsio);
5411         return (retval);
5412 }
5413
5414 /*
5415  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5416  * we don't really do anything with the LBA and length fields if the user
5417  * passes them in.  Instead we'll just flush out the cache for the entire
5418  * LUN.
5419  */
5420 int
5421 ctl_sync_cache(struct ctl_scsiio *ctsio)
5422 {
5423         struct ctl_lun *lun = CTL_LUN(ctsio);
5424         struct ctl_lba_len_flags *lbalen;
5425         uint64_t starting_lba;
5426         uint32_t block_count;
5427         int retval;
5428         uint8_t byte2;
5429
5430         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5431
5432         retval = 0;
5433
5434         switch (ctsio->cdb[0]) {
5435         case SYNCHRONIZE_CACHE: {
5436                 struct scsi_sync_cache *cdb;
5437                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5438
5439                 starting_lba = scsi_4btoul(cdb->begin_lba);
5440                 block_count = scsi_2btoul(cdb->lb_count);
5441                 byte2 = cdb->byte2;
5442                 break;
5443         }
5444         case SYNCHRONIZE_CACHE_16: {
5445                 struct scsi_sync_cache_16 *cdb;
5446                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5447
5448                 starting_lba = scsi_8btou64(cdb->begin_lba);
5449                 block_count = scsi_4btoul(cdb->lb_count);
5450                 byte2 = cdb->byte2;
5451                 break;
5452         }
5453         default:
5454                 ctl_set_invalid_opcode(ctsio);
5455                 ctl_done((union ctl_io *)ctsio);
5456                 goto bailout;
5457                 break; /* NOTREACHED */
5458         }
5459
5460         /*
5461          * We check the LBA and length, but don't do anything with them.
5462          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5463          * get flushed.  This check will just help satisfy anyone who wants
5464          * to see an error for an out of range LBA.
5465          */
5466         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5467                 ctl_set_lba_out_of_range(ctsio,
5468                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5469                 ctl_done((union ctl_io *)ctsio);
5470                 goto bailout;
5471         }
5472
5473         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5474         lbalen->lba = starting_lba;
5475         lbalen->len = block_count;
5476         lbalen->flags = byte2;
5477         retval = lun->backend->config_write((union ctl_io *)ctsio);
5478
5479 bailout:
5480         return (retval);
5481 }
5482
5483 int
5484 ctl_format(struct ctl_scsiio *ctsio)
5485 {
5486         struct scsi_format *cdb;
5487         int length, defect_list_len;
5488
5489         CTL_DEBUG_PRINT(("ctl_format\n"));
5490
5491         cdb = (struct scsi_format *)ctsio->cdb;
5492
5493         length = 0;
5494         if (cdb->byte2 & SF_FMTDATA) {
5495                 if (cdb->byte2 & SF_LONGLIST)
5496                         length = sizeof(struct scsi_format_header_long);
5497                 else
5498                         length = sizeof(struct scsi_format_header_short);
5499         }
5500
5501         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5502          && (length > 0)) {
5503                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5504                 ctsio->kern_data_len = length;
5505                 ctsio->kern_total_len = length;
5506                 ctsio->kern_rel_offset = 0;
5507                 ctsio->kern_sg_entries = 0;
5508                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5509                 ctsio->be_move_done = ctl_config_move_done;
5510                 ctl_datamove((union ctl_io *)ctsio);
5511
5512                 return (CTL_RETVAL_COMPLETE);
5513         }
5514
5515         defect_list_len = 0;
5516
5517         if (cdb->byte2 & SF_FMTDATA) {
5518                 if (cdb->byte2 & SF_LONGLIST) {
5519                         struct scsi_format_header_long *header;
5520
5521                         header = (struct scsi_format_header_long *)
5522                                 ctsio->kern_data_ptr;
5523
5524                         defect_list_len = scsi_4btoul(header->defect_list_len);
5525                         if (defect_list_len != 0) {
5526                                 ctl_set_invalid_field(ctsio,
5527                                                       /*sks_valid*/ 1,
5528                                                       /*command*/ 0,
5529                                                       /*field*/ 2,
5530                                                       /*bit_valid*/ 0,
5531                                                       /*bit*/ 0);
5532                                 goto bailout;
5533                         }
5534                 } else {
5535                         struct scsi_format_header_short *header;
5536
5537                         header = (struct scsi_format_header_short *)
5538                                 ctsio->kern_data_ptr;
5539
5540                         defect_list_len = scsi_2btoul(header->defect_list_len);
5541                         if (defect_list_len != 0) {
5542                                 ctl_set_invalid_field(ctsio,
5543                                                       /*sks_valid*/ 1,
5544                                                       /*command*/ 0,
5545                                                       /*field*/ 2,
5546                                                       /*bit_valid*/ 0,
5547                                                       /*bit*/ 0);
5548                                 goto bailout;
5549                         }
5550                 }
5551         }
5552
5553         ctl_set_success(ctsio);
5554 bailout:
5555
5556         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5557                 free(ctsio->kern_data_ptr, M_CTL);
5558                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5559         }
5560
5561         ctl_done((union ctl_io *)ctsio);
5562         return (CTL_RETVAL_COMPLETE);
5563 }
5564
5565 int
5566 ctl_read_buffer(struct ctl_scsiio *ctsio)
5567 {
5568         struct ctl_lun *lun = CTL_LUN(ctsio);
5569         uint64_t buffer_offset;
5570         uint32_t len;
5571         uint8_t byte2;
5572         static uint8_t descr[4];
5573         static uint8_t echo_descr[4] = { 0 };
5574
5575         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5576
5577         switch (ctsio->cdb[0]) {
5578         case READ_BUFFER: {
5579                 struct scsi_read_buffer *cdb;
5580
5581                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5582                 buffer_offset = scsi_3btoul(cdb->offset);
5583                 len = scsi_3btoul(cdb->length);
5584                 byte2 = cdb->byte2;
5585                 break;
5586         }
5587         case READ_BUFFER_16: {
5588                 struct scsi_read_buffer_16 *cdb;
5589
5590                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5591                 buffer_offset = scsi_8btou64(cdb->offset);
5592                 len = scsi_4btoul(cdb->length);
5593                 byte2 = cdb->byte2;
5594                 break;
5595         }
5596         default: /* This shouldn't happen. */
5597                 ctl_set_invalid_opcode(ctsio);
5598                 ctl_done((union ctl_io *)ctsio);
5599                 return (CTL_RETVAL_COMPLETE);
5600         }
5601
5602         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5603             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5604                 ctl_set_invalid_field(ctsio,
5605                                       /*sks_valid*/ 1,
5606                                       /*command*/ 1,
5607                                       /*field*/ 6,
5608                                       /*bit_valid*/ 0,
5609                                       /*bit*/ 0);
5610                 ctl_done((union ctl_io *)ctsio);
5611                 return (CTL_RETVAL_COMPLETE);
5612         }
5613
5614         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5615                 descr[0] = 0;
5616                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5617                 ctsio->kern_data_ptr = descr;
5618                 len = min(len, sizeof(descr));
5619         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5620                 ctsio->kern_data_ptr = echo_descr;
5621                 len = min(len, sizeof(echo_descr));
5622         } else {
5623                 if (lun->write_buffer == NULL) {
5624                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5625                             M_CTL, M_WAITOK);
5626                 }
5627                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5628         }
5629         ctsio->kern_data_len = len;
5630         ctsio->kern_total_len = len;
5631         ctsio->kern_rel_offset = 0;
5632         ctsio->kern_sg_entries = 0;
5633         ctl_set_success(ctsio);
5634         ctsio->be_move_done = ctl_config_move_done;
5635         ctl_datamove((union ctl_io *)ctsio);
5636         return (CTL_RETVAL_COMPLETE);
5637 }
5638
5639 int
5640 ctl_write_buffer(struct ctl_scsiio *ctsio)
5641 {
5642         struct ctl_lun *lun = CTL_LUN(ctsio);
5643         struct scsi_write_buffer *cdb;
5644         int buffer_offset, len;
5645
5646         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5647
5648         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5649
5650         len = scsi_3btoul(cdb->length);
5651         buffer_offset = scsi_3btoul(cdb->offset);
5652
5653         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5654                 ctl_set_invalid_field(ctsio,
5655                                       /*sks_valid*/ 1,
5656                                       /*command*/ 1,
5657                                       /*field*/ 6,
5658                                       /*bit_valid*/ 0,
5659                                       /*bit*/ 0);
5660                 ctl_done((union ctl_io *)ctsio);
5661                 return (CTL_RETVAL_COMPLETE);
5662         }
5663
5664         /*
5665          * If we've got a kernel request that hasn't been malloced yet,
5666          * malloc it and tell the caller the data buffer is here.
5667          */
5668         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5669                 if (lun->write_buffer == NULL) {
5670                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5671                             M_CTL, M_WAITOK);
5672                 }
5673                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5674                 ctsio->kern_data_len = len;
5675                 ctsio->kern_total_len = len;
5676                 ctsio->kern_rel_offset = 0;
5677                 ctsio->kern_sg_entries = 0;
5678                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5679                 ctsio->be_move_done = ctl_config_move_done;
5680                 ctl_datamove((union ctl_io *)ctsio);
5681
5682                 return (CTL_RETVAL_COMPLETE);
5683         }
5684
5685         ctl_set_success(ctsio);
5686         ctl_done((union ctl_io *)ctsio);
5687         return (CTL_RETVAL_COMPLETE);
5688 }
5689
5690 static int
5691 ctl_write_same_cont(union ctl_io *io)
5692 {
5693         struct ctl_lun *lun = CTL_LUN(io);
5694         struct ctl_scsiio *ctsio;
5695         struct ctl_lba_len_flags *lbalen;
5696         int retval;
5697
5698         ctsio = &io->scsiio;
5699         ctsio->io_hdr.status = CTL_STATUS_NONE;
5700         lbalen = (struct ctl_lba_len_flags *)
5701             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5702         lbalen->lba += lbalen->len;
5703         if ((lun->be_lun->maxlba + 1) - lbalen->lba <= UINT32_MAX) {
5704                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
5705                 lbalen->len = (lun->be_lun->maxlba + 1) - lbalen->lba;
5706         }
5707
5708         CTL_DEBUG_PRINT(("ctl_write_same_cont: calling config_write()\n"));
5709         retval = lun->backend->config_write((union ctl_io *)ctsio);
5710         return (retval);
5711 }
5712
5713 int
5714 ctl_write_same(struct ctl_scsiio *ctsio)
5715 {
5716         struct ctl_lun *lun = CTL_LUN(ctsio);
5717         struct ctl_lba_len_flags *lbalen;
5718         const char *val;
5719         uint64_t lba, ival;
5720         uint32_t num_blocks;
5721         int len, retval;
5722         uint8_t byte2;
5723
5724         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5725
5726         switch (ctsio->cdb[0]) {
5727         case WRITE_SAME_10: {
5728                 struct scsi_write_same_10 *cdb;
5729
5730                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5731
5732                 lba = scsi_4btoul(cdb->addr);
5733                 num_blocks = scsi_2btoul(cdb->length);
5734                 byte2 = cdb->byte2;
5735                 break;
5736         }
5737         case WRITE_SAME_16: {
5738                 struct scsi_write_same_16 *cdb;
5739
5740                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5741
5742                 lba = scsi_8btou64(cdb->addr);
5743                 num_blocks = scsi_4btoul(cdb->length);
5744                 byte2 = cdb->byte2;
5745                 break;
5746         }
5747         default:
5748                 /*
5749                  * We got a command we don't support.  This shouldn't
5750                  * happen, commands should be filtered out above us.
5751                  */
5752                 ctl_set_invalid_opcode(ctsio);
5753                 ctl_done((union ctl_io *)ctsio);
5754
5755                 return (CTL_RETVAL_COMPLETE);
5756                 break; /* NOTREACHED */
5757         }
5758
5759         /* ANCHOR flag can be used only together with UNMAP */
5760         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5761                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5762                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5763                 ctl_done((union ctl_io *)ctsio);
5764                 return (CTL_RETVAL_COMPLETE);
5765         }
5766
5767         /*
5768          * The first check is to make sure we're in bounds, the second
5769          * check is to catch wrap-around problems.  If the lba + num blocks
5770          * is less than the lba, then we've wrapped around and the block
5771          * range is invalid anyway.
5772          */
5773         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5774          || ((lba + num_blocks) < lba)) {
5775                 ctl_set_lba_out_of_range(ctsio,
5776                     MAX(lba, lun->be_lun->maxlba + 1));
5777                 ctl_done((union ctl_io *)ctsio);
5778                 return (CTL_RETVAL_COMPLETE);
5779         }
5780
5781         /* Zero number of blocks means "to the last logical block" */
5782         if (num_blocks == 0) {
5783                 ival = UINT64_MAX;
5784                 val = dnvlist_get_string(lun->be_lun->options,
5785                     "write_same_max_lba", NULL);
5786                 if (val != NULL)
5787                         ctl_expand_number(val, &ival);
5788                 if ((lun->be_lun->maxlba + 1) - lba > ival) {
5789                         ctl_set_invalid_field(ctsio,
5790                             /*sks_valid*/ 1, /*command*/ 1,
5791                             /*field*/ ctsio->cdb[0] == WRITE_SAME_10 ? 7 : 10,
5792                             /*bit_valid*/ 0, /*bit*/ 0);
5793                         ctl_done((union ctl_io *)ctsio);
5794                         return (CTL_RETVAL_COMPLETE);
5795                 }
5796                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5797                         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
5798                         ctsio->io_cont = ctl_write_same_cont;
5799                         num_blocks = 1 << 31;
5800                 } else
5801                         num_blocks = (lun->be_lun->maxlba + 1) - lba;
5802         }
5803
5804         len = lun->be_lun->blocksize;
5805
5806         /*
5807          * If we've got a kernel request that hasn't been malloced yet,
5808          * malloc it and tell the caller the data buffer is here.
5809          */
5810         if ((byte2 & SWS_NDOB) == 0 &&
5811             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5812                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5813                 ctsio->kern_data_len = len;
5814                 ctsio->kern_total_len = len;
5815                 ctsio->kern_rel_offset = 0;
5816                 ctsio->kern_sg_entries = 0;
5817                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5818                 ctsio->be_move_done = ctl_config_move_done;
5819                 ctl_datamove((union ctl_io *)ctsio);
5820
5821                 return (CTL_RETVAL_COMPLETE);
5822         }
5823
5824         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5825         lbalen->lba = lba;
5826         lbalen->len = num_blocks;
5827         lbalen->flags = byte2;
5828         retval = lun->backend->config_write((union ctl_io *)ctsio);
5829
5830         return (retval);
5831 }
5832
5833 int
5834 ctl_unmap(struct ctl_scsiio *ctsio)
5835 {
5836         struct ctl_lun *lun = CTL_LUN(ctsio);
5837         struct scsi_unmap *cdb;
5838         struct ctl_ptr_len_flags *ptrlen;
5839         struct scsi_unmap_header *hdr;
5840         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5841         uint64_t lba;
5842         uint32_t num_blocks;
5843         int len, retval;
5844         uint8_t byte2;
5845
5846         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5847
5848         cdb = (struct scsi_unmap *)ctsio->cdb;
5849         len = scsi_2btoul(cdb->length);
5850         byte2 = cdb->byte2;
5851
5852         /*
5853          * If we've got a kernel request that hasn't been malloced yet,
5854          * malloc it and tell the caller the data buffer is here.
5855          */
5856         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5857                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5858                 ctsio->kern_data_len = len;
5859                 ctsio->kern_total_len = len;
5860                 ctsio->kern_rel_offset = 0;
5861                 ctsio->kern_sg_entries = 0;
5862                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5863                 ctsio->be_move_done = ctl_config_move_done;
5864                 ctl_datamove((union ctl_io *)ctsio);
5865
5866                 return (CTL_RETVAL_COMPLETE);
5867         }
5868
5869         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5870         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5871         if (len < sizeof (*hdr) ||
5872             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5873             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5874             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5875                 ctl_set_invalid_field(ctsio,
5876                                       /*sks_valid*/ 0,
5877                                       /*command*/ 0,
5878                                       /*field*/ 0,
5879                                       /*bit_valid*/ 0,
5880                                       /*bit*/ 0);
5881                 goto done;
5882         }
5883         len = scsi_2btoul(hdr->desc_length);
5884         buf = (struct scsi_unmap_desc *)(hdr + 1);
5885         end = buf + len / sizeof(*buf);
5886
5887         endnz = buf;
5888         for (range = buf; range < end; range++) {
5889                 lba = scsi_8btou64(range->lba);
5890                 num_blocks = scsi_4btoul(range->length);
5891                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5892                  || ((lba + num_blocks) < lba)) {
5893                         ctl_set_lba_out_of_range(ctsio,
5894                             MAX(lba, lun->be_lun->maxlba + 1));
5895                         ctl_done((union ctl_io *)ctsio);
5896                         return (CTL_RETVAL_COMPLETE);
5897                 }
5898                 if (num_blocks != 0)
5899                         endnz = range + 1;
5900         }
5901
5902         /*
5903          * Block backend can not handle zero last range.
5904          * Filter it out and return if there is nothing left.
5905          */
5906         len = (uint8_t *)endnz - (uint8_t *)buf;
5907         if (len == 0) {
5908                 ctl_set_success(ctsio);
5909                 goto done;
5910         }
5911
5912         mtx_lock(&lun->lun_lock);
5913         ptrlen = (struct ctl_ptr_len_flags *)
5914             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5915         ptrlen->ptr = (void *)buf;
5916         ptrlen->len = len;
5917         ptrlen->flags = byte2;
5918         ctl_try_unblock_others(lun, (union ctl_io *)ctsio, FALSE);
5919         mtx_unlock(&lun->lun_lock);
5920
5921         retval = lun->backend->config_write((union ctl_io *)ctsio);
5922         return (retval);
5923
5924 done:
5925         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5926                 free(ctsio->kern_data_ptr, M_CTL);
5927                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5928         }
5929         ctl_done((union ctl_io *)ctsio);
5930         return (CTL_RETVAL_COMPLETE);
5931 }
5932
5933 int
5934 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5935                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5936 {
5937         struct ctl_lun *lun = CTL_LUN(ctsio);
5938         uint8_t *current_cp;
5939         int set_ua;
5940         uint32_t initidx;
5941
5942         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5943         set_ua = 0;
5944
5945         current_cp = (page_index->page_data + (page_index->page_len *
5946             CTL_PAGE_CURRENT));
5947
5948         mtx_lock(&lun->lun_lock);
5949         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5950                 memcpy(current_cp, page_ptr, page_index->page_len);
5951                 set_ua = 1;
5952         }
5953         if (set_ua != 0)
5954                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5955         mtx_unlock(&lun->lun_lock);
5956         if (set_ua) {
5957                 ctl_isc_announce_mode(lun,
5958                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5959                     page_index->page_code, page_index->subpage);
5960         }
5961         return (CTL_RETVAL_COMPLETE);
5962 }
5963
5964 static void
5965 ctl_ie_timer(void *arg)
5966 {
5967         struct ctl_lun *lun = arg;
5968         uint64_t t;
5969
5970         if (lun->ie_asc == 0)
5971                 return;
5972
5973         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5974                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5975         else
5976                 lun->ie_reported = 0;
5977
5978         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5979                 lun->ie_reportcnt++;
5980                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5981                 if (t == 0 || t == UINT32_MAX)
5982                         t = 3000;  /* 5 min */
5983                 callout_schedule(&lun->ie_callout, t * hz / 10);
5984         }
5985 }
5986
5987 int
5988 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5989                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5990 {
5991         struct ctl_lun *lun = CTL_LUN(ctsio);
5992         struct scsi_info_exceptions_page *pg;
5993         uint64_t t;
5994
5995         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5996
5997         pg = (struct scsi_info_exceptions_page *)page_ptr;
5998         mtx_lock(&lun->lun_lock);
5999         if (pg->info_flags & SIEP_FLAGS_TEST) {
6000                 lun->ie_asc = 0x5d;
6001                 lun->ie_ascq = 0xff;
6002                 if (pg->mrie == SIEP_MRIE_UA) {
6003                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
6004                         lun->ie_reported = 1;
6005                 } else {
6006                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
6007                         lun->ie_reported = -1;
6008                 }
6009                 lun->ie_reportcnt = 1;
6010                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
6011                         lun->ie_reportcnt++;
6012                         t = scsi_4btoul(pg->interval_timer);
6013                         if (t == 0 || t == UINT32_MAX)
6014                                 t = 3000;  /* 5 min */
6015                         callout_reset(&lun->ie_callout, t * hz / 10,
6016                             ctl_ie_timer, lun);
6017                 }
6018         } else {
6019                 lun->ie_asc = 0;
6020                 lun->ie_ascq = 0;
6021                 lun->ie_reported = 1;
6022                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
6023                 lun->ie_reportcnt = UINT32_MAX;
6024                 callout_stop(&lun->ie_callout);
6025         }
6026         mtx_unlock(&lun->lun_lock);
6027         return (CTL_RETVAL_COMPLETE);
6028 }
6029
6030 static int
6031 ctl_do_mode_select(union ctl_io *io)
6032 {
6033         struct ctl_lun *lun = CTL_LUN(io);
6034         struct scsi_mode_page_header *page_header;
6035         struct ctl_page_index *page_index;
6036         struct ctl_scsiio *ctsio;
6037         int page_len, page_len_offset, page_len_size;
6038         union ctl_modepage_info *modepage_info;
6039         uint16_t *len_left, *len_used;
6040         int retval, i;
6041
6042         ctsio = &io->scsiio;
6043         page_index = NULL;
6044         page_len = 0;
6045
6046         modepage_info = (union ctl_modepage_info *)
6047                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6048         len_left = &modepage_info->header.len_left;
6049         len_used = &modepage_info->header.len_used;
6050
6051 do_next_page:
6052
6053         page_header = (struct scsi_mode_page_header *)
6054                 (ctsio->kern_data_ptr + *len_used);
6055
6056         if (*len_left == 0) {
6057                 free(ctsio->kern_data_ptr, M_CTL);
6058                 ctl_set_success(ctsio);
6059                 ctl_done((union ctl_io *)ctsio);
6060                 return (CTL_RETVAL_COMPLETE);
6061         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6062
6063                 free(ctsio->kern_data_ptr, M_CTL);
6064                 ctl_set_param_len_error(ctsio);
6065                 ctl_done((union ctl_io *)ctsio);
6066                 return (CTL_RETVAL_COMPLETE);
6067
6068         } else if ((page_header->page_code & SMPH_SPF)
6069                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6070
6071                 free(ctsio->kern_data_ptr, M_CTL);
6072                 ctl_set_param_len_error(ctsio);
6073                 ctl_done((union ctl_io *)ctsio);
6074                 return (CTL_RETVAL_COMPLETE);
6075         }
6076
6077
6078         /*
6079          * XXX KDM should we do something with the block descriptor?
6080          */
6081         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6082                 page_index = &lun->mode_pages.index[i];
6083                 if (lun->be_lun->lun_type == T_DIRECT &&
6084                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6085                         continue;
6086                 if (lun->be_lun->lun_type == T_PROCESSOR &&
6087                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6088                         continue;
6089                 if (lun->be_lun->lun_type == T_CDROM &&
6090                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6091                         continue;
6092
6093                 if ((page_index->page_code & SMPH_PC_MASK) !=
6094                     (page_header->page_code & SMPH_PC_MASK))
6095                         continue;
6096
6097                 /*
6098                  * If neither page has a subpage code, then we've got a
6099                  * match.
6100                  */
6101                 if (((page_index->page_code & SMPH_SPF) == 0)
6102                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6103                         page_len = page_header->page_length;
6104                         break;
6105                 }
6106
6107                 /*
6108                  * If both pages have subpages, then the subpage numbers
6109                  * have to match.
6110                  */
6111                 if ((page_index->page_code & SMPH_SPF)
6112                   && (page_header->page_code & SMPH_SPF)) {
6113                         struct scsi_mode_page_header_sp *sph;
6114
6115                         sph = (struct scsi_mode_page_header_sp *)page_header;
6116                         if (page_index->subpage == sph->subpage) {
6117                                 page_len = scsi_2btoul(sph->page_length);
6118                                 break;
6119                         }
6120                 }
6121         }
6122
6123         /*
6124          * If we couldn't find the page, or if we don't have a mode select
6125          * handler for it, send back an error to the user.
6126          */
6127         if ((i >= CTL_NUM_MODE_PAGES)
6128          || (page_index->select_handler == NULL)) {
6129                 ctl_set_invalid_field(ctsio,
6130                                       /*sks_valid*/ 1,
6131                                       /*command*/ 0,
6132                                       /*field*/ *len_used,
6133                                       /*bit_valid*/ 0,
6134                                       /*bit*/ 0);
6135                 free(ctsio->kern_data_ptr, M_CTL);
6136                 ctl_done((union ctl_io *)ctsio);
6137                 return (CTL_RETVAL_COMPLETE);
6138         }
6139
6140         if (page_index->page_code & SMPH_SPF) {
6141                 page_len_offset = 2;
6142                 page_len_size = 2;
6143         } else {
6144                 page_len_size = 1;
6145                 page_len_offset = 1;
6146         }
6147
6148         /*
6149          * If the length the initiator gives us isn't the one we specify in
6150          * the mode page header, or if they didn't specify enough data in
6151          * the CDB to avoid truncating this page, kick out the request.
6152          */
6153         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6154                 ctl_set_invalid_field(ctsio,
6155                                       /*sks_valid*/ 1,
6156                                       /*command*/ 0,
6157                                       /*field*/ *len_used + page_len_offset,
6158                                       /*bit_valid*/ 0,
6159                                       /*bit*/ 0);
6160                 free(ctsio->kern_data_ptr, M_CTL);
6161                 ctl_done((union ctl_io *)ctsio);
6162                 return (CTL_RETVAL_COMPLETE);
6163         }
6164         if (*len_left < page_index->page_len) {
6165                 free(ctsio->kern_data_ptr, M_CTL);
6166                 ctl_set_param_len_error(ctsio);
6167                 ctl_done((union ctl_io *)ctsio);
6168                 return (CTL_RETVAL_COMPLETE);
6169         }
6170
6171         /*
6172          * Run through the mode page, checking to make sure that the bits
6173          * the user changed are actually legal for him to change.
6174          */
6175         for (i = 0; i < page_index->page_len; i++) {
6176                 uint8_t *user_byte, *change_mask, *current_byte;
6177                 int bad_bit;
6178                 int j;
6179
6180                 user_byte = (uint8_t *)page_header + i;
6181                 change_mask = page_index->page_data +
6182                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6183                 current_byte = page_index->page_data +
6184                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6185
6186                 /*
6187                  * Check to see whether the user set any bits in this byte
6188                  * that he is not allowed to set.
6189                  */
6190                 if ((*user_byte & ~(*change_mask)) ==
6191                     (*current_byte & ~(*change_mask)))
6192                         continue;
6193
6194                 /*
6195                  * Go through bit by bit to determine which one is illegal.
6196                  */
6197                 bad_bit = 0;
6198                 for (j = 7; j >= 0; j--) {
6199                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6200                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6201                                 bad_bit = i;
6202                                 break;
6203                         }
6204                 }
6205                 ctl_set_invalid_field(ctsio,
6206                                       /*sks_valid*/ 1,
6207                                       /*command*/ 0,
6208                                       /*field*/ *len_used + i,
6209                                       /*bit_valid*/ 1,
6210                                       /*bit*/ bad_bit);
6211                 free(ctsio->kern_data_ptr, M_CTL);
6212                 ctl_done((union ctl_io *)ctsio);
6213                 return (CTL_RETVAL_COMPLETE);
6214         }
6215
6216         /*
6217          * Decrement these before we call the page handler, since we may
6218          * end up getting called back one way or another before the handler
6219          * returns to this context.
6220          */
6221         *len_left -= page_index->page_len;
6222         *len_used += page_index->page_len;
6223
6224         retval = page_index->select_handler(ctsio, page_index,
6225                                             (uint8_t *)page_header);
6226
6227         /*
6228          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6229          * wait until this queued command completes to finish processing
6230          * the mode page.  If it returns anything other than
6231          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6232          * already set the sense information, freed the data pointer, and
6233          * completed the io for us.
6234          */
6235         if (retval != CTL_RETVAL_COMPLETE)
6236                 goto bailout_no_done;
6237
6238         /*
6239          * If the initiator sent us more than one page, parse the next one.
6240          */
6241         if (*len_left > 0)
6242                 goto do_next_page;
6243
6244         ctl_set_success(ctsio);
6245         free(ctsio->kern_data_ptr, M_CTL);
6246         ctl_done((union ctl_io *)ctsio);
6247
6248 bailout_no_done:
6249
6250         return (CTL_RETVAL_COMPLETE);
6251
6252 }
6253
6254 int
6255 ctl_mode_select(struct ctl_scsiio *ctsio)
6256 {
6257         struct ctl_lun *lun = CTL_LUN(ctsio);
6258         union ctl_modepage_info *modepage_info;
6259         int bd_len, i, header_size, param_len, rtd;
6260         uint32_t initidx;
6261
6262         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6263         switch (ctsio->cdb[0]) {
6264         case MODE_SELECT_6: {
6265                 struct scsi_mode_select_6 *cdb;
6266
6267                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6268
6269                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6270                 param_len = cdb->length;
6271                 header_size = sizeof(struct scsi_mode_header_6);
6272                 break;
6273         }
6274         case MODE_SELECT_10: {
6275                 struct scsi_mode_select_10 *cdb;
6276
6277                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6278
6279                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6280                 param_len = scsi_2btoul(cdb->length);
6281                 header_size = sizeof(struct scsi_mode_header_10);
6282                 break;
6283         }
6284         default:
6285                 ctl_set_invalid_opcode(ctsio);
6286                 ctl_done((union ctl_io *)ctsio);
6287                 return (CTL_RETVAL_COMPLETE);
6288         }
6289
6290         if (rtd) {
6291                 if (param_len != 0) {
6292                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 0,
6293                             /*command*/ 1, /*field*/ 0,
6294                             /*bit_valid*/ 0, /*bit*/ 0);
6295                         ctl_done((union ctl_io *)ctsio);
6296                         return (CTL_RETVAL_COMPLETE);
6297                 }
6298
6299                 /* Revert to defaults. */
6300                 ctl_init_page_index(lun);
6301                 mtx_lock(&lun->lun_lock);
6302                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6303                 mtx_unlock(&lun->lun_lock);
6304                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6305                         ctl_isc_announce_mode(lun, -1,
6306                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
6307                             lun->mode_pages.index[i].subpage);
6308                 }
6309                 ctl_set_success(ctsio);
6310                 ctl_done((union ctl_io *)ctsio);
6311                 return (CTL_RETVAL_COMPLETE);
6312         }
6313
6314         /*
6315          * From SPC-3:
6316          * "A parameter list length of zero indicates that the Data-Out Buffer
6317          * shall be empty. This condition shall not be considered as an error."
6318          */
6319         if (param_len == 0) {
6320                 ctl_set_success(ctsio);
6321                 ctl_done((union ctl_io *)ctsio);
6322                 return (CTL_RETVAL_COMPLETE);
6323         }
6324
6325         /*
6326          * Since we'll hit this the first time through, prior to
6327          * allocation, we don't need to free a data buffer here.
6328          */
6329         if (param_len < header_size) {
6330                 ctl_set_param_len_error(ctsio);
6331                 ctl_done((union ctl_io *)ctsio);
6332                 return (CTL_RETVAL_COMPLETE);
6333         }
6334
6335         /*
6336          * Allocate the data buffer and grab the user's data.  In theory,
6337          * we shouldn't have to sanity check the parameter list length here
6338          * because the maximum size is 64K.  We should be able to malloc
6339          * that much without too many problems.
6340          */
6341         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6342                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6343                 ctsio->kern_data_len = param_len;
6344                 ctsio->kern_total_len = param_len;
6345                 ctsio->kern_rel_offset = 0;
6346                 ctsio->kern_sg_entries = 0;
6347                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6348                 ctsio->be_move_done = ctl_config_move_done;
6349                 ctl_datamove((union ctl_io *)ctsio);
6350
6351                 return (CTL_RETVAL_COMPLETE);
6352         }
6353
6354         switch (ctsio->cdb[0]) {
6355         case MODE_SELECT_6: {
6356                 struct scsi_mode_header_6 *mh6;
6357
6358                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6359                 bd_len = mh6->blk_desc_len;
6360                 break;
6361         }
6362         case MODE_SELECT_10: {
6363                 struct scsi_mode_header_10 *mh10;
6364
6365                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6366                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6367                 break;
6368         }
6369         default:
6370                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6371         }
6372
6373         if (param_len < (header_size + bd_len)) {
6374                 free(ctsio->kern_data_ptr, M_CTL);
6375                 ctl_set_param_len_error(ctsio);
6376                 ctl_done((union ctl_io *)ctsio);
6377                 return (CTL_RETVAL_COMPLETE);
6378         }
6379
6380         /*
6381          * Set the IO_CONT flag, so that if this I/O gets passed to
6382          * ctl_config_write_done(), it'll get passed back to
6383          * ctl_do_mode_select() for further processing, or completion if
6384          * we're all done.
6385          */
6386         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6387         ctsio->io_cont = ctl_do_mode_select;
6388
6389         modepage_info = (union ctl_modepage_info *)
6390                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6391         memset(modepage_info, 0, sizeof(*modepage_info));
6392         modepage_info->header.len_left = param_len - header_size - bd_len;
6393         modepage_info->header.len_used = header_size + bd_len;
6394
6395         return (ctl_do_mode_select((union ctl_io *)ctsio));
6396 }
6397
6398 int
6399 ctl_mode_sense(struct ctl_scsiio *ctsio)
6400 {
6401         struct ctl_lun *lun = CTL_LUN(ctsio);
6402         int pc, page_code, llba, subpage;
6403         int alloc_len, page_len, header_len, bd_len, total_len;
6404         void *block_desc;
6405         struct ctl_page_index *page_index;
6406
6407         llba = 0;
6408
6409         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6410
6411         switch (ctsio->cdb[0]) {
6412         case MODE_SENSE_6: {
6413                 struct scsi_mode_sense_6 *cdb;
6414
6415                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6416
6417                 header_len = sizeof(struct scsi_mode_hdr_6);
6418                 if (cdb->byte2 & SMS_DBD)
6419                         bd_len = 0;
6420                 else
6421                         bd_len = sizeof(struct scsi_mode_block_descr);
6422                 header_len += bd_len;
6423
6424                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6425                 page_code = cdb->page & SMS_PAGE_CODE;
6426                 subpage = cdb->subpage;
6427                 alloc_len = cdb->length;
6428                 break;
6429         }
6430         case MODE_SENSE_10: {
6431                 struct scsi_mode_sense_10 *cdb;
6432
6433                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6434
6435                 header_len = sizeof(struct scsi_mode_hdr_10);
6436                 if (cdb->byte2 & SMS_DBD) {
6437                         bd_len = 0;
6438                 } else if (lun->be_lun->lun_type == T_DIRECT) {
6439                         if (cdb->byte2 & SMS10_LLBAA) {
6440                                 llba = 1;
6441                                 bd_len = sizeof(struct scsi_mode_block_descr_dlong);
6442                         } else
6443                                 bd_len = sizeof(struct scsi_mode_block_descr_dshort);
6444                 } else
6445                         bd_len = sizeof(struct scsi_mode_block_descr);
6446                 header_len += bd_len;
6447
6448                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6449                 page_code = cdb->page & SMS_PAGE_CODE;
6450                 subpage = cdb->subpage;
6451                 alloc_len = scsi_2btoul(cdb->length);
6452                 break;
6453         }
6454         default:
6455                 ctl_set_invalid_opcode(ctsio);
6456                 ctl_done((union ctl_io *)ctsio);
6457                 return (CTL_RETVAL_COMPLETE);
6458                 break; /* NOTREACHED */
6459         }
6460
6461         /*
6462          * We have to make a first pass through to calculate the size of
6463          * the pages that match the user's query.  Then we allocate enough
6464          * memory to hold it, and actually copy the data into the buffer.
6465          */
6466         switch (page_code) {
6467         case SMS_ALL_PAGES_PAGE: {
6468                 u_int i;
6469
6470                 page_len = 0;
6471
6472                 /*
6473                  * At the moment, values other than 0 and 0xff here are
6474                  * reserved according to SPC-3.
6475                  */
6476                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6477                  && (subpage != SMS_SUBPAGE_ALL)) {
6478                         ctl_set_invalid_field(ctsio,
6479                                               /*sks_valid*/ 1,
6480                                               /*command*/ 1,
6481                                               /*field*/ 3,
6482                                               /*bit_valid*/ 0,
6483                                               /*bit*/ 0);
6484                         ctl_done((union ctl_io *)ctsio);
6485                         return (CTL_RETVAL_COMPLETE);
6486                 }
6487
6488                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6489                         page_index = &lun->mode_pages.index[i];
6490
6491                         /* Make sure the page is supported for this dev type */
6492                         if (lun->be_lun->lun_type == T_DIRECT &&
6493                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6494                                 continue;
6495                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6496                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6497                                 continue;
6498                         if (lun->be_lun->lun_type == T_CDROM &&
6499                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6500                                 continue;
6501
6502                         /*
6503                          * We don't use this subpage if the user didn't
6504                          * request all subpages.
6505                          */
6506                         if ((page_index->subpage != 0)
6507                          && (subpage == SMS_SUBPAGE_PAGE_0))
6508                                 continue;
6509
6510                         page_len += page_index->page_len;
6511                 }
6512                 break;
6513         }
6514         default: {
6515                 u_int i;
6516
6517                 page_len = 0;
6518
6519                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6520                         page_index = &lun->mode_pages.index[i];
6521
6522                         /* Make sure the page is supported for this dev type */
6523                         if (lun->be_lun->lun_type == T_DIRECT &&
6524                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6525                                 continue;
6526                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6527                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6528                                 continue;
6529                         if (lun->be_lun->lun_type == T_CDROM &&
6530                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6531                                 continue;
6532
6533                         /* Look for the right page code */
6534                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6535                                 continue;
6536
6537                         /* Look for the right subpage or the subpage wildcard*/
6538                         if ((page_index->subpage != subpage)
6539                          && (subpage != SMS_SUBPAGE_ALL))
6540                                 continue;
6541
6542                         page_len += page_index->page_len;
6543                 }
6544
6545                 if (page_len == 0) {
6546                         ctl_set_invalid_field(ctsio,
6547                                               /*sks_valid*/ 1,
6548                                               /*command*/ 1,
6549                                               /*field*/ 2,
6550                                               /*bit_valid*/ 1,
6551                                               /*bit*/ 5);
6552                         ctl_done((union ctl_io *)ctsio);
6553                         return (CTL_RETVAL_COMPLETE);
6554                 }
6555                 break;
6556         }
6557         }
6558
6559         total_len = header_len + page_len;
6560
6561         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6562         ctsio->kern_sg_entries = 0;
6563         ctsio->kern_rel_offset = 0;
6564         ctsio->kern_data_len = min(total_len, alloc_len);
6565         ctsio->kern_total_len = ctsio->kern_data_len;
6566
6567         switch (ctsio->cdb[0]) {
6568         case MODE_SENSE_6: {
6569                 struct scsi_mode_hdr_6 *header;
6570
6571                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6572
6573                 header->datalen = MIN(total_len - 1, 254);
6574                 if (lun->be_lun->lun_type == T_DIRECT) {
6575                         header->dev_specific = 0x10; /* DPOFUA */
6576                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6577                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6578                                 header->dev_specific |= 0x80; /* WP */
6579                 }
6580                 header->block_descr_len = bd_len;
6581                 block_desc = &header[1];
6582                 break;
6583         }
6584         case MODE_SENSE_10: {
6585                 struct scsi_mode_hdr_10 *header;
6586                 int datalen;
6587
6588                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6589
6590                 datalen = MIN(total_len - 2, 65533);
6591                 scsi_ulto2b(datalen, header->datalen);
6592                 if (lun->be_lun->lun_type == T_DIRECT) {
6593                         header->dev_specific = 0x10; /* DPOFUA */
6594                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6595                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6596                                 header->dev_specific |= 0x80; /* WP */
6597                 }
6598                 if (llba)
6599                         header->flags |= SMH_LONGLBA;
6600                 scsi_ulto2b(bd_len, header->block_descr_len);
6601                 block_desc = &header[1];
6602                 break;
6603         }
6604         default:
6605                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6606         }
6607
6608         /*
6609          * If we've got a disk, use its blocksize in the block
6610          * descriptor.  Otherwise, just set it to 0.
6611          */
6612         if (bd_len > 0) {
6613                 if (lun->be_lun->lun_type == T_DIRECT) {
6614                         if (llba) {
6615                                 struct scsi_mode_block_descr_dlong *bd = block_desc;
6616                                 if (lun->be_lun->maxlba != 0)
6617                                         scsi_u64to8b(lun->be_lun->maxlba + 1,
6618                                             bd->num_blocks);
6619                                 scsi_ulto4b(lun->be_lun->blocksize,
6620                                     bd->block_len);
6621                         } else {
6622                                 struct scsi_mode_block_descr_dshort *bd = block_desc;
6623                                 if (lun->be_lun->maxlba != 0)
6624                                         scsi_ulto4b(MIN(lun->be_lun->maxlba+1,
6625                                             UINT32_MAX), bd->num_blocks);
6626                                 scsi_ulto3b(lun->be_lun->blocksize,
6627                                     bd->block_len);
6628                         }
6629                 } else {
6630                         struct scsi_mode_block_descr *bd = block_desc;
6631                         scsi_ulto3b(0, bd->block_len);
6632                 }
6633         }
6634
6635         switch (page_code) {
6636         case SMS_ALL_PAGES_PAGE: {
6637                 int i, data_used;
6638
6639                 data_used = header_len;
6640                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6641                         struct ctl_page_index *page_index;
6642
6643                         page_index = &lun->mode_pages.index[i];
6644                         if (lun->be_lun->lun_type == T_DIRECT &&
6645                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6646                                 continue;
6647                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6648                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6649                                 continue;
6650                         if (lun->be_lun->lun_type == T_CDROM &&
6651                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6652                                 continue;
6653
6654                         /*
6655                          * We don't use this subpage if the user didn't
6656                          * request all subpages.  We already checked (above)
6657                          * to make sure the user only specified a subpage
6658                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6659                          */
6660                         if ((page_index->subpage != 0)
6661                          && (subpage == SMS_SUBPAGE_PAGE_0))
6662                                 continue;
6663
6664                         /*
6665                          * Call the handler, if it exists, to update the
6666                          * page to the latest values.
6667                          */
6668                         if (page_index->sense_handler != NULL)
6669                                 page_index->sense_handler(ctsio, page_index,pc);
6670
6671                         memcpy(ctsio->kern_data_ptr + data_used,
6672                                page_index->page_data +
6673                                (page_index->page_len * pc),
6674                                page_index->page_len);
6675                         data_used += page_index->page_len;
6676                 }
6677                 break;
6678         }
6679         default: {
6680                 int i, data_used;
6681
6682                 data_used = header_len;
6683
6684                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6685                         struct ctl_page_index *page_index;
6686
6687                         page_index = &lun->mode_pages.index[i];
6688
6689                         /* Look for the right page code */
6690                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6691                                 continue;
6692
6693                         /* Look for the right subpage or the subpage wildcard*/
6694                         if ((page_index->subpage != subpage)
6695                          && (subpage != SMS_SUBPAGE_ALL))
6696                                 continue;
6697
6698                         /* Make sure the page is supported for this dev type */
6699                         if (lun->be_lun->lun_type == T_DIRECT &&
6700                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6701                                 continue;
6702                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6703                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6704                                 continue;
6705                         if (lun->be_lun->lun_type == T_CDROM &&
6706                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6707                                 continue;
6708
6709                         /*
6710                          * Call the handler, if it exists, to update the
6711                          * page to the latest values.
6712                          */
6713                         if (page_index->sense_handler != NULL)
6714                                 page_index->sense_handler(ctsio, page_index,pc);
6715
6716                         memcpy(ctsio->kern_data_ptr + data_used,
6717                                page_index->page_data +
6718                                (page_index->page_len * pc),
6719                                page_index->page_len);
6720                         data_used += page_index->page_len;
6721                 }
6722                 break;
6723         }
6724         }
6725
6726         ctl_set_success(ctsio);
6727         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6728         ctsio->be_move_done = ctl_config_move_done;
6729         ctl_datamove((union ctl_io *)ctsio);
6730         return (CTL_RETVAL_COMPLETE);
6731 }
6732
6733 int
6734 ctl_temp_log_sense_handler(struct ctl_scsiio *ctsio,
6735                                struct ctl_page_index *page_index,
6736                                int pc)
6737 {
6738         struct ctl_lun *lun = CTL_LUN(ctsio);
6739         struct scsi_log_temperature *data;
6740         const char *value;
6741
6742         data = (struct scsi_log_temperature *)page_index->page_data;
6743
6744         scsi_ulto2b(SLP_TEMPERATURE, data->hdr.param_code);
6745         data->hdr.param_control = SLP_LBIN;
6746         data->hdr.param_len = sizeof(struct scsi_log_temperature) -
6747             sizeof(struct scsi_log_param_header);
6748         if ((value = dnvlist_get_string(lun->be_lun->options, "temperature",
6749             NULL)) != NULL)
6750                 data->temperature = strtol(value, NULL, 0);
6751         else
6752                 data->temperature = 0xff;
6753         data++;
6754
6755         scsi_ulto2b(SLP_REFTEMPERATURE, data->hdr.param_code);
6756         data->hdr.param_control = SLP_LBIN;
6757         data->hdr.param_len = sizeof(struct scsi_log_temperature) -
6758             sizeof(struct scsi_log_param_header);
6759         if ((value = dnvlist_get_string(lun->be_lun->options, "reftemperature",
6760             NULL)) != NULL)
6761                 data->temperature = strtol(value, NULL, 0);
6762         else
6763                 data->temperature = 0xff;
6764         return (0);
6765 }
6766
6767 int
6768 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6769                                struct ctl_page_index *page_index,
6770                                int pc)
6771 {
6772         struct ctl_lun *lun = CTL_LUN(ctsio);
6773         struct scsi_log_param_header *phdr;
6774         uint8_t *data;
6775         uint64_t val;
6776
6777         data = page_index->page_data;
6778
6779         if (lun->backend->lun_attr != NULL &&
6780             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6781              != UINT64_MAX) {
6782                 phdr = (struct scsi_log_param_header *)data;
6783                 scsi_ulto2b(0x0001, phdr->param_code);
6784                 phdr->param_control = SLP_LBIN | SLP_LP;
6785                 phdr->param_len = 8;
6786                 data = (uint8_t *)(phdr + 1);
6787                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6788                 data[4] = 0x02; /* per-pool */
6789                 data += phdr->param_len;
6790         }
6791
6792         if (lun->backend->lun_attr != NULL &&
6793             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6794              != UINT64_MAX) {
6795                 phdr = (struct scsi_log_param_header *)data;
6796                 scsi_ulto2b(0x0002, phdr->param_code);
6797                 phdr->param_control = SLP_LBIN | SLP_LP;
6798                 phdr->param_len = 8;
6799                 data = (uint8_t *)(phdr + 1);
6800                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6801                 data[4] = 0x01; /* per-LUN */
6802                 data += phdr->param_len;
6803         }
6804
6805         if (lun->backend->lun_attr != NULL &&
6806             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6807              != UINT64_MAX) {
6808                 phdr = (struct scsi_log_param_header *)data;
6809                 scsi_ulto2b(0x00f1, phdr->param_code);
6810                 phdr->param_control = SLP_LBIN | SLP_LP;
6811                 phdr->param_len = 8;
6812                 data = (uint8_t *)(phdr + 1);
6813                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6814                 data[4] = 0x02; /* per-pool */
6815                 data += phdr->param_len;
6816         }
6817
6818         if (lun->backend->lun_attr != NULL &&
6819             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6820              != UINT64_MAX) {
6821                 phdr = (struct scsi_log_param_header *)data;
6822                 scsi_ulto2b(0x00f2, phdr->param_code);
6823                 phdr->param_control = SLP_LBIN | SLP_LP;
6824                 phdr->param_len = 8;
6825                 data = (uint8_t *)(phdr + 1);
6826                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6827                 data[4] = 0x02; /* per-pool */
6828                 data += phdr->param_len;
6829         }
6830
6831         page_index->page_len = data - page_index->page_data;
6832         return (0);
6833 }
6834
6835 int
6836 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6837                                struct ctl_page_index *page_index,
6838                                int pc)
6839 {
6840         struct ctl_lun *lun = CTL_LUN(ctsio);
6841         struct stat_page *data;
6842         struct bintime *t;
6843
6844         data = (struct stat_page *)page_index->page_data;
6845
6846         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6847         data->sap.hdr.param_control = SLP_LBIN;
6848         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6849             sizeof(struct scsi_log_param_header);
6850         scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
6851             data->sap.read_num);
6852         scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
6853             data->sap.write_num);
6854         if (lun->be_lun->blocksize > 0) {
6855                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
6856                     lun->be_lun->blocksize, data->sap.recvieved_lba);
6857                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
6858                     lun->be_lun->blocksize, data->sap.transmitted_lba);
6859         }
6860         t = &lun->stats.time[CTL_STATS_READ];
6861         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6862             data->sap.read_int);
6863         t = &lun->stats.time[CTL_STATS_WRITE];
6864         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6865             data->sap.write_int);
6866         scsi_u64to8b(0, data->sap.weighted_num);
6867         scsi_u64to8b(0, data->sap.weighted_int);
6868         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6869         data->it.hdr.param_control = SLP_LBIN;
6870         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6871             sizeof(struct scsi_log_param_header);
6872 #ifdef CTL_TIME_IO
6873         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6874 #endif
6875         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6876         data->it.hdr.param_control = SLP_LBIN;
6877         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6878             sizeof(struct scsi_log_param_header);
6879         scsi_ulto4b(3, data->ti.exponent);
6880         scsi_ulto4b(1, data->ti.integer);
6881         return (0);
6882 }
6883
6884 int
6885 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6886                                struct ctl_page_index *page_index,
6887                                int pc)
6888 {
6889         struct ctl_lun *lun = CTL_LUN(ctsio);
6890         struct scsi_log_informational_exceptions *data;
6891         const char *value;
6892
6893         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6894
6895         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6896         data->hdr.param_control = SLP_LBIN;
6897         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6898             sizeof(struct scsi_log_param_header);
6899         data->ie_asc = lun->ie_asc;
6900         data->ie_ascq = lun->ie_ascq;
6901         if ((value = dnvlist_get_string(lun->be_lun->options, "temperature",
6902             NULL)) != NULL)
6903                 data->temperature = strtol(value, NULL, 0);
6904         else
6905                 data->temperature = 0xff;
6906         return (0);
6907 }
6908
6909 int
6910 ctl_log_sense(struct ctl_scsiio *ctsio)
6911 {
6912         struct ctl_lun *lun = CTL_LUN(ctsio);
6913         int i, pc, page_code, subpage;
6914         int alloc_len, total_len;
6915         struct ctl_page_index *page_index;
6916         struct scsi_log_sense *cdb;
6917         struct scsi_log_header *header;
6918
6919         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6920
6921         cdb = (struct scsi_log_sense *)ctsio->cdb;
6922         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6923         page_code = cdb->page & SLS_PAGE_CODE;
6924         subpage = cdb->subpage;
6925         alloc_len = scsi_2btoul(cdb->length);
6926
6927         page_index = NULL;
6928         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6929                 page_index = &lun->log_pages.index[i];
6930
6931                 /* Look for the right page code */
6932                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6933                         continue;
6934
6935                 /* Look for the right subpage or the subpage wildcard*/
6936                 if (page_index->subpage != subpage)
6937                         continue;
6938
6939                 break;
6940         }
6941         if (i >= CTL_NUM_LOG_PAGES) {
6942                 ctl_set_invalid_field(ctsio,
6943                                       /*sks_valid*/ 1,
6944                                       /*command*/ 1,
6945                                       /*field*/ 2,
6946                                       /*bit_valid*/ 0,
6947                                       /*bit*/ 0);
6948                 ctl_done((union ctl_io *)ctsio);
6949                 return (CTL_RETVAL_COMPLETE);
6950         }
6951
6952         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6953
6954         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6955         ctsio->kern_sg_entries = 0;
6956         ctsio->kern_rel_offset = 0;
6957         ctsio->kern_data_len = min(total_len, alloc_len);
6958         ctsio->kern_total_len = ctsio->kern_data_len;
6959
6960         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6961         header->page = page_index->page_code;
6962         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6963                 header->page |= SL_DS;
6964         if (page_index->subpage) {
6965                 header->page |= SL_SPF;
6966                 header->subpage = page_index->subpage;
6967         }
6968         scsi_ulto2b(page_index->page_len, header->datalen);
6969
6970         /*
6971          * Call the handler, if it exists, to update the
6972          * page to the latest values.
6973          */
6974         if (page_index->sense_handler != NULL)
6975                 page_index->sense_handler(ctsio, page_index, pc);
6976
6977         memcpy(header + 1, page_index->page_data, page_index->page_len);
6978
6979         ctl_set_success(ctsio);
6980         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6981         ctsio->be_move_done = ctl_config_move_done;
6982         ctl_datamove((union ctl_io *)ctsio);
6983         return (CTL_RETVAL_COMPLETE);
6984 }
6985
6986 int
6987 ctl_read_capacity(struct ctl_scsiio *ctsio)
6988 {
6989         struct ctl_lun *lun = CTL_LUN(ctsio);
6990         struct scsi_read_capacity *cdb;
6991         struct scsi_read_capacity_data *data;
6992         uint32_t lba;
6993
6994         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6995
6996         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6997
6998         lba = scsi_4btoul(cdb->addr);
6999         if (((cdb->pmi & SRC_PMI) == 0)
7000          && (lba != 0)) {
7001                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7002                                       /*sks_valid*/ 1,
7003                                       /*command*/ 1,
7004                                       /*field*/ 2,
7005                                       /*bit_valid*/ 0,
7006                                       /*bit*/ 0);
7007                 ctl_done((union ctl_io *)ctsio);
7008                 return (CTL_RETVAL_COMPLETE);
7009         }
7010
7011         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7012         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7013         ctsio->kern_data_len = sizeof(*data);
7014         ctsio->kern_total_len = sizeof(*data);
7015         ctsio->kern_rel_offset = 0;
7016         ctsio->kern_sg_entries = 0;
7017
7018         /*
7019          * If the maximum LBA is greater than 0xfffffffe, the user must
7020          * issue a SERVICE ACTION IN (16) command, with the read capacity
7021          * serivce action set.
7022          */
7023         if (lun->be_lun->maxlba > 0xfffffffe)
7024                 scsi_ulto4b(0xffffffff, data->addr);
7025         else
7026                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7027
7028         /*
7029          * XXX KDM this may not be 512 bytes...
7030          */
7031         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7032
7033         ctl_set_success(ctsio);
7034         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7035         ctsio->be_move_done = ctl_config_move_done;
7036         ctl_datamove((union ctl_io *)ctsio);
7037         return (CTL_RETVAL_COMPLETE);
7038 }
7039
7040 int
7041 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7042 {
7043         struct ctl_lun *lun = CTL_LUN(ctsio);
7044         struct scsi_read_capacity_16 *cdb;
7045         struct scsi_read_capacity_data_long *data;
7046         uint64_t lba;
7047         uint32_t alloc_len;
7048
7049         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7050
7051         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7052
7053         alloc_len = scsi_4btoul(cdb->alloc_len);
7054         lba = scsi_8btou64(cdb->addr);
7055
7056         if ((cdb->reladr & SRC16_PMI)
7057          && (lba != 0)) {
7058                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7059                                       /*sks_valid*/ 1,
7060                                       /*command*/ 1,
7061                                       /*field*/ 2,
7062                                       /*bit_valid*/ 0,
7063                                       /*bit*/ 0);
7064                 ctl_done((union ctl_io *)ctsio);
7065                 return (CTL_RETVAL_COMPLETE);
7066         }
7067
7068         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7069         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7070         ctsio->kern_rel_offset = 0;
7071         ctsio->kern_sg_entries = 0;
7072         ctsio->kern_data_len = min(sizeof(*data), alloc_len);
7073         ctsio->kern_total_len = ctsio->kern_data_len;
7074
7075         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7076         /* XXX KDM this may not be 512 bytes... */
7077         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7078         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7079         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7080         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7081                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7082
7083         ctl_set_success(ctsio);
7084         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7085         ctsio->be_move_done = ctl_config_move_done;
7086         ctl_datamove((union ctl_io *)ctsio);
7087         return (CTL_RETVAL_COMPLETE);
7088 }
7089
7090 int
7091 ctl_get_lba_status(struct ctl_scsiio *ctsio)
7092 {
7093         struct ctl_lun *lun = CTL_LUN(ctsio);
7094         struct scsi_get_lba_status *cdb;
7095         struct scsi_get_lba_status_data *data;
7096         struct ctl_lba_len_flags *lbalen;
7097         uint64_t lba;
7098         uint32_t alloc_len, total_len;
7099         int retval;
7100
7101         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7102
7103         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7104         lba = scsi_8btou64(cdb->addr);
7105         alloc_len = scsi_4btoul(cdb->alloc_len);
7106
7107         if (lba > lun->be_lun->maxlba) {
7108                 ctl_set_lba_out_of_range(ctsio, lba);
7109                 ctl_done((union ctl_io *)ctsio);
7110                 return (CTL_RETVAL_COMPLETE);
7111         }
7112
7113         total_len = sizeof(*data) + sizeof(data->descr[0]);
7114         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7115         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7116         ctsio->kern_rel_offset = 0;
7117         ctsio->kern_sg_entries = 0;
7118         ctsio->kern_data_len = min(total_len, alloc_len);
7119         ctsio->kern_total_len = ctsio->kern_data_len;
7120
7121         /* Fill dummy data in case backend can't tell anything. */
7122         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7123         scsi_u64to8b(lba, data->descr[0].addr);
7124         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7125             data->descr[0].length);
7126         data->descr[0].status = 0; /* Mapped or unknown. */
7127
7128         ctl_set_success(ctsio);
7129         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7130         ctsio->be_move_done = ctl_config_move_done;
7131
7132         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7133         lbalen->lba = lba;
7134         lbalen->len = total_len;
7135         lbalen->flags = 0;
7136         retval = lun->backend->config_read((union ctl_io *)ctsio);
7137         return (retval);
7138 }
7139
7140 int
7141 ctl_read_defect(struct ctl_scsiio *ctsio)
7142 {
7143         struct scsi_read_defect_data_10 *ccb10;
7144         struct scsi_read_defect_data_12 *ccb12;
7145         struct scsi_read_defect_data_hdr_10 *data10;
7146         struct scsi_read_defect_data_hdr_12 *data12;
7147         uint32_t alloc_len, data_len;
7148         uint8_t format;
7149
7150         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7151
7152         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7153                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7154                 format = ccb10->format;
7155                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7156                 data_len = sizeof(*data10);
7157         } else {
7158                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7159                 format = ccb12->format;
7160                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7161                 data_len = sizeof(*data12);
7162         }
7163         if (alloc_len == 0) {
7164                 ctl_set_success(ctsio);
7165                 ctl_done((union ctl_io *)ctsio);
7166                 return (CTL_RETVAL_COMPLETE);
7167         }
7168
7169         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7170         ctsio->kern_rel_offset = 0;
7171         ctsio->kern_sg_entries = 0;
7172         ctsio->kern_data_len = min(data_len, alloc_len);
7173         ctsio->kern_total_len = ctsio->kern_data_len;
7174
7175         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7176                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7177                     ctsio->kern_data_ptr;
7178                 data10->format = format;
7179                 scsi_ulto2b(0, data10->length);
7180         } else {
7181                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7182                     ctsio->kern_data_ptr;
7183                 data12->format = format;
7184                 scsi_ulto2b(0, data12->generation);
7185                 scsi_ulto4b(0, data12->length);
7186         }
7187
7188         ctl_set_success(ctsio);
7189         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7190         ctsio->be_move_done = ctl_config_move_done;
7191         ctl_datamove((union ctl_io *)ctsio);
7192         return (CTL_RETVAL_COMPLETE);
7193 }
7194
7195 int
7196 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7197 {
7198         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7199         struct ctl_lun *lun = CTL_LUN(ctsio);
7200         struct scsi_maintenance_in *cdb;
7201         int retval;
7202         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7203         int num_ha_groups, num_target_ports, shared_group;
7204         struct ctl_port *port;
7205         struct scsi_target_group_data *rtg_ptr;
7206         struct scsi_target_group_data_extended *rtg_ext_ptr;
7207         struct scsi_target_port_group_descriptor *tpg_desc;
7208
7209         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7210
7211         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7212         retval = CTL_RETVAL_COMPLETE;
7213
7214         switch (cdb->byte2 & STG_PDF_MASK) {
7215         case STG_PDF_LENGTH:
7216                 ext = 0;
7217                 break;
7218         case STG_PDF_EXTENDED:
7219                 ext = 1;
7220                 break;
7221         default:
7222                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7223                                       /*sks_valid*/ 1,
7224                                       /*command*/ 1,
7225                                       /*field*/ 2,
7226                                       /*bit_valid*/ 1,
7227                                       /*bit*/ 5);
7228                 ctl_done((union ctl_io *)ctsio);
7229                 return(retval);
7230         }
7231
7232         num_target_ports = 0;
7233         shared_group = (softc->is_single != 0);
7234         mtx_lock(&softc->ctl_lock);
7235         STAILQ_FOREACH(port, &softc->port_list, links) {
7236                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7237                         continue;
7238                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7239                         continue;
7240                 num_target_ports++;
7241                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7242                         shared_group = 1;
7243         }
7244         mtx_unlock(&softc->ctl_lock);
7245         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7246
7247         if (ext)
7248                 total_len = sizeof(struct scsi_target_group_data_extended);
7249         else
7250                 total_len = sizeof(struct scsi_target_group_data);
7251         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7252                 (shared_group + num_ha_groups) +
7253             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7254
7255         alloc_len = scsi_4btoul(cdb->length);
7256
7257         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7258         ctsio->kern_sg_entries = 0;
7259         ctsio->kern_rel_offset = 0;
7260         ctsio->kern_data_len = min(total_len, alloc_len);
7261         ctsio->kern_total_len = ctsio->kern_data_len;
7262
7263         if (ext) {
7264                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7265                     ctsio->kern_data_ptr;
7266                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7267                 rtg_ext_ptr->format_type = 0x10;
7268                 rtg_ext_ptr->implicit_transition_time = 0;
7269                 tpg_desc = &rtg_ext_ptr->groups[0];
7270         } else {
7271                 rtg_ptr = (struct scsi_target_group_data *)
7272                     ctsio->kern_data_ptr;
7273                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7274                 tpg_desc = &rtg_ptr->groups[0];
7275         }
7276
7277         mtx_lock(&softc->ctl_lock);
7278         pg = softc->port_min / softc->port_cnt;
7279         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7280                 /* Some shelf is known to be primary. */
7281                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7282                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7283                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7284                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7285                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7286                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7287                 else
7288                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7289                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7290                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7291                 } else {
7292                         ts = os;
7293                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7294                 }
7295         } else {
7296                 /* No known primary shelf. */
7297                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7298                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7299                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7300                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7301                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7302                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7303                 } else {
7304                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7305                 }
7306         }
7307         if (shared_group) {
7308                 tpg_desc->pref_state = ts;
7309                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7310                     TPG_U_SUP | TPG_T_SUP;
7311                 scsi_ulto2b(1, tpg_desc->target_port_group);
7312                 tpg_desc->status = TPG_IMPLICIT;
7313                 pc = 0;
7314                 STAILQ_FOREACH(port, &softc->port_list, links) {
7315                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7316                                 continue;
7317                         if (!softc->is_single &&
7318                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7319                                 continue;
7320                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7321                                 continue;
7322                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7323                             relative_target_port_identifier);
7324                         pc++;
7325                 }
7326                 tpg_desc->target_port_count = pc;
7327                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7328                     &tpg_desc->descriptors[pc];
7329         }
7330         for (g = 0; g < num_ha_groups; g++) {
7331                 tpg_desc->pref_state = (g == pg) ? ts : os;
7332                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7333                     TPG_U_SUP | TPG_T_SUP;
7334                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7335                 tpg_desc->status = TPG_IMPLICIT;
7336                 pc = 0;
7337                 STAILQ_FOREACH(port, &softc->port_list, links) {
7338                         if (port->targ_port < g * softc->port_cnt ||
7339                             port->targ_port >= (g + 1) * softc->port_cnt)
7340                                 continue;
7341                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7342                                 continue;
7343                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7344                                 continue;
7345                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7346                                 continue;
7347                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7348                             relative_target_port_identifier);
7349                         pc++;
7350                 }
7351                 tpg_desc->target_port_count = pc;
7352                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7353                     &tpg_desc->descriptors[pc];
7354         }
7355         mtx_unlock(&softc->ctl_lock);
7356
7357         ctl_set_success(ctsio);
7358         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7359         ctsio->be_move_done = ctl_config_move_done;
7360         ctl_datamove((union ctl_io *)ctsio);
7361         return(retval);
7362 }
7363
7364 int
7365 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7366 {
7367         struct ctl_lun *lun = CTL_LUN(ctsio);
7368         struct scsi_report_supported_opcodes *cdb;
7369         const struct ctl_cmd_entry *entry, *sentry;
7370         struct scsi_report_supported_opcodes_all *all;
7371         struct scsi_report_supported_opcodes_descr *descr;
7372         struct scsi_report_supported_opcodes_one *one;
7373         int retval;
7374         int alloc_len, total_len;
7375         int opcode, service_action, i, j, num;
7376
7377         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7378
7379         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7380         retval = CTL_RETVAL_COMPLETE;
7381
7382         opcode = cdb->requested_opcode;
7383         service_action = scsi_2btoul(cdb->requested_service_action);
7384         switch (cdb->options & RSO_OPTIONS_MASK) {
7385         case RSO_OPTIONS_ALL:
7386                 num = 0;
7387                 for (i = 0; i < 256; i++) {
7388                         entry = &ctl_cmd_table[i];
7389                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7390                                 for (j = 0; j < 32; j++) {
7391                                         sentry = &((const struct ctl_cmd_entry *)
7392                                             entry->execute)[j];
7393                                         if (ctl_cmd_applicable(
7394                                             lun->be_lun->lun_type, sentry))
7395                                                 num++;
7396                                 }
7397                         } else {
7398                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7399                                     entry))
7400                                         num++;
7401                         }
7402                 }
7403                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7404                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7405                 break;
7406         case RSO_OPTIONS_OC:
7407                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7408                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7409                                               /*sks_valid*/ 1,
7410                                               /*command*/ 1,
7411                                               /*field*/ 2,
7412                                               /*bit_valid*/ 1,
7413                                               /*bit*/ 2);
7414                         ctl_done((union ctl_io *)ctsio);
7415                         return (CTL_RETVAL_COMPLETE);
7416                 }
7417                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7418                 break;
7419         case RSO_OPTIONS_OC_SA:
7420                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7421                     service_action >= 32) {
7422                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7423                                               /*sks_valid*/ 1,
7424                                               /*command*/ 1,
7425                                               /*field*/ 2,
7426                                               /*bit_valid*/ 1,
7427                                               /*bit*/ 2);
7428                         ctl_done((union ctl_io *)ctsio);
7429                         return (CTL_RETVAL_COMPLETE);
7430                 }
7431                 /* FALLTHROUGH */
7432         case RSO_OPTIONS_OC_ASA:
7433                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7434                 break;
7435         default:
7436                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7437                                       /*sks_valid*/ 1,
7438                                       /*command*/ 1,
7439                                       /*field*/ 2,
7440                                       /*bit_valid*/ 1,
7441                                       /*bit*/ 2);
7442                 ctl_done((union ctl_io *)ctsio);
7443                 return (CTL_RETVAL_COMPLETE);
7444         }
7445
7446         alloc_len = scsi_4btoul(cdb->length);
7447
7448         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7449         ctsio->kern_sg_entries = 0;
7450         ctsio->kern_rel_offset = 0;
7451         ctsio->kern_data_len = min(total_len, alloc_len);
7452         ctsio->kern_total_len = ctsio->kern_data_len;
7453
7454         switch (cdb->options & RSO_OPTIONS_MASK) {
7455         case RSO_OPTIONS_ALL:
7456                 all = (struct scsi_report_supported_opcodes_all *)
7457                     ctsio->kern_data_ptr;
7458                 num = 0;
7459                 for (i = 0; i < 256; i++) {
7460                         entry = &ctl_cmd_table[i];
7461                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7462                                 for (j = 0; j < 32; j++) {
7463                                         sentry = &((const struct ctl_cmd_entry *)
7464                                             entry->execute)[j];
7465                                         if (!ctl_cmd_applicable(
7466                                             lun->be_lun->lun_type, sentry))
7467                                                 continue;
7468                                         descr = &all->descr[num++];
7469                                         descr->opcode = i;
7470                                         scsi_ulto2b(j, descr->service_action);
7471                                         descr->flags = RSO_SERVACTV;
7472                                         scsi_ulto2b(sentry->length,
7473                                             descr->cdb_length);
7474                                 }
7475                         } else {
7476                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7477                                     entry))
7478                                         continue;
7479                                 descr = &all->descr[num++];
7480                                 descr->opcode = i;
7481                                 scsi_ulto2b(0, descr->service_action);
7482                                 descr->flags = 0;
7483                                 scsi_ulto2b(entry->length, descr->cdb_length);
7484                         }
7485                 }
7486                 scsi_ulto4b(
7487                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7488                     all->length);
7489                 break;
7490         case RSO_OPTIONS_OC:
7491                 one = (struct scsi_report_supported_opcodes_one *)
7492                     ctsio->kern_data_ptr;
7493                 entry = &ctl_cmd_table[opcode];
7494                 goto fill_one;
7495         case RSO_OPTIONS_OC_SA:
7496                 one = (struct scsi_report_supported_opcodes_one *)
7497                     ctsio->kern_data_ptr;
7498                 entry = &ctl_cmd_table[opcode];
7499                 entry = &((const struct ctl_cmd_entry *)
7500                     entry->execute)[service_action];
7501 fill_one:
7502                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7503                         one->support = 3;
7504                         scsi_ulto2b(entry->length, one->cdb_length);
7505                         one->cdb_usage[0] = opcode;
7506                         memcpy(&one->cdb_usage[1], entry->usage,
7507                             entry->length - 1);
7508                 } else
7509                         one->support = 1;
7510                 break;
7511         case RSO_OPTIONS_OC_ASA:
7512                 one = (struct scsi_report_supported_opcodes_one *)
7513                     ctsio->kern_data_ptr;
7514                 entry = &ctl_cmd_table[opcode];
7515                 if (entry->flags & CTL_CMD_FLAG_SA5) {
7516                         entry = &((const struct ctl_cmd_entry *)
7517                             entry->execute)[service_action];
7518                 } else if (service_action != 0) {
7519                         one->support = 1;
7520                         break;
7521                 }
7522                 goto fill_one;
7523         }
7524
7525         ctl_set_success(ctsio);
7526         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7527         ctsio->be_move_done = ctl_config_move_done;
7528         ctl_datamove((union ctl_io *)ctsio);
7529         return(retval);
7530 }
7531
7532 int
7533 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7534 {
7535         struct scsi_report_supported_tmf *cdb;
7536         struct scsi_report_supported_tmf_ext_data *data;
7537         int retval;
7538         int alloc_len, total_len;
7539
7540         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7541
7542         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7543
7544         retval = CTL_RETVAL_COMPLETE;
7545
7546         if (cdb->options & RST_REPD)
7547                 total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7548         else
7549                 total_len = sizeof(struct scsi_report_supported_tmf_data);
7550         alloc_len = scsi_4btoul(cdb->length);
7551
7552         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7553         ctsio->kern_sg_entries = 0;
7554         ctsio->kern_rel_offset = 0;
7555         ctsio->kern_data_len = min(total_len, alloc_len);
7556         ctsio->kern_total_len = ctsio->kern_data_len;
7557
7558         data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7559         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7560             RST_TRS;
7561         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7562         data->length = total_len - 4;
7563
7564         ctl_set_success(ctsio);
7565         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7566         ctsio->be_move_done = ctl_config_move_done;
7567         ctl_datamove((union ctl_io *)ctsio);
7568         return (retval);
7569 }
7570
7571 int
7572 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7573 {
7574         struct scsi_report_timestamp *cdb;
7575         struct scsi_report_timestamp_data *data;
7576         struct timeval tv;
7577         int64_t timestamp;
7578         int retval;
7579         int alloc_len, total_len;
7580
7581         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7582
7583         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7584
7585         retval = CTL_RETVAL_COMPLETE;
7586
7587         total_len = sizeof(struct scsi_report_timestamp_data);
7588         alloc_len = scsi_4btoul(cdb->length);
7589
7590         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7591         ctsio->kern_sg_entries = 0;
7592         ctsio->kern_rel_offset = 0;
7593         ctsio->kern_data_len = min(total_len, alloc_len);
7594         ctsio->kern_total_len = ctsio->kern_data_len;
7595
7596         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7597         scsi_ulto2b(sizeof(*data) - 2, data->length);
7598         data->origin = RTS_ORIG_OUTSIDE;
7599         getmicrotime(&tv);
7600         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7601         scsi_ulto4b(timestamp >> 16, data->timestamp);
7602         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7603
7604         ctl_set_success(ctsio);
7605         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7606         ctsio->be_move_done = ctl_config_move_done;
7607         ctl_datamove((union ctl_io *)ctsio);
7608         return (retval);
7609 }
7610
7611 int
7612 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7613 {
7614         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7615         struct ctl_lun *lun = CTL_LUN(ctsio);
7616         struct scsi_per_res_in *cdb;
7617         int alloc_len, total_len = 0;
7618         /* struct scsi_per_res_in_rsrv in_data; */
7619         uint64_t key;
7620
7621         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7622
7623         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7624
7625         alloc_len = scsi_2btoul(cdb->length);
7626
7627 retry:
7628         mtx_lock(&lun->lun_lock);
7629         switch (cdb->action) {
7630         case SPRI_RK: /* read keys */
7631                 total_len = sizeof(struct scsi_per_res_in_keys) +
7632                         lun->pr_key_count *
7633                         sizeof(struct scsi_per_res_key);
7634                 break;
7635         case SPRI_RR: /* read reservation */
7636                 if (lun->flags & CTL_LUN_PR_RESERVED)
7637                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7638                 else
7639                         total_len = sizeof(struct scsi_per_res_in_header);
7640                 break;
7641         case SPRI_RC: /* report capabilities */
7642                 total_len = sizeof(struct scsi_per_res_cap);
7643                 break;
7644         case SPRI_RS: /* read full status */
7645                 total_len = sizeof(struct scsi_per_res_in_header) +
7646                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7647                     lun->pr_key_count;
7648                 break;
7649         default:
7650                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7651         }
7652         mtx_unlock(&lun->lun_lock);
7653
7654         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7655         ctsio->kern_rel_offset = 0;
7656         ctsio->kern_sg_entries = 0;
7657         ctsio->kern_data_len = min(total_len, alloc_len);
7658         ctsio->kern_total_len = ctsio->kern_data_len;
7659
7660         mtx_lock(&lun->lun_lock);
7661         switch (cdb->action) {
7662         case SPRI_RK: { // read keys
7663         struct scsi_per_res_in_keys *res_keys;
7664                 int i, key_count;
7665
7666                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7667
7668                 /*
7669                  * We had to drop the lock to allocate our buffer, which
7670                  * leaves time for someone to come in with another
7671                  * persistent reservation.  (That is unlikely, though,
7672                  * since this should be the only persistent reservation
7673                  * command active right now.)
7674                  */
7675                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7676                     (lun->pr_key_count *
7677                      sizeof(struct scsi_per_res_key)))){
7678                         mtx_unlock(&lun->lun_lock);
7679                         free(ctsio->kern_data_ptr, M_CTL);
7680                         printf("%s: reservation length changed, retrying\n",
7681                                __func__);
7682                         goto retry;
7683                 }
7684
7685                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7686
7687                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7688                              lun->pr_key_count, res_keys->header.length);
7689
7690                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7691                         if ((key = ctl_get_prkey(lun, i)) == 0)
7692                                 continue;
7693
7694                         /*
7695                          * We used lun->pr_key_count to calculate the
7696                          * size to allocate.  If it turns out the number of
7697                          * initiators with the registered flag set is
7698                          * larger than that (i.e. they haven't been kept in
7699                          * sync), we've got a problem.
7700                          */
7701                         if (key_count >= lun->pr_key_count) {
7702                                 key_count++;
7703                                 continue;
7704                         }
7705                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7706                         key_count++;
7707                 }
7708                 break;
7709         }
7710         case SPRI_RR: { // read reservation
7711                 struct scsi_per_res_in_rsrv *res;
7712                 int tmp_len, header_only;
7713
7714                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7715
7716                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7717
7718                 if (lun->flags & CTL_LUN_PR_RESERVED)
7719                 {
7720                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7721                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7722                                     res->header.length);
7723                         header_only = 0;
7724                 } else {
7725                         tmp_len = sizeof(struct scsi_per_res_in_header);
7726                         scsi_ulto4b(0, res->header.length);
7727                         header_only = 1;
7728                 }
7729
7730                 /*
7731                  * We had to drop the lock to allocate our buffer, which
7732                  * leaves time for someone to come in with another
7733                  * persistent reservation.  (That is unlikely, though,
7734                  * since this should be the only persistent reservation
7735                  * command active right now.)
7736                  */
7737                 if (tmp_len != total_len) {
7738                         mtx_unlock(&lun->lun_lock);
7739                         free(ctsio->kern_data_ptr, M_CTL);
7740                         printf("%s: reservation status changed, retrying\n",
7741                                __func__);
7742                         goto retry;
7743                 }
7744
7745                 /*
7746                  * No reservation held, so we're done.
7747                  */
7748                 if (header_only != 0)
7749                         break;
7750
7751                 /*
7752                  * If the registration is an All Registrants type, the key
7753                  * is 0, since it doesn't really matter.
7754                  */
7755                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7756                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7757                             res->data.reservation);
7758                 }
7759                 res->data.scopetype = lun->pr_res_type;
7760                 break;
7761         }
7762         case SPRI_RC:     //report capabilities
7763         {
7764                 struct scsi_per_res_cap *res_cap;
7765                 uint16_t type_mask;
7766
7767                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7768                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7769                 res_cap->flags1 = SPRI_CRH;
7770                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7771                 type_mask = SPRI_TM_WR_EX_AR |
7772                             SPRI_TM_EX_AC_RO |
7773                             SPRI_TM_WR_EX_RO |
7774                             SPRI_TM_EX_AC |
7775                             SPRI_TM_WR_EX |
7776                             SPRI_TM_EX_AC_AR;
7777                 scsi_ulto2b(type_mask, res_cap->type_mask);
7778                 break;
7779         }
7780         case SPRI_RS: { // read full status
7781                 struct scsi_per_res_in_full *res_status;
7782                 struct scsi_per_res_in_full_desc *res_desc;
7783                 struct ctl_port *port;
7784                 int i, len;
7785
7786                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7787
7788                 /*
7789                  * We had to drop the lock to allocate our buffer, which
7790                  * leaves time for someone to come in with another
7791                  * persistent reservation.  (That is unlikely, though,
7792                  * since this should be the only persistent reservation
7793                  * command active right now.)
7794                  */
7795                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7796                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7797                      lun->pr_key_count)){
7798                         mtx_unlock(&lun->lun_lock);
7799                         free(ctsio->kern_data_ptr, M_CTL);
7800                         printf("%s: reservation length changed, retrying\n",
7801                                __func__);
7802                         goto retry;
7803                 }
7804
7805                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7806
7807                 res_desc = &res_status->desc[0];
7808                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7809                         if ((key = ctl_get_prkey(lun, i)) == 0)
7810                                 continue;
7811
7812                         scsi_u64to8b(key, res_desc->res_key.key);
7813                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7814                             (lun->pr_res_idx == i ||
7815                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7816                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7817                                 res_desc->scopetype = lun->pr_res_type;
7818                         }
7819                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7820                             res_desc->rel_trgt_port_id);
7821                         len = 0;
7822                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7823                         if (port != NULL)
7824                                 len = ctl_create_iid(port,
7825                                     i % CTL_MAX_INIT_PER_PORT,
7826                                     res_desc->transport_id);
7827                         scsi_ulto4b(len, res_desc->additional_length);
7828                         res_desc = (struct scsi_per_res_in_full_desc *)
7829                             &res_desc->transport_id[len];
7830                 }
7831                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7832                     res_status->header.length);
7833                 break;
7834         }
7835         default:
7836                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7837         }
7838         mtx_unlock(&lun->lun_lock);
7839
7840         ctl_set_success(ctsio);
7841         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7842         ctsio->be_move_done = ctl_config_move_done;
7843         ctl_datamove((union ctl_io *)ctsio);
7844         return (CTL_RETVAL_COMPLETE);
7845 }
7846
7847 /*
7848  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7849  * it should return.
7850  */
7851 static int
7852 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7853                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7854                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7855                 struct scsi_per_res_out_parms* param)
7856 {
7857         union ctl_ha_msg persis_io;
7858         int i;
7859
7860         mtx_lock(&lun->lun_lock);
7861         if (sa_res_key == 0) {
7862                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7863                         /* validate scope and type */
7864                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7865                              SPR_LU_SCOPE) {
7866                                 mtx_unlock(&lun->lun_lock);
7867                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7868                                                       /*sks_valid*/ 1,
7869                                                       /*command*/ 1,
7870                                                       /*field*/ 2,
7871                                                       /*bit_valid*/ 1,
7872                                                       /*bit*/ 4);
7873                                 ctl_done((union ctl_io *)ctsio);
7874                                 return (1);
7875                         }
7876
7877                         if (type>8 || type==2 || type==4 || type==0) {
7878                                 mtx_unlock(&lun->lun_lock);
7879                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7880                                                       /*sks_valid*/ 1,
7881                                                       /*command*/ 1,
7882                                                       /*field*/ 2,
7883                                                       /*bit_valid*/ 1,
7884                                                       /*bit*/ 0);
7885                                 ctl_done((union ctl_io *)ctsio);
7886                                 return (1);
7887                         }
7888
7889                         /*
7890                          * Unregister everybody else and build UA for
7891                          * them
7892                          */
7893                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7894                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7895                                         continue;
7896
7897                                 ctl_clr_prkey(lun, i);
7898                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7899                         }
7900                         lun->pr_key_count = 1;
7901                         lun->pr_res_type = type;
7902                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7903                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7904                                 lun->pr_res_idx = residx;
7905                         lun->pr_generation++;
7906                         mtx_unlock(&lun->lun_lock);
7907
7908                         /* send msg to other side */
7909                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7910                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7911                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7912                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7913                         persis_io.pr.pr_info.res_type = type;
7914                         memcpy(persis_io.pr.pr_info.sa_res_key,
7915                                param->serv_act_res_key,
7916                                sizeof(param->serv_act_res_key));
7917                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7918                             sizeof(persis_io.pr), M_WAITOK);
7919                 } else {
7920                         /* not all registrants */
7921                         mtx_unlock(&lun->lun_lock);
7922                         free(ctsio->kern_data_ptr, M_CTL);
7923                         ctl_set_invalid_field(ctsio,
7924                                               /*sks_valid*/ 1,
7925                                               /*command*/ 0,
7926                                               /*field*/ 8,
7927                                               /*bit_valid*/ 0,
7928                                               /*bit*/ 0);
7929                         ctl_done((union ctl_io *)ctsio);
7930                         return (1);
7931                 }
7932         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7933                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7934                 int found = 0;
7935
7936                 if (res_key == sa_res_key) {
7937                         /* special case */
7938                         /*
7939                          * The spec implies this is not good but doesn't
7940                          * say what to do. There are two choices either
7941                          * generate a res conflict or check condition
7942                          * with illegal field in parameter data. Since
7943                          * that is what is done when the sa_res_key is
7944                          * zero I'll take that approach since this has
7945                          * to do with the sa_res_key.
7946                          */
7947                         mtx_unlock(&lun->lun_lock);
7948                         free(ctsio->kern_data_ptr, M_CTL);
7949                         ctl_set_invalid_field(ctsio,
7950                                               /*sks_valid*/ 1,
7951                                               /*command*/ 0,
7952                                               /*field*/ 8,
7953                                               /*bit_valid*/ 0,
7954                                               /*bit*/ 0);
7955                         ctl_done((union ctl_io *)ctsio);
7956                         return (1);
7957                 }
7958
7959                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7960                         if (ctl_get_prkey(lun, i) != sa_res_key)
7961                                 continue;
7962
7963                         found = 1;
7964                         ctl_clr_prkey(lun, i);
7965                         lun->pr_key_count--;
7966                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7967                 }
7968                 if (!found) {
7969                         mtx_unlock(&lun->lun_lock);
7970                         free(ctsio->kern_data_ptr, M_CTL);
7971                         ctl_set_reservation_conflict(ctsio);
7972                         ctl_done((union ctl_io *)ctsio);
7973                         return (CTL_RETVAL_COMPLETE);
7974                 }
7975                 lun->pr_generation++;
7976                 mtx_unlock(&lun->lun_lock);
7977
7978                 /* send msg to other side */
7979                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7980                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7981                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7982                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7983                 persis_io.pr.pr_info.res_type = type;
7984                 memcpy(persis_io.pr.pr_info.sa_res_key,
7985                        param->serv_act_res_key,
7986                        sizeof(param->serv_act_res_key));
7987                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7988                     sizeof(persis_io.pr), M_WAITOK);
7989         } else {
7990                 /* Reserved but not all registrants */
7991                 /* sa_res_key is res holder */
7992                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7993                         /* validate scope and type */
7994                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7995                              SPR_LU_SCOPE) {
7996                                 mtx_unlock(&lun->lun_lock);
7997                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7998                                                       /*sks_valid*/ 1,
7999                                                       /*command*/ 1,
8000                                                       /*field*/ 2,
8001                                                       /*bit_valid*/ 1,
8002                                                       /*bit*/ 4);
8003                                 ctl_done((union ctl_io *)ctsio);
8004                                 return (1);
8005                         }
8006
8007                         if (type>8 || type==2 || type==4 || type==0) {
8008                                 mtx_unlock(&lun->lun_lock);
8009                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8010                                                       /*sks_valid*/ 1,
8011                                                       /*command*/ 1,
8012                                                       /*field*/ 2,
8013                                                       /*bit_valid*/ 1,
8014                                                       /*bit*/ 0);
8015                                 ctl_done((union ctl_io *)ctsio);
8016                                 return (1);
8017                         }
8018
8019                         /*
8020                          * Do the following:
8021                          * if sa_res_key != res_key remove all
8022                          * registrants w/sa_res_key and generate UA
8023                          * for these registrants(Registrations
8024                          * Preempted) if it wasn't an exclusive
8025                          * reservation generate UA(Reservations
8026                          * Preempted) for all other registered nexuses
8027                          * if the type has changed. Establish the new
8028                          * reservation and holder. If res_key and
8029                          * sa_res_key are the same do the above
8030                          * except don't unregister the res holder.
8031                          */
8032
8033                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8034                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8035                                         continue;
8036
8037                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
8038                                         ctl_clr_prkey(lun, i);
8039                                         lun->pr_key_count--;
8040                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8041                                 } else if (type != lun->pr_res_type &&
8042                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8043                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8044                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8045                                 }
8046                         }
8047                         lun->pr_res_type = type;
8048                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8049                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8050                                 lun->pr_res_idx = residx;
8051                         else
8052                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8053                         lun->pr_generation++;
8054                         mtx_unlock(&lun->lun_lock);
8055
8056                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8057                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8058                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8059                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8060                         persis_io.pr.pr_info.res_type = type;
8061                         memcpy(persis_io.pr.pr_info.sa_res_key,
8062                                param->serv_act_res_key,
8063                                sizeof(param->serv_act_res_key));
8064                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8065                             sizeof(persis_io.pr), M_WAITOK);
8066                 } else {
8067                         /*
8068                          * sa_res_key is not the res holder just
8069                          * remove registrants
8070                          */
8071                         int found=0;
8072
8073                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8074                                 if (sa_res_key != ctl_get_prkey(lun, i))
8075                                         continue;
8076
8077                                 found = 1;
8078                                 ctl_clr_prkey(lun, i);
8079                                 lun->pr_key_count--;
8080                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8081                         }
8082
8083                         if (!found) {
8084                                 mtx_unlock(&lun->lun_lock);
8085                                 free(ctsio->kern_data_ptr, M_CTL);
8086                                 ctl_set_reservation_conflict(ctsio);
8087                                 ctl_done((union ctl_io *)ctsio);
8088                                 return (1);
8089                         }
8090                         lun->pr_generation++;
8091                         mtx_unlock(&lun->lun_lock);
8092
8093                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8094                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8095                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8096                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8097                         persis_io.pr.pr_info.res_type = type;
8098                         memcpy(persis_io.pr.pr_info.sa_res_key,
8099                                param->serv_act_res_key,
8100                                sizeof(param->serv_act_res_key));
8101                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8102                             sizeof(persis_io.pr), M_WAITOK);
8103                 }
8104         }
8105         return (0);
8106 }
8107
8108 static void
8109 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8110 {
8111         uint64_t sa_res_key;
8112         int i;
8113
8114         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8115
8116         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8117          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8118          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8119                 if (sa_res_key == 0) {
8120                         /*
8121                          * Unregister everybody else and build UA for
8122                          * them
8123                          */
8124                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8125                                 if (i == msg->pr.pr_info.residx ||
8126                                     ctl_get_prkey(lun, i) == 0)
8127                                         continue;
8128
8129                                 ctl_clr_prkey(lun, i);
8130                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8131                         }
8132
8133                         lun->pr_key_count = 1;
8134                         lun->pr_res_type = msg->pr.pr_info.res_type;
8135                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8136                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8137                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8138                 } else {
8139                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8140                                 if (sa_res_key == ctl_get_prkey(lun, i))
8141                                         continue;
8142
8143                                 ctl_clr_prkey(lun, i);
8144                                 lun->pr_key_count--;
8145                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8146                         }
8147                 }
8148         } else {
8149                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8150                         if (i == msg->pr.pr_info.residx ||
8151                             ctl_get_prkey(lun, i) == 0)
8152                                 continue;
8153
8154                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8155                                 ctl_clr_prkey(lun, i);
8156                                 lun->pr_key_count--;
8157                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8158                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8159                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8160                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8161                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8162                         }
8163                 }
8164                 lun->pr_res_type = msg->pr.pr_info.res_type;
8165                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8166                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8167                         lun->pr_res_idx = msg->pr.pr_info.residx;
8168                 else
8169                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8170         }
8171         lun->pr_generation++;
8172
8173 }
8174
8175
8176 int
8177 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8178 {
8179         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8180         struct ctl_lun *lun = CTL_LUN(ctsio);
8181         int retval;
8182         u_int32_t param_len;
8183         struct scsi_per_res_out *cdb;
8184         struct scsi_per_res_out_parms* param;
8185         uint32_t residx;
8186         uint64_t res_key, sa_res_key, key;
8187         uint8_t type;
8188         union ctl_ha_msg persis_io;
8189         int    i;
8190
8191         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8192
8193         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8194         retval = CTL_RETVAL_COMPLETE;
8195
8196         /*
8197          * We only support whole-LUN scope.  The scope & type are ignored for
8198          * register, register and ignore existing key and clear.
8199          * We sometimes ignore scope and type on preempts too!!
8200          * Verify reservation type here as well.
8201          */
8202         type = cdb->scope_type & SPR_TYPE_MASK;
8203         if ((cdb->action == SPRO_RESERVE)
8204          || (cdb->action == SPRO_RELEASE)) {
8205                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8206                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8207                                               /*sks_valid*/ 1,
8208                                               /*command*/ 1,
8209                                               /*field*/ 2,
8210                                               /*bit_valid*/ 1,
8211                                               /*bit*/ 4);
8212                         ctl_done((union ctl_io *)ctsio);
8213                         return (CTL_RETVAL_COMPLETE);
8214                 }
8215
8216                 if (type>8 || type==2 || type==4 || type==0) {
8217                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8218                                               /*sks_valid*/ 1,
8219                                               /*command*/ 1,
8220                                               /*field*/ 2,
8221                                               /*bit_valid*/ 1,
8222                                               /*bit*/ 0);
8223                         ctl_done((union ctl_io *)ctsio);
8224                         return (CTL_RETVAL_COMPLETE);
8225                 }
8226         }
8227
8228         param_len = scsi_4btoul(cdb->length);
8229
8230         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8231                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8232                 ctsio->kern_data_len = param_len;
8233                 ctsio->kern_total_len = param_len;
8234                 ctsio->kern_rel_offset = 0;
8235                 ctsio->kern_sg_entries = 0;
8236                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8237                 ctsio->be_move_done = ctl_config_move_done;
8238                 ctl_datamove((union ctl_io *)ctsio);
8239
8240                 return (CTL_RETVAL_COMPLETE);
8241         }
8242
8243         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8244
8245         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8246         res_key = scsi_8btou64(param->res_key.key);
8247         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8248
8249         /*
8250          * Validate the reservation key here except for SPRO_REG_IGNO
8251          * This must be done for all other service actions
8252          */
8253         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8254                 mtx_lock(&lun->lun_lock);
8255                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8256                         if (res_key != key) {
8257                                 /*
8258                                  * The current key passed in doesn't match
8259                                  * the one the initiator previously
8260                                  * registered.
8261                                  */
8262                                 mtx_unlock(&lun->lun_lock);
8263                                 free(ctsio->kern_data_ptr, M_CTL);
8264                                 ctl_set_reservation_conflict(ctsio);
8265                                 ctl_done((union ctl_io *)ctsio);
8266                                 return (CTL_RETVAL_COMPLETE);
8267                         }
8268                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8269                         /*
8270                          * We are not registered
8271                          */
8272                         mtx_unlock(&lun->lun_lock);
8273                         free(ctsio->kern_data_ptr, M_CTL);
8274                         ctl_set_reservation_conflict(ctsio);
8275                         ctl_done((union ctl_io *)ctsio);
8276                         return (CTL_RETVAL_COMPLETE);
8277                 } else if (res_key != 0) {
8278                         /*
8279                          * We are not registered and trying to register but
8280                          * the register key isn't zero.
8281                          */
8282                         mtx_unlock(&lun->lun_lock);
8283                         free(ctsio->kern_data_ptr, M_CTL);
8284                         ctl_set_reservation_conflict(ctsio);
8285                         ctl_done((union ctl_io *)ctsio);
8286                         return (CTL_RETVAL_COMPLETE);
8287                 }
8288                 mtx_unlock(&lun->lun_lock);
8289         }
8290
8291         switch (cdb->action & SPRO_ACTION_MASK) {
8292         case SPRO_REGISTER:
8293         case SPRO_REG_IGNO: {
8294
8295                 /*
8296                  * We don't support any of these options, as we report in
8297                  * the read capabilities request (see
8298                  * ctl_persistent_reserve_in(), above).
8299                  */
8300                 if ((param->flags & SPR_SPEC_I_PT)
8301                  || (param->flags & SPR_ALL_TG_PT)
8302                  || (param->flags & SPR_APTPL)) {
8303                         int bit_ptr;
8304
8305                         if (param->flags & SPR_APTPL)
8306                                 bit_ptr = 0;
8307                         else if (param->flags & SPR_ALL_TG_PT)
8308                                 bit_ptr = 2;
8309                         else /* SPR_SPEC_I_PT */
8310                                 bit_ptr = 3;
8311
8312                         free(ctsio->kern_data_ptr, M_CTL);
8313                         ctl_set_invalid_field(ctsio,
8314                                               /*sks_valid*/ 1,
8315                                               /*command*/ 0,
8316                                               /*field*/ 20,
8317                                               /*bit_valid*/ 1,
8318                                               /*bit*/ bit_ptr);
8319                         ctl_done((union ctl_io *)ctsio);
8320                         return (CTL_RETVAL_COMPLETE);
8321                 }
8322
8323                 mtx_lock(&lun->lun_lock);
8324
8325                 /*
8326                  * The initiator wants to clear the
8327                  * key/unregister.
8328                  */
8329                 if (sa_res_key == 0) {
8330                         if ((res_key == 0
8331                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8332                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8333                           && ctl_get_prkey(lun, residx) == 0)) {
8334                                 mtx_unlock(&lun->lun_lock);
8335                                 goto done;
8336                         }
8337
8338                         ctl_clr_prkey(lun, residx);
8339                         lun->pr_key_count--;
8340
8341                         if (residx == lun->pr_res_idx) {
8342                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8343                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8344
8345                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8346                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8347                                     lun->pr_key_count) {
8348                                         /*
8349                                          * If the reservation is a registrants
8350                                          * only type we need to generate a UA
8351                                          * for other registered inits.  The
8352                                          * sense code should be RESERVATIONS
8353                                          * RELEASED
8354                                          */
8355
8356                                         for (i = softc->init_min; i < softc->init_max; i++){
8357                                                 if (ctl_get_prkey(lun, i) == 0)
8358                                                         continue;
8359                                                 ctl_est_ua(lun, i,
8360                                                     CTL_UA_RES_RELEASE);
8361                                         }
8362                                 }
8363                                 lun->pr_res_type = 0;
8364                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8365                                 if (lun->pr_key_count==0) {
8366                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8367                                         lun->pr_res_type = 0;
8368                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8369                                 }
8370                         }
8371                         lun->pr_generation++;
8372                         mtx_unlock(&lun->lun_lock);
8373
8374                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8375                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8376                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8377                         persis_io.pr.pr_info.residx = residx;
8378                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8379                             sizeof(persis_io.pr), M_WAITOK);
8380                 } else /* sa_res_key != 0 */ {
8381
8382                         /*
8383                          * If we aren't registered currently then increment
8384                          * the key count and set the registered flag.
8385                          */
8386                         ctl_alloc_prkey(lun, residx);
8387                         if (ctl_get_prkey(lun, residx) == 0)
8388                                 lun->pr_key_count++;
8389                         ctl_set_prkey(lun, residx, sa_res_key);
8390                         lun->pr_generation++;
8391                         mtx_unlock(&lun->lun_lock);
8392
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_REG_KEY;
8396                         persis_io.pr.pr_info.residx = residx;
8397                         memcpy(persis_io.pr.pr_info.sa_res_key,
8398                                param->serv_act_res_key,
8399                                sizeof(param->serv_act_res_key));
8400                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8401                             sizeof(persis_io.pr), M_WAITOK);
8402                 }
8403
8404                 break;
8405         }
8406         case SPRO_RESERVE:
8407                 mtx_lock(&lun->lun_lock);
8408                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8409                         /*
8410                          * if this isn't the reservation holder and it's
8411                          * not a "all registrants" type or if the type is
8412                          * different then we have a conflict
8413                          */
8414                         if ((lun->pr_res_idx != residx
8415                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8416                          || lun->pr_res_type != type) {
8417                                 mtx_unlock(&lun->lun_lock);
8418                                 free(ctsio->kern_data_ptr, M_CTL);
8419                                 ctl_set_reservation_conflict(ctsio);
8420                                 ctl_done((union ctl_io *)ctsio);
8421                                 return (CTL_RETVAL_COMPLETE);
8422                         }
8423                         mtx_unlock(&lun->lun_lock);
8424                 } else /* create a reservation */ {
8425                         /*
8426                          * If it's not an "all registrants" type record
8427                          * reservation holder
8428                          */
8429                         if (type != SPR_TYPE_WR_EX_AR
8430                          && type != SPR_TYPE_EX_AC_AR)
8431                                 lun->pr_res_idx = residx; /* Res holder */
8432                         else
8433                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8434
8435                         lun->flags |= CTL_LUN_PR_RESERVED;
8436                         lun->pr_res_type = type;
8437
8438                         mtx_unlock(&lun->lun_lock);
8439
8440                         /* send msg to other side */
8441                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8442                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8443                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8444                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8445                         persis_io.pr.pr_info.res_type = type;
8446                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8447                             sizeof(persis_io.pr), M_WAITOK);
8448                 }
8449                 break;
8450
8451         case SPRO_RELEASE:
8452                 mtx_lock(&lun->lun_lock);
8453                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8454                         /* No reservation exists return good status */
8455                         mtx_unlock(&lun->lun_lock);
8456                         goto done;
8457                 }
8458                 /*
8459                  * Is this nexus a reservation holder?
8460                  */
8461                 if (lun->pr_res_idx != residx
8462                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8463                         /*
8464                          * not a res holder return good status but
8465                          * do nothing
8466                          */
8467                         mtx_unlock(&lun->lun_lock);
8468                         goto done;
8469                 }
8470
8471                 if (lun->pr_res_type != type) {
8472                         mtx_unlock(&lun->lun_lock);
8473                         free(ctsio->kern_data_ptr, M_CTL);
8474                         ctl_set_illegal_pr_release(ctsio);
8475                         ctl_done((union ctl_io *)ctsio);
8476                         return (CTL_RETVAL_COMPLETE);
8477                 }
8478
8479                 /* okay to release */
8480                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8481                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8482                 lun->pr_res_type = 0;
8483
8484                 /*
8485                  * If this isn't an exclusive access reservation and NUAR
8486                  * is not set, generate UA for all other registrants.
8487                  */
8488                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8489                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8490                         for (i = softc->init_min; i < softc->init_max; i++) {
8491                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8492                                         continue;
8493                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8494                         }
8495                 }
8496                 mtx_unlock(&lun->lun_lock);
8497
8498                 /* Send msg to other side */
8499                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8500                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8501                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8502                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8503                      sizeof(persis_io.pr), M_WAITOK);
8504                 break;
8505
8506         case SPRO_CLEAR:
8507                 /* send msg to other side */
8508
8509                 mtx_lock(&lun->lun_lock);
8510                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8511                 lun->pr_res_type = 0;
8512                 lun->pr_key_count = 0;
8513                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8514
8515                 ctl_clr_prkey(lun, residx);
8516                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8517                         if (ctl_get_prkey(lun, i) != 0) {
8518                                 ctl_clr_prkey(lun, i);
8519                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8520                         }
8521                 lun->pr_generation++;
8522                 mtx_unlock(&lun->lun_lock);
8523
8524                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8525                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8526                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8527                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8528                      sizeof(persis_io.pr), M_WAITOK);
8529                 break;
8530
8531         case SPRO_PREEMPT:
8532         case SPRO_PRE_ABO: {
8533                 int nretval;
8534
8535                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8536                                           residx, ctsio, cdb, param);
8537                 if (nretval != 0)
8538                         return (CTL_RETVAL_COMPLETE);
8539                 break;
8540         }
8541         default:
8542                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8543         }
8544
8545 done:
8546         free(ctsio->kern_data_ptr, M_CTL);
8547         ctl_set_success(ctsio);
8548         ctl_done((union ctl_io *)ctsio);
8549
8550         return (retval);
8551 }
8552
8553 /*
8554  * This routine is for handling a message from the other SC pertaining to
8555  * persistent reserve out. All the error checking will have been done
8556  * so only perorming the action need be done here to keep the two
8557  * in sync.
8558  */
8559 static void
8560 ctl_hndl_per_res_out_on_other_sc(union ctl_io *io)
8561 {
8562         struct ctl_softc *softc = CTL_SOFTC(io);
8563         union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg;
8564         struct ctl_lun *lun;
8565         int i;
8566         uint32_t residx, targ_lun;
8567
8568         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8569         mtx_lock(&softc->ctl_lock);
8570         if (targ_lun >= ctl_max_luns ||
8571             (lun = softc->ctl_luns[targ_lun]) == NULL) {
8572                 mtx_unlock(&softc->ctl_lock);
8573                 return;
8574         }
8575         mtx_lock(&lun->lun_lock);
8576         mtx_unlock(&softc->ctl_lock);
8577         if (lun->flags & CTL_LUN_DISABLED) {
8578                 mtx_unlock(&lun->lun_lock);
8579                 return;
8580         }
8581         residx = ctl_get_initindex(&msg->hdr.nexus);
8582         switch(msg->pr.pr_info.action) {
8583         case CTL_PR_REG_KEY:
8584                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8585                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8586                         lun->pr_key_count++;
8587                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8588                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8589                 lun->pr_generation++;
8590                 break;
8591
8592         case CTL_PR_UNREG_KEY:
8593                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8594                 lun->pr_key_count--;
8595
8596                 /* XXX Need to see if the reservation has been released */
8597                 /* if so do we need to generate UA? */
8598                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8599                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8600                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8601
8602                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8603                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8604                             lun->pr_key_count) {
8605                                 /*
8606                                  * If the reservation is a registrants
8607                                  * only type we need to generate a UA
8608                                  * for other registered inits.  The
8609                                  * sense code should be RESERVATIONS
8610                                  * RELEASED
8611                                  */
8612
8613                                 for (i = softc->init_min; i < softc->init_max; i++) {
8614                                         if (ctl_get_prkey(lun, i) == 0)
8615                                                 continue;
8616
8617                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8618                                 }
8619                         }
8620                         lun->pr_res_type = 0;
8621                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8622                         if (lun->pr_key_count==0) {
8623                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8624                                 lun->pr_res_type = 0;
8625                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8626                         }
8627                 }
8628                 lun->pr_generation++;
8629                 break;
8630
8631         case CTL_PR_RESERVE:
8632                 lun->flags |= CTL_LUN_PR_RESERVED;
8633                 lun->pr_res_type = msg->pr.pr_info.res_type;
8634                 lun->pr_res_idx = msg->pr.pr_info.residx;
8635
8636                 break;
8637
8638         case CTL_PR_RELEASE:
8639                 /*
8640                  * If this isn't an exclusive access reservation and NUAR
8641                  * is not set, generate UA for all other registrants.
8642                  */
8643                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8644                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8645                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8646                         for (i = softc->init_min; i < softc->init_max; i++) {
8647                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8648                                         continue;
8649                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8650                         }
8651                 }
8652
8653                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8654                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8655                 lun->pr_res_type = 0;
8656                 break;
8657
8658         case CTL_PR_PREEMPT:
8659                 ctl_pro_preempt_other(lun, msg);
8660                 break;
8661         case CTL_PR_CLEAR:
8662                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8663                 lun->pr_res_type = 0;
8664                 lun->pr_key_count = 0;
8665                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8666
8667                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8668                         if (ctl_get_prkey(lun, i) == 0)
8669                                 continue;
8670                         ctl_clr_prkey(lun, i);
8671                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8672                 }
8673                 lun->pr_generation++;
8674                 break;
8675         }
8676
8677         mtx_unlock(&lun->lun_lock);
8678 }
8679
8680 int
8681 ctl_read_write(struct ctl_scsiio *ctsio)
8682 {
8683         struct ctl_lun *lun = CTL_LUN(ctsio);
8684         struct ctl_lba_len_flags *lbalen;
8685         uint64_t lba;
8686         uint32_t num_blocks;
8687         int flags, retval;
8688         int isread;
8689
8690         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8691
8692         flags = 0;
8693         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8694               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8695         switch (ctsio->cdb[0]) {
8696         case READ_6:
8697         case WRITE_6: {
8698                 struct scsi_rw_6 *cdb;
8699
8700                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8701
8702                 lba = scsi_3btoul(cdb->addr);
8703                 /* only 5 bits are valid in the most significant address byte */
8704                 lba &= 0x1fffff;
8705                 num_blocks = cdb->length;
8706                 /*
8707                  * This is correct according to SBC-2.
8708                  */
8709                 if (num_blocks == 0)
8710                         num_blocks = 256;
8711                 break;
8712         }
8713         case READ_10:
8714         case WRITE_10: {
8715                 struct scsi_rw_10 *cdb;
8716
8717                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8718                 if (cdb->byte2 & SRW10_FUA)
8719                         flags |= CTL_LLF_FUA;
8720                 if (cdb->byte2 & SRW10_DPO)
8721                         flags |= CTL_LLF_DPO;
8722                 lba = scsi_4btoul(cdb->addr);
8723                 num_blocks = scsi_2btoul(cdb->length);
8724                 break;
8725         }
8726         case WRITE_VERIFY_10: {
8727                 struct scsi_write_verify_10 *cdb;
8728
8729                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8730                 flags |= CTL_LLF_FUA;
8731                 if (cdb->byte2 & SWV_DPO)
8732                         flags |= CTL_LLF_DPO;
8733                 lba = scsi_4btoul(cdb->addr);
8734                 num_blocks = scsi_2btoul(cdb->length);
8735                 break;
8736         }
8737         case READ_12:
8738         case WRITE_12: {
8739                 struct scsi_rw_12 *cdb;
8740
8741                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8742                 if (cdb->byte2 & SRW12_FUA)
8743                         flags |= CTL_LLF_FUA;
8744                 if (cdb->byte2 & SRW12_DPO)
8745                         flags |= CTL_LLF_DPO;
8746                 lba = scsi_4btoul(cdb->addr);
8747                 num_blocks = scsi_4btoul(cdb->length);
8748                 break;
8749         }
8750         case WRITE_VERIFY_12: {
8751                 struct scsi_write_verify_12 *cdb;
8752
8753                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8754                 flags |= CTL_LLF_FUA;
8755                 if (cdb->byte2 & SWV_DPO)
8756                         flags |= CTL_LLF_DPO;
8757                 lba = scsi_4btoul(cdb->addr);
8758                 num_blocks = scsi_4btoul(cdb->length);
8759                 break;
8760         }
8761         case READ_16:
8762         case WRITE_16: {
8763                 struct scsi_rw_16 *cdb;
8764
8765                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8766                 if (cdb->byte2 & SRW12_FUA)
8767                         flags |= CTL_LLF_FUA;
8768                 if (cdb->byte2 & SRW12_DPO)
8769                         flags |= CTL_LLF_DPO;
8770                 lba = scsi_8btou64(cdb->addr);
8771                 num_blocks = scsi_4btoul(cdb->length);
8772                 break;
8773         }
8774         case WRITE_ATOMIC_16: {
8775                 struct scsi_write_atomic_16 *cdb;
8776
8777                 if (lun->be_lun->atomicblock == 0) {
8778                         ctl_set_invalid_opcode(ctsio);
8779                         ctl_done((union ctl_io *)ctsio);
8780                         return (CTL_RETVAL_COMPLETE);
8781                 }
8782
8783                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8784                 if (cdb->byte2 & SRW12_FUA)
8785                         flags |= CTL_LLF_FUA;
8786                 if (cdb->byte2 & SRW12_DPO)
8787                         flags |= CTL_LLF_DPO;
8788                 lba = scsi_8btou64(cdb->addr);
8789                 num_blocks = scsi_2btoul(cdb->length);
8790                 if (num_blocks > lun->be_lun->atomicblock) {
8791                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8792                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8793                             /*bit*/ 0);
8794                         ctl_done((union ctl_io *)ctsio);
8795                         return (CTL_RETVAL_COMPLETE);
8796                 }
8797                 break;
8798         }
8799         case WRITE_VERIFY_16: {
8800                 struct scsi_write_verify_16 *cdb;
8801
8802                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8803                 flags |= CTL_LLF_FUA;
8804                 if (cdb->byte2 & SWV_DPO)
8805                         flags |= CTL_LLF_DPO;
8806                 lba = scsi_8btou64(cdb->addr);
8807                 num_blocks = scsi_4btoul(cdb->length);
8808                 break;
8809         }
8810         default:
8811                 /*
8812                  * We got a command we don't support.  This shouldn't
8813                  * happen, commands should be filtered out above us.
8814                  */
8815                 ctl_set_invalid_opcode(ctsio);
8816                 ctl_done((union ctl_io *)ctsio);
8817
8818                 return (CTL_RETVAL_COMPLETE);
8819                 break; /* NOTREACHED */
8820         }
8821
8822         /*
8823          * The first check is to make sure we're in bounds, the second
8824          * check is to catch wrap-around problems.  If the lba + num blocks
8825          * is less than the lba, then we've wrapped around and the block
8826          * range is invalid anyway.
8827          */
8828         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8829          || ((lba + num_blocks) < lba)) {
8830                 ctl_set_lba_out_of_range(ctsio,
8831                     MAX(lba, lun->be_lun->maxlba + 1));
8832                 ctl_done((union ctl_io *)ctsio);
8833                 return (CTL_RETVAL_COMPLETE);
8834         }
8835
8836         /*
8837          * According to SBC-3, a transfer length of 0 is not an error.
8838          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8839          * translates to 256 blocks for those commands.
8840          */
8841         if (num_blocks == 0) {
8842                 ctl_set_success(ctsio);
8843                 ctl_done((union ctl_io *)ctsio);
8844                 return (CTL_RETVAL_COMPLETE);
8845         }
8846
8847         /* Set FUA and/or DPO if caches are disabled. */
8848         if (isread) {
8849                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8850                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8851         } else {
8852                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8853                         flags |= CTL_LLF_FUA;
8854         }
8855
8856         lbalen = (struct ctl_lba_len_flags *)
8857             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8858         lbalen->lba = lba;
8859         lbalen->len = num_blocks;
8860         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8861
8862         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8863         ctsio->kern_rel_offset = 0;
8864
8865         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8866
8867         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8868         return (retval);
8869 }
8870
8871 static int
8872 ctl_cnw_cont(union ctl_io *io)
8873 {
8874         struct ctl_lun *lun = CTL_LUN(io);
8875         struct ctl_scsiio *ctsio;
8876         struct ctl_lba_len_flags *lbalen;
8877         int retval;
8878
8879         ctsio = &io->scsiio;
8880         ctsio->io_hdr.status = CTL_STATUS_NONE;
8881         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8882         lbalen = (struct ctl_lba_len_flags *)
8883             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8884         lbalen->flags &= ~CTL_LLF_COMPARE;
8885         lbalen->flags |= CTL_LLF_WRITE;
8886
8887         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8888         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8889         return (retval);
8890 }
8891
8892 int
8893 ctl_cnw(struct ctl_scsiio *ctsio)
8894 {
8895         struct ctl_lun *lun = CTL_LUN(ctsio);
8896         struct ctl_lba_len_flags *lbalen;
8897         uint64_t lba;
8898         uint32_t num_blocks;
8899         int flags, retval;
8900
8901         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8902
8903         flags = 0;
8904         switch (ctsio->cdb[0]) {
8905         case COMPARE_AND_WRITE: {
8906                 struct scsi_compare_and_write *cdb;
8907
8908                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8909                 if (cdb->byte2 & SRW10_FUA)
8910                         flags |= CTL_LLF_FUA;
8911                 if (cdb->byte2 & SRW10_DPO)
8912                         flags |= CTL_LLF_DPO;
8913                 lba = scsi_8btou64(cdb->addr);
8914                 num_blocks = cdb->length;
8915                 break;
8916         }
8917         default:
8918                 /*
8919                  * We got a command we don't support.  This shouldn't
8920                  * happen, commands should be filtered out above us.
8921                  */
8922                 ctl_set_invalid_opcode(ctsio);
8923                 ctl_done((union ctl_io *)ctsio);
8924
8925                 return (CTL_RETVAL_COMPLETE);
8926                 break; /* NOTREACHED */
8927         }
8928
8929         /*
8930          * The first check is to make sure we're in bounds, the second
8931          * check is to catch wrap-around problems.  If the lba + num blocks
8932          * is less than the lba, then we've wrapped around and the block
8933          * range is invalid anyway.
8934          */
8935         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8936          || ((lba + num_blocks) < lba)) {
8937                 ctl_set_lba_out_of_range(ctsio,
8938                     MAX(lba, lun->be_lun->maxlba + 1));
8939                 ctl_done((union ctl_io *)ctsio);
8940                 return (CTL_RETVAL_COMPLETE);
8941         }
8942
8943         /*
8944          * According to SBC-3, a transfer length of 0 is not an error.
8945          */
8946         if (num_blocks == 0) {
8947                 ctl_set_success(ctsio);
8948                 ctl_done((union ctl_io *)ctsio);
8949                 return (CTL_RETVAL_COMPLETE);
8950         }
8951
8952         /* Set FUA if write cache is disabled. */
8953         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8954                 flags |= CTL_LLF_FUA;
8955
8956         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8957         ctsio->kern_rel_offset = 0;
8958
8959         /*
8960          * Set the IO_CONT flag, so that if this I/O gets passed to
8961          * ctl_data_submit_done(), it'll get passed back to
8962          * ctl_ctl_cnw_cont() for further processing.
8963          */
8964         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8965         ctsio->io_cont = ctl_cnw_cont;
8966
8967         lbalen = (struct ctl_lba_len_flags *)
8968             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8969         lbalen->lba = lba;
8970         lbalen->len = num_blocks;
8971         lbalen->flags = CTL_LLF_COMPARE | flags;
8972
8973         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8974         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8975         return (retval);
8976 }
8977
8978 int
8979 ctl_verify(struct ctl_scsiio *ctsio)
8980 {
8981         struct ctl_lun *lun = CTL_LUN(ctsio);
8982         struct ctl_lba_len_flags *lbalen;
8983         uint64_t lba;
8984         uint32_t num_blocks;
8985         int bytchk, flags;
8986         int retval;
8987
8988         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8989
8990         bytchk = 0;
8991         flags = CTL_LLF_FUA;
8992         switch (ctsio->cdb[0]) {
8993         case VERIFY_10: {
8994                 struct scsi_verify_10 *cdb;
8995
8996                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8997                 if (cdb->byte2 & SVFY_BYTCHK)
8998                         bytchk = 1;
8999                 if (cdb->byte2 & SVFY_DPO)
9000                         flags |= CTL_LLF_DPO;
9001                 lba = scsi_4btoul(cdb->addr);
9002                 num_blocks = scsi_2btoul(cdb->length);
9003                 break;
9004         }
9005         case VERIFY_12: {
9006                 struct scsi_verify_12 *cdb;
9007
9008                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
9009                 if (cdb->byte2 & SVFY_BYTCHK)
9010                         bytchk = 1;
9011                 if (cdb->byte2 & SVFY_DPO)
9012                         flags |= CTL_LLF_DPO;
9013                 lba = scsi_4btoul(cdb->addr);
9014                 num_blocks = scsi_4btoul(cdb->length);
9015                 break;
9016         }
9017         case VERIFY_16: {
9018                 struct scsi_rw_16 *cdb;
9019
9020                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9021                 if (cdb->byte2 & SVFY_BYTCHK)
9022                         bytchk = 1;
9023                 if (cdb->byte2 & SVFY_DPO)
9024                         flags |= CTL_LLF_DPO;
9025                 lba = scsi_8btou64(cdb->addr);
9026                 num_blocks = scsi_4btoul(cdb->length);
9027                 break;
9028         }
9029         default:
9030                 /*
9031                  * We got a command we don't support.  This shouldn't
9032                  * happen, commands should be filtered out above us.
9033                  */
9034                 ctl_set_invalid_opcode(ctsio);
9035                 ctl_done((union ctl_io *)ctsio);
9036                 return (CTL_RETVAL_COMPLETE);
9037         }
9038
9039         /*
9040          * The first check is to make sure we're in bounds, the second
9041          * check is to catch wrap-around problems.  If the lba + num blocks
9042          * is less than the lba, then we've wrapped around and the block
9043          * range is invalid anyway.
9044          */
9045         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9046          || ((lba + num_blocks) < lba)) {
9047                 ctl_set_lba_out_of_range(ctsio,
9048                     MAX(lba, lun->be_lun->maxlba + 1));
9049                 ctl_done((union ctl_io *)ctsio);
9050                 return (CTL_RETVAL_COMPLETE);
9051         }
9052
9053         /*
9054          * According to SBC-3, a transfer length of 0 is not an error.
9055          */
9056         if (num_blocks == 0) {
9057                 ctl_set_success(ctsio);
9058                 ctl_done((union ctl_io *)ctsio);
9059                 return (CTL_RETVAL_COMPLETE);
9060         }
9061
9062         lbalen = (struct ctl_lba_len_flags *)
9063             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9064         lbalen->lba = lba;
9065         lbalen->len = num_blocks;
9066         if (bytchk) {
9067                 lbalen->flags = CTL_LLF_COMPARE | flags;
9068                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9069         } else {
9070                 lbalen->flags = CTL_LLF_VERIFY | flags;
9071                 ctsio->kern_total_len = 0;
9072         }
9073         ctsio->kern_rel_offset = 0;
9074
9075         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9076         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9077         return (retval);
9078 }
9079
9080 int
9081 ctl_report_luns(struct ctl_scsiio *ctsio)
9082 {
9083         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9084         struct ctl_port *port = CTL_PORT(ctsio);
9085         struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio);
9086         struct scsi_report_luns *cdb;
9087         struct scsi_report_luns_data *lun_data;
9088         int num_filled, num_luns, num_port_luns, retval;
9089         uint32_t alloc_len, lun_datalen;
9090         uint32_t initidx, targ_lun_id, lun_id;
9091
9092         retval = CTL_RETVAL_COMPLETE;
9093         cdb = (struct scsi_report_luns *)ctsio->cdb;
9094
9095         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9096
9097         num_luns = 0;
9098         num_port_luns = port->lun_map ? port->lun_map_size : ctl_max_luns;
9099         mtx_lock(&softc->ctl_lock);
9100         for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) {
9101                 if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX)
9102                         num_luns++;
9103         }
9104         mtx_unlock(&softc->ctl_lock);
9105
9106         switch (cdb->select_report) {
9107         case RPL_REPORT_DEFAULT:
9108         case RPL_REPORT_ALL:
9109         case RPL_REPORT_NONSUBSID:
9110                 break;
9111         case RPL_REPORT_WELLKNOWN:
9112         case RPL_REPORT_ADMIN:
9113         case RPL_REPORT_CONGLOM:
9114                 num_luns = 0;
9115                 break;
9116         default:
9117                 ctl_set_invalid_field(ctsio,
9118                                       /*sks_valid*/ 1,
9119                                       /*command*/ 1,
9120                                       /*field*/ 2,
9121                                       /*bit_valid*/ 0,
9122                                       /*bit*/ 0);
9123                 ctl_done((union ctl_io *)ctsio);
9124                 return (retval);
9125                 break; /* NOTREACHED */
9126         }
9127
9128         alloc_len = scsi_4btoul(cdb->length);
9129         /*
9130          * The initiator has to allocate at least 16 bytes for this request,
9131          * so he can at least get the header and the first LUN.  Otherwise
9132          * we reject the request (per SPC-3 rev 14, section 6.21).
9133          */
9134         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9135             sizeof(struct scsi_report_luns_lundata))) {
9136                 ctl_set_invalid_field(ctsio,
9137                                       /*sks_valid*/ 1,
9138                                       /*command*/ 1,
9139                                       /*field*/ 6,
9140                                       /*bit_valid*/ 0,
9141                                       /*bit*/ 0);
9142                 ctl_done((union ctl_io *)ctsio);
9143                 return (retval);
9144         }
9145
9146         lun_datalen = sizeof(*lun_data) +
9147                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9148
9149         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9150         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9151         ctsio->kern_sg_entries = 0;
9152
9153         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9154
9155         mtx_lock(&softc->ctl_lock);
9156         for (targ_lun_id = 0, num_filled = 0;
9157             targ_lun_id < num_port_luns && num_filled < num_luns;
9158             targ_lun_id++) {
9159                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9160                 if (lun_id == UINT32_MAX)
9161                         continue;
9162                 lun = softc->ctl_luns[lun_id];
9163                 if (lun == NULL)
9164                         continue;
9165
9166                 be64enc(lun_data->luns[num_filled++].lundata,
9167                     ctl_encode_lun(targ_lun_id));
9168
9169                 /*
9170                  * According to SPC-3, rev 14 section 6.21:
9171                  *
9172                  * "The execution of a REPORT LUNS command to any valid and
9173                  * installed logical unit shall clear the REPORTED LUNS DATA
9174                  * HAS CHANGED unit attention condition for all logical
9175                  * units of that target with respect to the requesting
9176                  * initiator. A valid and installed logical unit is one
9177                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9178                  * INQUIRY data (see 6.4.2)."
9179                  *
9180                  * If request_lun is NULL, the LUN this report luns command
9181                  * was issued to is either disabled or doesn't exist. In that
9182                  * case, we shouldn't clear any pending lun change unit
9183                  * attention.
9184                  */
9185                 if (request_lun != NULL) {
9186                         mtx_lock(&lun->lun_lock);
9187                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9188                         mtx_unlock(&lun->lun_lock);
9189                 }
9190         }
9191         mtx_unlock(&softc->ctl_lock);
9192
9193         /*
9194          * It's quite possible that we've returned fewer LUNs than we allocated
9195          * space for.  Trim it.
9196          */
9197         lun_datalen = sizeof(*lun_data) +
9198                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9199         ctsio->kern_rel_offset = 0;
9200         ctsio->kern_sg_entries = 0;
9201         ctsio->kern_data_len = min(lun_datalen, alloc_len);
9202         ctsio->kern_total_len = ctsio->kern_data_len;
9203
9204         /*
9205          * We set this to the actual data length, regardless of how much
9206          * space we actually have to return results.  If the user looks at
9207          * this value, he'll know whether or not he allocated enough space
9208          * and reissue the command if necessary.  We don't support well
9209          * known logical units, so if the user asks for that, return none.
9210          */
9211         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9212
9213         /*
9214          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9215          * this request.
9216          */
9217         ctl_set_success(ctsio);
9218         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9219         ctsio->be_move_done = ctl_config_move_done;
9220         ctl_datamove((union ctl_io *)ctsio);
9221         return (retval);
9222 }
9223
9224 int
9225 ctl_request_sense(struct ctl_scsiio *ctsio)
9226 {
9227         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9228         struct ctl_lun *lun = CTL_LUN(ctsio);
9229         struct scsi_request_sense *cdb;
9230         struct scsi_sense_data *sense_ptr, *ps;
9231         uint32_t initidx;
9232         int have_error;
9233         u_int sense_len = SSD_FULL_SIZE;
9234         scsi_sense_data_type sense_format;
9235         ctl_ua_type ua_type;
9236         uint8_t asc = 0, ascq = 0;
9237
9238         cdb = (struct scsi_request_sense *)ctsio->cdb;
9239
9240         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9241
9242         /*
9243          * Determine which sense format the user wants.
9244          */
9245         if (cdb->byte2 & SRS_DESC)
9246                 sense_format = SSD_TYPE_DESC;
9247         else
9248                 sense_format = SSD_TYPE_FIXED;
9249
9250         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9251         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9252         ctsio->kern_sg_entries = 0;
9253         ctsio->kern_rel_offset = 0;
9254
9255         /*
9256          * struct scsi_sense_data, which is currently set to 256 bytes, is
9257          * larger than the largest allowed value for the length field in the
9258          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9259          */
9260         ctsio->kern_data_len = cdb->length;
9261         ctsio->kern_total_len = cdb->length;
9262
9263         /*
9264          * If we don't have a LUN, we don't have any pending sense.
9265          */
9266         if (lun == NULL ||
9267             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9268              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9269                 /* "Logical unit not supported" */
9270                 ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9271                     /*current_error*/ 1,
9272                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9273                     /*asc*/ 0x25,
9274                     /*ascq*/ 0x00,
9275                     SSD_ELEM_NONE);
9276                 goto send;
9277         }
9278
9279         have_error = 0;
9280         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9281         /*
9282          * Check for pending sense, and then for pending unit attentions.
9283          * Pending sense gets returned first, then pending unit attentions.
9284          */
9285         mtx_lock(&lun->lun_lock);
9286         ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
9287         if (ps != NULL)
9288                 ps += initidx % CTL_MAX_INIT_PER_PORT;
9289         if (ps != NULL && ps->error_code != 0) {
9290                 scsi_sense_data_type stored_format;
9291
9292                 /*
9293                  * Check to see which sense format was used for the stored
9294                  * sense data.
9295                  */
9296                 stored_format = scsi_sense_type(ps);
9297
9298                 /*
9299                  * If the user requested a different sense format than the
9300                  * one we stored, then we need to convert it to the other
9301                  * format.  If we're going from descriptor to fixed format
9302                  * sense data, we may lose things in translation, depending
9303                  * on what options were used.
9304                  *
9305                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9306                  * for some reason we'll just copy it out as-is.
9307                  */
9308                 if ((stored_format == SSD_TYPE_FIXED)
9309                  && (sense_format == SSD_TYPE_DESC))
9310                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9311                             ps, (struct scsi_sense_data_desc *)sense_ptr);
9312                 else if ((stored_format == SSD_TYPE_DESC)
9313                       && (sense_format == SSD_TYPE_FIXED))
9314                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9315                             ps, (struct scsi_sense_data_fixed *)sense_ptr);
9316                 else
9317                         memcpy(sense_ptr, ps, sizeof(*sense_ptr));
9318
9319                 ps->error_code = 0;
9320                 have_error = 1;
9321         } else {
9322                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9323                     sense_format);
9324                 if (ua_type != CTL_UA_NONE)
9325                         have_error = 1;
9326         }
9327         if (have_error == 0) {
9328                 /*
9329                  * Report informational exception if have one and allowed.
9330                  */
9331                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9332                         asc = lun->ie_asc;
9333                         ascq = lun->ie_ascq;
9334                 }
9335                 ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9336                     /*current_error*/ 1,
9337                     /*sense_key*/ SSD_KEY_NO_SENSE,
9338                     /*asc*/ asc,
9339                     /*ascq*/ ascq,
9340                     SSD_ELEM_NONE);
9341         }
9342         mtx_unlock(&lun->lun_lock);
9343
9344 send:
9345         /*
9346          * We report the SCSI status as OK, since the status of the command
9347          * itself is OK.  We're reporting sense as parameter data.
9348          */
9349         ctl_set_success(ctsio);
9350         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9351         ctsio->be_move_done = ctl_config_move_done;
9352         ctl_datamove((union ctl_io *)ctsio);
9353         return (CTL_RETVAL_COMPLETE);
9354 }
9355
9356 int
9357 ctl_tur(struct ctl_scsiio *ctsio)
9358 {
9359
9360         CTL_DEBUG_PRINT(("ctl_tur\n"));
9361
9362         ctl_set_success(ctsio);
9363         ctl_done((union ctl_io *)ctsio);
9364
9365         return (CTL_RETVAL_COMPLETE);
9366 }
9367
9368 /*
9369  * SCSI VPD page 0x00, the Supported VPD Pages page.
9370  */
9371 static int
9372 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9373 {
9374         struct ctl_lun *lun = CTL_LUN(ctsio);
9375         struct scsi_vpd_supported_pages *pages;
9376         int sup_page_size;
9377         int p;
9378
9379         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9380             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9381         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9382         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9383         ctsio->kern_rel_offset = 0;
9384         ctsio->kern_sg_entries = 0;
9385         ctsio->kern_data_len = min(sup_page_size, alloc_len);
9386         ctsio->kern_total_len = ctsio->kern_data_len;
9387
9388         /*
9389          * The control device is always connected.  The disk device, on the
9390          * other hand, may not be online all the time.  Need to change this
9391          * to figure out whether the disk device is actually online or not.
9392          */
9393         if (lun != NULL)
9394                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9395                                 lun->be_lun->lun_type;
9396         else
9397                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9398
9399         p = 0;
9400         /* Supported VPD pages */
9401         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9402         /* Serial Number */
9403         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9404         /* Device Identification */
9405         pages->page_list[p++] = SVPD_DEVICE_ID;
9406         /* Extended INQUIRY Data */
9407         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9408         /* Mode Page Policy */
9409         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9410         /* SCSI Ports */
9411         pages->page_list[p++] = SVPD_SCSI_PORTS;
9412         /* Third-party Copy */
9413         pages->page_list[p++] = SVPD_SCSI_TPC;
9414         /* SCSI Feature Sets */
9415         pages->page_list[p++] = SVPD_SCSI_SFS;
9416         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9417                 /* Block limits */
9418                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9419                 /* Block Device Characteristics */
9420                 pages->page_list[p++] = SVPD_BDC;
9421                 /* Logical Block Provisioning */
9422                 pages->page_list[p++] = SVPD_LBP;
9423         }
9424         pages->length = p;
9425
9426         ctl_set_success(ctsio);
9427         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9428         ctsio->be_move_done = ctl_config_move_done;
9429         ctl_datamove((union ctl_io *)ctsio);
9430         return (CTL_RETVAL_COMPLETE);
9431 }
9432
9433 /*
9434  * SCSI VPD page 0x80, the Unit Serial Number page.
9435  */
9436 static int
9437 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9438 {
9439         struct ctl_lun *lun = CTL_LUN(ctsio);
9440         struct scsi_vpd_unit_serial_number *sn_ptr;
9441         int data_len;
9442
9443         data_len = 4 + CTL_SN_LEN;
9444         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9445         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9446         ctsio->kern_rel_offset = 0;
9447         ctsio->kern_sg_entries = 0;
9448         ctsio->kern_data_len = min(data_len, alloc_len);
9449         ctsio->kern_total_len = ctsio->kern_data_len;
9450
9451         /*
9452          * The control device is always connected.  The disk device, on the
9453          * other hand, may not be online all the time.  Need to change this
9454          * to figure out whether the disk device is actually online or not.
9455          */
9456         if (lun != NULL)
9457                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9458                                   lun->be_lun->lun_type;
9459         else
9460                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9461
9462         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9463         sn_ptr->length = CTL_SN_LEN;
9464         /*
9465          * If we don't have a LUN, we just leave the serial number as
9466          * all spaces.
9467          */
9468         if (lun != NULL) {
9469                 strncpy((char *)sn_ptr->serial_num,
9470                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9471         } else
9472                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
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 /*
9483  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9484  */
9485 static int
9486 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9487 {
9488         struct ctl_lun *lun = CTL_LUN(ctsio);
9489         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9490         int data_len;
9491
9492         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9493         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9494         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9495         ctsio->kern_sg_entries = 0;
9496         ctsio->kern_rel_offset = 0;
9497         ctsio->kern_data_len = min(data_len, alloc_len);
9498         ctsio->kern_total_len = ctsio->kern_data_len;
9499
9500         /*
9501          * The control device is always connected.  The disk device, on the
9502          * other hand, may not be online all the time.
9503          */
9504         if (lun != NULL)
9505                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9506                                      lun->be_lun->lun_type;
9507         else
9508                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9509         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9510         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9511         /*
9512          * We support head of queue, ordered and simple tags.
9513          */
9514         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9515         /*
9516          * Volatile cache supported.
9517          */
9518         eid_ptr->flags3 = SVPD_EID_V_SUP;
9519
9520         /*
9521          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9522          * attention for a particular IT nexus on all LUNs once we report
9523          * it to that nexus once.  This bit is required as of SPC-4.
9524          */
9525         eid_ptr->flags4 = SVPD_EID_LUICLR;
9526
9527         /*
9528          * We support revert to defaults (RTD) bit in MODE SELECT.
9529          */
9530         eid_ptr->flags5 = SVPD_EID_RTD_SUP;
9531
9532         /*
9533          * XXX KDM in order to correctly answer this, we would need
9534          * information from the SIM to determine how much sense data it
9535          * can send.  So this would really be a path inquiry field, most
9536          * likely.  This can be set to a maximum of 252 according to SPC-4,
9537          * but the hardware may or may not be able to support that much.
9538          * 0 just means that the maximum sense data length is not reported.
9539          */
9540         eid_ptr->max_sense_length = 0;
9541
9542         ctl_set_success(ctsio);
9543         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9544         ctsio->be_move_done = ctl_config_move_done;
9545         ctl_datamove((union ctl_io *)ctsio);
9546         return (CTL_RETVAL_COMPLETE);
9547 }
9548
9549 static int
9550 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9551 {
9552         struct ctl_lun *lun = CTL_LUN(ctsio);
9553         struct scsi_vpd_mode_page_policy *mpp_ptr;
9554         int data_len;
9555
9556         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9557             sizeof(struct scsi_vpd_mode_page_policy_descr);
9558
9559         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9560         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9561         ctsio->kern_rel_offset = 0;
9562         ctsio->kern_sg_entries = 0;
9563         ctsio->kern_data_len = min(data_len, alloc_len);
9564         ctsio->kern_total_len = ctsio->kern_data_len;
9565
9566         /*
9567          * The control device is always connected.  The disk device, on the
9568          * other hand, may not be online all the time.
9569          */
9570         if (lun != NULL)
9571                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9572                                      lun->be_lun->lun_type;
9573         else
9574                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9575         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9576         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9577         mpp_ptr->descr[0].page_code = 0x3f;
9578         mpp_ptr->descr[0].subpage_code = 0xff;
9579         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9580
9581         ctl_set_success(ctsio);
9582         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9583         ctsio->be_move_done = ctl_config_move_done;
9584         ctl_datamove((union ctl_io *)ctsio);
9585         return (CTL_RETVAL_COMPLETE);
9586 }
9587
9588 /*
9589  * SCSI VPD page 0x83, the Device Identification page.
9590  */
9591 static int
9592 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9593 {
9594         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9595         struct ctl_port *port = CTL_PORT(ctsio);
9596         struct ctl_lun *lun = CTL_LUN(ctsio);
9597         struct scsi_vpd_device_id *devid_ptr;
9598         struct scsi_vpd_id_descriptor *desc;
9599         int data_len, g;
9600         uint8_t proto;
9601
9602         data_len = sizeof(struct scsi_vpd_device_id) +
9603             sizeof(struct scsi_vpd_id_descriptor) +
9604                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9605             sizeof(struct scsi_vpd_id_descriptor) +
9606                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9607         if (lun && lun->lun_devid)
9608                 data_len += lun->lun_devid->len;
9609         if (port && port->port_devid)
9610                 data_len += port->port_devid->len;
9611         if (port && port->target_devid)
9612                 data_len += port->target_devid->len;
9613
9614         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9615         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9616         ctsio->kern_sg_entries = 0;
9617         ctsio->kern_rel_offset = 0;
9618         ctsio->kern_sg_entries = 0;
9619         ctsio->kern_data_len = min(data_len, alloc_len);
9620         ctsio->kern_total_len = ctsio->kern_data_len;
9621
9622         /*
9623          * The control device is always connected.  The disk device, on the
9624          * other hand, may not be online all the time.
9625          */
9626         if (lun != NULL)
9627                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9628                                      lun->be_lun->lun_type;
9629         else
9630                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9631         devid_ptr->page_code = SVPD_DEVICE_ID;
9632         scsi_ulto2b(data_len - 4, devid_ptr->length);
9633
9634         if (port && port->port_type == CTL_PORT_FC)
9635                 proto = SCSI_PROTO_FC << 4;
9636         else if (port && port->port_type == CTL_PORT_SAS)
9637                 proto = SCSI_PROTO_SAS << 4;
9638         else if (port && port->port_type == CTL_PORT_ISCSI)
9639                 proto = SCSI_PROTO_ISCSI << 4;
9640         else
9641                 proto = SCSI_PROTO_SPI << 4;
9642         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9643
9644         /*
9645          * We're using a LUN association here.  i.e., this device ID is a
9646          * per-LUN identifier.
9647          */
9648         if (lun && lun->lun_devid) {
9649                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9650                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9651                     lun->lun_devid->len);
9652         }
9653
9654         /*
9655          * This is for the WWPN which is a port association.
9656          */
9657         if (port && port->port_devid) {
9658                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9659                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9660                     port->port_devid->len);
9661         }
9662
9663         /*
9664          * This is for the Relative Target Port(type 4h) identifier
9665          */
9666         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9667         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9668             SVPD_ID_TYPE_RELTARG;
9669         desc->length = 4;
9670         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9671         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9672             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9673
9674         /*
9675          * This is for the Target Port Group(type 5h) identifier
9676          */
9677         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9678         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9679             SVPD_ID_TYPE_TPORTGRP;
9680         desc->length = 4;
9681         if (softc->is_single ||
9682             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9683                 g = 1;
9684         else
9685                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9686         scsi_ulto2b(g, &desc->identifier[2]);
9687         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9688             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9689
9690         /*
9691          * This is for the Target identifier
9692          */
9693         if (port && port->target_devid) {
9694                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9695         }
9696
9697         ctl_set_success(ctsio);
9698         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9699         ctsio->be_move_done = ctl_config_move_done;
9700         ctl_datamove((union ctl_io *)ctsio);
9701         return (CTL_RETVAL_COMPLETE);
9702 }
9703
9704 static int
9705 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9706 {
9707         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9708         struct ctl_lun *lun = CTL_LUN(ctsio);
9709         struct scsi_vpd_scsi_ports *sp;
9710         struct scsi_vpd_port_designation *pd;
9711         struct scsi_vpd_port_designation_cont *pdc;
9712         struct ctl_port *port;
9713         int data_len, num_target_ports, iid_len, id_len;
9714
9715         num_target_ports = 0;
9716         iid_len = 0;
9717         id_len = 0;
9718         mtx_lock(&softc->ctl_lock);
9719         STAILQ_FOREACH(port, &softc->port_list, links) {
9720                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9721                         continue;
9722                 if (lun != NULL &&
9723                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9724                         continue;
9725                 num_target_ports++;
9726                 if (port->init_devid)
9727                         iid_len += port->init_devid->len;
9728                 if (port->port_devid)
9729                         id_len += port->port_devid->len;
9730         }
9731         mtx_unlock(&softc->ctl_lock);
9732
9733         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9734             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9735              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9736         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9737         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9738         ctsio->kern_sg_entries = 0;
9739         ctsio->kern_rel_offset = 0;
9740         ctsio->kern_sg_entries = 0;
9741         ctsio->kern_data_len = min(data_len, alloc_len);
9742         ctsio->kern_total_len = ctsio->kern_data_len;
9743
9744         /*
9745          * The control device is always connected.  The disk device, on the
9746          * other hand, may not be online all the time.  Need to change this
9747          * to figure out whether the disk device is actually online or not.
9748          */
9749         if (lun != NULL)
9750                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9751                                   lun->be_lun->lun_type;
9752         else
9753                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9754
9755         sp->page_code = SVPD_SCSI_PORTS;
9756         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9757             sp->page_length);
9758         pd = &sp->design[0];
9759
9760         mtx_lock(&softc->ctl_lock);
9761         STAILQ_FOREACH(port, &softc->port_list, links) {
9762                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9763                         continue;
9764                 if (lun != NULL &&
9765                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9766                         continue;
9767                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9768                 if (port->init_devid) {
9769                         iid_len = port->init_devid->len;
9770                         memcpy(pd->initiator_transportid,
9771                             port->init_devid->data, port->init_devid->len);
9772                 } else
9773                         iid_len = 0;
9774                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9775                 pdc = (struct scsi_vpd_port_designation_cont *)
9776                     (&pd->initiator_transportid[iid_len]);
9777                 if (port->port_devid) {
9778                         id_len = port->port_devid->len;
9779                         memcpy(pdc->target_port_descriptors,
9780                             port->port_devid->data, port->port_devid->len);
9781                 } else
9782                         id_len = 0;
9783                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9784                 pd = (struct scsi_vpd_port_designation *)
9785                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9786         }
9787         mtx_unlock(&softc->ctl_lock);
9788
9789         ctl_set_success(ctsio);
9790         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9791         ctsio->be_move_done = ctl_config_move_done;
9792         ctl_datamove((union ctl_io *)ctsio);
9793         return (CTL_RETVAL_COMPLETE);
9794 }
9795
9796 static int
9797 ctl_inquiry_evpd_sfs(struct ctl_scsiio *ctsio, int alloc_len)
9798 {
9799         struct ctl_lun *lun = CTL_LUN(ctsio);
9800         struct scsi_vpd_sfs *sfs_ptr;
9801         int sfs_page_size, n;
9802
9803         sfs_page_size = sizeof(*sfs_ptr) + 5 * 2;
9804         ctsio->kern_data_ptr = malloc(sfs_page_size, M_CTL, M_WAITOK | M_ZERO);
9805         sfs_ptr = (struct scsi_vpd_sfs *)ctsio->kern_data_ptr;
9806         ctsio->kern_sg_entries = 0;
9807         ctsio->kern_rel_offset = 0;
9808         ctsio->kern_sg_entries = 0;
9809         ctsio->kern_data_len = min(sfs_page_size, alloc_len);
9810         ctsio->kern_total_len = ctsio->kern_data_len;
9811
9812         /*
9813          * The control device is always connected.  The disk device, on the
9814          * other hand, may not be online all the time.  Need to change this
9815          * to figure out whether the disk device is actually online or not.
9816          */
9817         if (lun != NULL)
9818                 sfs_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9819                                   lun->be_lun->lun_type;
9820         else
9821                 sfs_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9822
9823         sfs_ptr->page_code = SVPD_SCSI_SFS;
9824         n = 0;
9825         /* Discovery 2016 */
9826         scsi_ulto2b(0x0001, &sfs_ptr->codes[2 * n++]);
9827         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9828                  /* SBC Base 2016 */
9829                 scsi_ulto2b(0x0101, &sfs_ptr->codes[2 * n++]);
9830                  /* SBC Base 2010 */
9831                 scsi_ulto2b(0x0102, &sfs_ptr->codes[2 * n++]);
9832                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9833                         /* Basic Provisioning 2016 */
9834                         scsi_ulto2b(0x0103, &sfs_ptr->codes[2 * n++]);
9835                 }
9836                 /* Drive Maintenance 2016 */
9837                 //scsi_ulto2b(0x0104, &sfs_ptr->codes[2 * n++]);
9838         }
9839         scsi_ulto2b(4 + 2 * n, sfs_ptr->page_length);
9840
9841         ctl_set_success(ctsio);
9842         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9843         ctsio->be_move_done = ctl_config_move_done;
9844         ctl_datamove((union ctl_io *)ctsio);
9845         return (CTL_RETVAL_COMPLETE);
9846 }
9847
9848 static int
9849 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9850 {
9851         struct ctl_lun *lun = CTL_LUN(ctsio);
9852         struct scsi_vpd_block_limits *bl_ptr;
9853         const char *val;
9854         uint64_t ival;
9855
9856         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9857         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9858         ctsio->kern_sg_entries = 0;
9859         ctsio->kern_rel_offset = 0;
9860         ctsio->kern_sg_entries = 0;
9861         ctsio->kern_data_len = min(sizeof(*bl_ptr), alloc_len);
9862         ctsio->kern_total_len = ctsio->kern_data_len;
9863
9864         /*
9865          * The control device is always connected.  The disk device, on the
9866          * other hand, may not be online all the time.  Need to change this
9867          * to figure out whether the disk device is actually online or not.
9868          */
9869         if (lun != NULL)
9870                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9871                                   lun->be_lun->lun_type;
9872         else
9873                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9874
9875         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9876         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9877         bl_ptr->max_cmp_write_len = 0xff;
9878         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9879         if (lun != NULL) {
9880                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9881                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9882                         ival = 0xffffffff;
9883                         val = dnvlist_get_string(lun->be_lun->options,
9884                             "unmap_max_lba", NULL);
9885                         if (val != NULL)
9886                                 ctl_expand_number(val, &ival);
9887                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9888                         ival = 0xffffffff;
9889                         val = dnvlist_get_string(lun->be_lun->options,
9890                             "unmap_max_descr", NULL);
9891                         if (val != NULL)
9892                                 ctl_expand_number(val, &ival);
9893                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9894                         if (lun->be_lun->ublockexp != 0) {
9895                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9896                                     bl_ptr->opt_unmap_grain);
9897                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9898                                     bl_ptr->unmap_grain_align);
9899                         }
9900                 }
9901                 scsi_ulto4b(lun->be_lun->atomicblock,
9902                     bl_ptr->max_atomic_transfer_length);
9903                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9904                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9905                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9906                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9907                 ival = UINT64_MAX;
9908                 val = dnvlist_get_string(lun->be_lun->options,
9909                     "write_same_max_lba", NULL);
9910                 if (val != NULL)
9911                         ctl_expand_number(val, &ival);
9912                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9913                 if (lun->be_lun->maxlba + 1 > ival)
9914                         bl_ptr->flags |= SVPD_BL_WSNZ;
9915         }
9916
9917         ctl_set_success(ctsio);
9918         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9919         ctsio->be_move_done = ctl_config_move_done;
9920         ctl_datamove((union ctl_io *)ctsio);
9921         return (CTL_RETVAL_COMPLETE);
9922 }
9923
9924 static int
9925 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9926 {
9927         struct ctl_lun *lun = CTL_LUN(ctsio);
9928         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9929         const char *value;
9930         u_int i;
9931
9932         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9933         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9934         ctsio->kern_sg_entries = 0;
9935         ctsio->kern_rel_offset = 0;
9936         ctsio->kern_data_len = min(sizeof(*bdc_ptr), alloc_len);
9937         ctsio->kern_total_len = ctsio->kern_data_len;
9938
9939         /*
9940          * The control device is always connected.  The disk device, on the
9941          * other hand, may not be online all the time.  Need to change this
9942          * to figure out whether the disk device is actually online or not.
9943          */
9944         if (lun != NULL)
9945                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9946                                   lun->be_lun->lun_type;
9947         else
9948                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9949         bdc_ptr->page_code = SVPD_BDC;
9950         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9951         if (lun != NULL &&
9952             (value = dnvlist_get_string(lun->be_lun->options, "rpm", NULL)) != NULL)
9953                 i = strtol(value, NULL, 0);
9954         else
9955                 i = CTL_DEFAULT_ROTATION_RATE;
9956         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9957         if (lun != NULL &&
9958             (value = dnvlist_get_string(lun->be_lun->options, "formfactor", NULL)) != NULL)
9959                 i = strtol(value, NULL, 0);
9960         else
9961                 i = 0;
9962         bdc_ptr->wab_wac_ff = (i & 0x0f);
9963         bdc_ptr->flags = SVPD_RBWZ | SVPD_FUAB | SVPD_VBULS;
9964
9965         ctl_set_success(ctsio);
9966         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9967         ctsio->be_move_done = ctl_config_move_done;
9968         ctl_datamove((union ctl_io *)ctsio);
9969         return (CTL_RETVAL_COMPLETE);
9970 }
9971
9972 static int
9973 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9974 {
9975         struct ctl_lun *lun = CTL_LUN(ctsio);
9976         struct scsi_vpd_logical_block_prov *lbp_ptr;
9977         const char *value;
9978
9979         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9980         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9981         ctsio->kern_sg_entries = 0;
9982         ctsio->kern_rel_offset = 0;
9983         ctsio->kern_data_len = min(sizeof(*lbp_ptr), alloc_len);
9984         ctsio->kern_total_len = ctsio->kern_data_len;
9985
9986         /*
9987          * The control device is always connected.  The disk device, on the
9988          * other hand, may not be online all the time.  Need to change this
9989          * to figure out whether the disk device is actually online or not.
9990          */
9991         if (lun != NULL)
9992                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9993                                   lun->be_lun->lun_type;
9994         else
9995                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9996
9997         lbp_ptr->page_code = SVPD_LBP;
9998         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9999         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10000         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10001                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10002                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10003                 value = dnvlist_get_string(lun->be_lun->options,
10004                     "provisioning_type", NULL);
10005                 if (value != NULL) {
10006                         if (strcmp(value, "resource") == 0)
10007                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10008                         else if (strcmp(value, "thin") == 0)
10009                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
10010                 } else
10011                         lbp_ptr->prov_type = SVPD_LBP_THIN;
10012         }
10013
10014         ctl_set_success(ctsio);
10015         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10016         ctsio->be_move_done = ctl_config_move_done;
10017         ctl_datamove((union ctl_io *)ctsio);
10018         return (CTL_RETVAL_COMPLETE);
10019 }
10020
10021 /*
10022  * INQUIRY with the EVPD bit set.
10023  */
10024 static int
10025 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10026 {
10027         struct ctl_lun *lun = CTL_LUN(ctsio);
10028         struct scsi_inquiry *cdb;
10029         int alloc_len, retval;
10030
10031         cdb = (struct scsi_inquiry *)ctsio->cdb;
10032         alloc_len = scsi_2btoul(cdb->length);
10033
10034         switch (cdb->page_code) {
10035         case SVPD_SUPPORTED_PAGES:
10036                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10037                 break;
10038         case SVPD_UNIT_SERIAL_NUMBER:
10039                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10040                 break;
10041         case SVPD_DEVICE_ID:
10042                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10043                 break;
10044         case SVPD_EXTENDED_INQUIRY_DATA:
10045                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10046                 break;
10047         case SVPD_MODE_PAGE_POLICY:
10048                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10049                 break;
10050         case SVPD_SCSI_PORTS:
10051                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10052                 break;
10053         case SVPD_SCSI_TPC:
10054                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10055                 break;
10056         case SVPD_SCSI_SFS:
10057                 retval = ctl_inquiry_evpd_sfs(ctsio, alloc_len);
10058                 break;
10059         case SVPD_BLOCK_LIMITS:
10060                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10061                         goto err;
10062                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10063                 break;
10064         case SVPD_BDC:
10065                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10066                         goto err;
10067                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10068                 break;
10069         case SVPD_LBP:
10070                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10071                         goto err;
10072                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10073                 break;
10074         default:
10075 err:
10076                 ctl_set_invalid_field(ctsio,
10077                                       /*sks_valid*/ 1,
10078                                       /*command*/ 1,
10079                                       /*field*/ 2,
10080                                       /*bit_valid*/ 0,
10081                                       /*bit*/ 0);
10082                 ctl_done((union ctl_io *)ctsio);
10083                 retval = CTL_RETVAL_COMPLETE;
10084                 break;
10085         }
10086
10087         return (retval);
10088 }
10089
10090 /*
10091  * Standard INQUIRY data.
10092  */
10093 static int
10094 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10095 {
10096         struct ctl_softc *softc = CTL_SOFTC(ctsio);
10097         struct ctl_port *port = CTL_PORT(ctsio);
10098         struct ctl_lun *lun = CTL_LUN(ctsio);
10099         struct scsi_inquiry_data *inq_ptr;
10100         struct scsi_inquiry *cdb;
10101         const char *val;
10102         uint32_t alloc_len, data_len;
10103         ctl_port_type port_type;
10104
10105         port_type = port->port_type;
10106         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10107                 port_type = CTL_PORT_SCSI;
10108
10109         cdb = (struct scsi_inquiry *)ctsio->cdb;
10110         alloc_len = scsi_2btoul(cdb->length);
10111
10112         /*
10113          * We malloc the full inquiry data size here and fill it
10114          * in.  If the user only asks for less, we'll give him
10115          * that much.
10116          */
10117         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10118         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10119         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10120         ctsio->kern_sg_entries = 0;
10121         ctsio->kern_rel_offset = 0;
10122         ctsio->kern_data_len = min(data_len, alloc_len);
10123         ctsio->kern_total_len = ctsio->kern_data_len;
10124
10125         if (lun != NULL) {
10126                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10127                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10128                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10129                             lun->be_lun->lun_type;
10130                 } else {
10131                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10132                             lun->be_lun->lun_type;
10133                 }
10134                 if (lun->flags & CTL_LUN_REMOVABLE)
10135                         inq_ptr->dev_qual2 |= SID_RMB;
10136         } else
10137                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10138
10139         /* RMB in byte 2 is 0 */
10140         inq_ptr->version = SCSI_REV_SPC5;
10141
10142         /*
10143          * According to SAM-3, even if a device only supports a single
10144          * level of LUN addressing, it should still set the HISUP bit:
10145          *
10146          * 4.9.1 Logical unit numbers overview
10147          *
10148          * All logical unit number formats described in this standard are
10149          * hierarchical in structure even when only a single level in that
10150          * hierarchy is used. The HISUP bit shall be set to one in the
10151          * standard INQUIRY data (see SPC-2) when any logical unit number
10152          * format described in this standard is used.  Non-hierarchical
10153          * formats are outside the scope of this standard.
10154          *
10155          * Therefore we set the HiSup bit here.
10156          *
10157          * The response format is 2, per SPC-3.
10158          */
10159         inq_ptr->response_format = SID_HiSup | 2;
10160
10161         inq_ptr->additional_length = data_len -
10162             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10163         CTL_DEBUG_PRINT(("additional_length = %d\n",
10164                          inq_ptr->additional_length));
10165
10166         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10167         if (port_type == CTL_PORT_SCSI)
10168                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10169         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10170         inq_ptr->flags = SID_CmdQue;
10171         if (port_type == CTL_PORT_SCSI)
10172                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10173
10174         /*
10175          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10176          * We have 8 bytes for the vendor name, and 16 bytes for the device
10177          * name and 4 bytes for the revision.
10178          */
10179         if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options,
10180             "vendor", NULL)) == NULL) {
10181                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10182         } else {
10183                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10184                 strncpy(inq_ptr->vendor, val,
10185                     min(sizeof(inq_ptr->vendor), strlen(val)));
10186         }
10187         if (lun == NULL) {
10188                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10189                     sizeof(inq_ptr->product));
10190         } else if ((val = dnvlist_get_string(lun->be_lun->options, "product",
10191             NULL)) == NULL) {
10192                 switch (lun->be_lun->lun_type) {
10193                 case T_DIRECT:
10194                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10195                             sizeof(inq_ptr->product));
10196                         break;
10197                 case T_PROCESSOR:
10198                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10199                             sizeof(inq_ptr->product));
10200                         break;
10201                 case T_CDROM:
10202                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10203                             sizeof(inq_ptr->product));
10204                         break;
10205                 default:
10206                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10207                             sizeof(inq_ptr->product));
10208                         break;
10209                 }
10210         } else {
10211                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10212                 strncpy(inq_ptr->product, val,
10213                     min(sizeof(inq_ptr->product), strlen(val)));
10214         }
10215
10216         /*
10217          * XXX make this a macro somewhere so it automatically gets
10218          * incremented when we make changes.
10219          */
10220         if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options,
10221             "revision", NULL)) == NULL) {
10222                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10223         } else {
10224                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10225                 strncpy(inq_ptr->revision, val,
10226                     min(sizeof(inq_ptr->revision), strlen(val)));
10227         }
10228
10229         /*
10230          * For parallel SCSI, we support double transition and single
10231          * transition clocking.  We also support QAS (Quick Arbitration
10232          * and Selection) and Information Unit transfers on both the
10233          * control and array devices.
10234          */
10235         if (port_type == CTL_PORT_SCSI)
10236                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10237                                     SID_SPI_IUS;
10238
10239         /* SAM-6 (no version claimed) */
10240         scsi_ulto2b(0x00C0, inq_ptr->version1);
10241         /* SPC-5 (no version claimed) */
10242         scsi_ulto2b(0x05C0, inq_ptr->version2);
10243         if (port_type == CTL_PORT_FC) {
10244                 /* FCP-2 ANSI INCITS.350:2003 */
10245                 scsi_ulto2b(0x0917, inq_ptr->version3);
10246         } else if (port_type == CTL_PORT_SCSI) {
10247                 /* SPI-4 ANSI INCITS.362:200x */
10248                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10249         } else if (port_type == CTL_PORT_ISCSI) {
10250                 /* iSCSI (no version claimed) */
10251                 scsi_ulto2b(0x0960, inq_ptr->version3);
10252         } else if (port_type == CTL_PORT_SAS) {
10253                 /* SAS (no version claimed) */
10254                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10255         } else if (port_type == CTL_PORT_UMASS) {
10256                 /* USB Mass Storage Class Bulk-Only Transport, Revision 1.0 */
10257                 scsi_ulto2b(0x1730, inq_ptr->version3);
10258         }
10259
10260         if (lun == NULL) {
10261                 /* SBC-4 (no version claimed) */
10262                 scsi_ulto2b(0x0600, inq_ptr->version4);
10263         } else {
10264                 switch (lun->be_lun->lun_type) {
10265                 case T_DIRECT:
10266                         /* SBC-4 (no version claimed) */
10267                         scsi_ulto2b(0x0600, inq_ptr->version4);
10268                         break;
10269                 case T_PROCESSOR:
10270                         break;
10271                 case T_CDROM:
10272                         /* MMC-6 (no version claimed) */
10273                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10274                         break;
10275                 default:
10276                         break;
10277                 }
10278         }
10279
10280         ctl_set_success(ctsio);
10281         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10282         ctsio->be_move_done = ctl_config_move_done;
10283         ctl_datamove((union ctl_io *)ctsio);
10284         return (CTL_RETVAL_COMPLETE);
10285 }
10286
10287 int
10288 ctl_inquiry(struct ctl_scsiio *ctsio)
10289 {
10290         struct scsi_inquiry *cdb;
10291         int retval;
10292
10293         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10294
10295         cdb = (struct scsi_inquiry *)ctsio->cdb;
10296         if (cdb->byte2 & SI_EVPD)
10297                 retval = ctl_inquiry_evpd(ctsio);
10298         else if (cdb->page_code == 0)
10299                 retval = ctl_inquiry_std(ctsio);
10300         else {
10301                 ctl_set_invalid_field(ctsio,
10302                                       /*sks_valid*/ 1,
10303                                       /*command*/ 1,
10304                                       /*field*/ 2,
10305                                       /*bit_valid*/ 0,
10306                                       /*bit*/ 0);
10307                 ctl_done((union ctl_io *)ctsio);
10308                 return (CTL_RETVAL_COMPLETE);
10309         }
10310
10311         return (retval);
10312 }
10313
10314 int
10315 ctl_get_config(struct ctl_scsiio *ctsio)
10316 {
10317         struct ctl_lun *lun = CTL_LUN(ctsio);
10318         struct scsi_get_config_header *hdr;
10319         struct scsi_get_config_feature *feature;
10320         struct scsi_get_config *cdb;
10321         uint32_t alloc_len, data_len;
10322         int rt, starting;
10323
10324         cdb = (struct scsi_get_config *)ctsio->cdb;
10325         rt = (cdb->rt & SGC_RT_MASK);
10326         starting = scsi_2btoul(cdb->starting_feature);
10327         alloc_len = scsi_2btoul(cdb->length);
10328
10329         data_len = sizeof(struct scsi_get_config_header) +
10330             sizeof(struct scsi_get_config_feature) + 8 +
10331             sizeof(struct scsi_get_config_feature) + 8 +
10332             sizeof(struct scsi_get_config_feature) + 4 +
10333             sizeof(struct scsi_get_config_feature) + 4 +
10334             sizeof(struct scsi_get_config_feature) + 8 +
10335             sizeof(struct scsi_get_config_feature) +
10336             sizeof(struct scsi_get_config_feature) + 4 +
10337             sizeof(struct scsi_get_config_feature) + 4 +
10338             sizeof(struct scsi_get_config_feature) + 4 +
10339             sizeof(struct scsi_get_config_feature) + 4 +
10340             sizeof(struct scsi_get_config_feature) + 4 +
10341             sizeof(struct scsi_get_config_feature) + 4;
10342         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10343         ctsio->kern_sg_entries = 0;
10344         ctsio->kern_rel_offset = 0;
10345
10346         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10347         if (lun->flags & CTL_LUN_NO_MEDIA)
10348                 scsi_ulto2b(0x0000, hdr->current_profile);
10349         else
10350                 scsi_ulto2b(0x0010, hdr->current_profile);
10351         feature = (struct scsi_get_config_feature *)(hdr + 1);
10352
10353         if (starting > 0x003b)
10354                 goto done;
10355         if (starting > 0x003a)
10356                 goto f3b;
10357         if (starting > 0x002b)
10358                 goto f3a;
10359         if (starting > 0x002a)
10360                 goto f2b;
10361         if (starting > 0x001f)
10362                 goto f2a;
10363         if (starting > 0x001e)
10364                 goto f1f;
10365         if (starting > 0x001d)
10366                 goto f1e;
10367         if (starting > 0x0010)
10368                 goto f1d;
10369         if (starting > 0x0003)
10370                 goto f10;
10371         if (starting > 0x0002)
10372                 goto f3;
10373         if (starting > 0x0001)
10374                 goto f2;
10375         if (starting > 0x0000)
10376                 goto f1;
10377
10378         /* Profile List */
10379         scsi_ulto2b(0x0000, feature->feature_code);
10380         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10381         feature->add_length = 8;
10382         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10383         feature->feature_data[2] = 0x00;
10384         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10385         feature->feature_data[6] = 0x01;
10386         feature = (struct scsi_get_config_feature *)
10387             &feature->feature_data[feature->add_length];
10388
10389 f1:     /* Core */
10390         scsi_ulto2b(0x0001, feature->feature_code);
10391         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10392         feature->add_length = 8;
10393         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10394         feature->feature_data[4] = 0x03;
10395         feature = (struct scsi_get_config_feature *)
10396             &feature->feature_data[feature->add_length];
10397
10398 f2:     /* Morphing */
10399         scsi_ulto2b(0x0002, feature->feature_code);
10400         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10401         feature->add_length = 4;
10402         feature->feature_data[0] = 0x02;
10403         feature = (struct scsi_get_config_feature *)
10404             &feature->feature_data[feature->add_length];
10405
10406 f3:     /* Removable Medium */
10407         scsi_ulto2b(0x0003, feature->feature_code);
10408         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10409         feature->add_length = 4;
10410         feature->feature_data[0] = 0x39;
10411         feature = (struct scsi_get_config_feature *)
10412             &feature->feature_data[feature->add_length];
10413
10414         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10415                 goto done;
10416
10417 f10:    /* Random Read */
10418         scsi_ulto2b(0x0010, feature->feature_code);
10419         feature->flags = 0x00;
10420         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10421                 feature->flags |= SGC_F_CURRENT;
10422         feature->add_length = 8;
10423         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10424         scsi_ulto2b(1, &feature->feature_data[4]);
10425         feature->feature_data[6] = 0x00;
10426         feature = (struct scsi_get_config_feature *)
10427             &feature->feature_data[feature->add_length];
10428
10429 f1d:    /* Multi-Read */
10430         scsi_ulto2b(0x001D, feature->feature_code);
10431         feature->flags = 0x00;
10432         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10433                 feature->flags |= SGC_F_CURRENT;
10434         feature->add_length = 0;
10435         feature = (struct scsi_get_config_feature *)
10436             &feature->feature_data[feature->add_length];
10437
10438 f1e:    /* CD Read */
10439         scsi_ulto2b(0x001E, feature->feature_code);
10440         feature->flags = 0x00;
10441         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10442                 feature->flags |= SGC_F_CURRENT;
10443         feature->add_length = 4;
10444         feature->feature_data[0] = 0x00;
10445         feature = (struct scsi_get_config_feature *)
10446             &feature->feature_data[feature->add_length];
10447
10448 f1f:    /* DVD Read */
10449         scsi_ulto2b(0x001F, feature->feature_code);
10450         feature->flags = 0x08;
10451         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10452                 feature->flags |= SGC_F_CURRENT;
10453         feature->add_length = 4;
10454         feature->feature_data[0] = 0x01;
10455         feature->feature_data[2] = 0x03;
10456         feature = (struct scsi_get_config_feature *)
10457             &feature->feature_data[feature->add_length];
10458
10459 f2a:    /* DVD+RW */
10460         scsi_ulto2b(0x002A, feature->feature_code);
10461         feature->flags = 0x04;
10462         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10463                 feature->flags |= SGC_F_CURRENT;
10464         feature->add_length = 4;
10465         feature->feature_data[0] = 0x00;
10466         feature->feature_data[1] = 0x00;
10467         feature = (struct scsi_get_config_feature *)
10468             &feature->feature_data[feature->add_length];
10469
10470 f2b:    /* DVD+R */
10471         scsi_ulto2b(0x002B, feature->feature_code);
10472         feature->flags = 0x00;
10473         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10474                 feature->flags |= SGC_F_CURRENT;
10475         feature->add_length = 4;
10476         feature->feature_data[0] = 0x00;
10477         feature = (struct scsi_get_config_feature *)
10478             &feature->feature_data[feature->add_length];
10479
10480 f3a:    /* DVD+RW Dual Layer */
10481         scsi_ulto2b(0x003A, feature->feature_code);
10482         feature->flags = 0x00;
10483         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10484                 feature->flags |= SGC_F_CURRENT;
10485         feature->add_length = 4;
10486         feature->feature_data[0] = 0x00;
10487         feature->feature_data[1] = 0x00;
10488         feature = (struct scsi_get_config_feature *)
10489             &feature->feature_data[feature->add_length];
10490
10491 f3b:    /* DVD+R Dual Layer */
10492         scsi_ulto2b(0x003B, feature->feature_code);
10493         feature->flags = 0x00;
10494         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10495                 feature->flags |= SGC_F_CURRENT;
10496         feature->add_length = 4;
10497         feature->feature_data[0] = 0x00;
10498         feature = (struct scsi_get_config_feature *)
10499             &feature->feature_data[feature->add_length];
10500
10501 done:
10502         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10503         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10504                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10505                 if (scsi_2btoul(feature->feature_code) == starting)
10506                         feature = (struct scsi_get_config_feature *)
10507                             &feature->feature_data[feature->add_length];
10508                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10509         }
10510         scsi_ulto4b(data_len - 4, hdr->data_length);
10511         ctsio->kern_data_len = min(data_len, alloc_len);
10512         ctsio->kern_total_len = ctsio->kern_data_len;
10513
10514         ctl_set_success(ctsio);
10515         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10516         ctsio->be_move_done = ctl_config_move_done;
10517         ctl_datamove((union ctl_io *)ctsio);
10518         return (CTL_RETVAL_COMPLETE);
10519 }
10520
10521 int
10522 ctl_get_event_status(struct ctl_scsiio *ctsio)
10523 {
10524         struct scsi_get_event_status_header *hdr;
10525         struct scsi_get_event_status *cdb;
10526         uint32_t alloc_len, data_len;
10527
10528         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10529         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10530                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10531                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10532                 ctl_done((union ctl_io *)ctsio);
10533                 return (CTL_RETVAL_COMPLETE);
10534         }
10535         alloc_len = scsi_2btoul(cdb->length);
10536
10537         data_len = sizeof(struct scsi_get_event_status_header);
10538         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10539         ctsio->kern_sg_entries = 0;
10540         ctsio->kern_rel_offset = 0;
10541         ctsio->kern_data_len = min(data_len, alloc_len);
10542         ctsio->kern_total_len = ctsio->kern_data_len;
10543
10544         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10545         scsi_ulto2b(0, hdr->descr_length);
10546         hdr->nea_class = SGESN_NEA;
10547         hdr->supported_class = 0;
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 int
10557 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10558 {
10559         struct scsi_mechanism_status_header *hdr;
10560         struct scsi_mechanism_status *cdb;
10561         uint32_t alloc_len, data_len;
10562
10563         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10564         alloc_len = scsi_2btoul(cdb->length);
10565
10566         data_len = sizeof(struct scsi_mechanism_status_header);
10567         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10568         ctsio->kern_sg_entries = 0;
10569         ctsio->kern_rel_offset = 0;
10570         ctsio->kern_data_len = min(data_len, alloc_len);
10571         ctsio->kern_total_len = ctsio->kern_data_len;
10572
10573         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10574         hdr->state1 = 0x00;
10575         hdr->state2 = 0xe0;
10576         scsi_ulto3b(0, hdr->lba);
10577         hdr->slots_num = 0;
10578         scsi_ulto2b(0, hdr->slots_length);
10579
10580         ctl_set_success(ctsio);
10581         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10582         ctsio->be_move_done = ctl_config_move_done;
10583         ctl_datamove((union ctl_io *)ctsio);
10584         return (CTL_RETVAL_COMPLETE);
10585 }
10586
10587 static void
10588 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10589 {
10590
10591         lba += 150;
10592         buf[0] = 0;
10593         buf[1] = bin2bcd((lba / 75) / 60);
10594         buf[2] = bin2bcd((lba / 75) % 60);
10595         buf[3] = bin2bcd(lba % 75);
10596 }
10597
10598 int
10599 ctl_read_toc(struct ctl_scsiio *ctsio)
10600 {
10601         struct ctl_lun *lun = CTL_LUN(ctsio);
10602         struct scsi_read_toc_hdr *hdr;
10603         struct scsi_read_toc_type01_descr *descr;
10604         struct scsi_read_toc *cdb;
10605         uint32_t alloc_len, data_len;
10606         int format, msf;
10607
10608         cdb = (struct scsi_read_toc *)ctsio->cdb;
10609         msf = (cdb->byte2 & CD_MSF) != 0;
10610         format = cdb->format;
10611         alloc_len = scsi_2btoul(cdb->data_len);
10612
10613         data_len = sizeof(struct scsi_read_toc_hdr);
10614         if (format == 0)
10615                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10616         else
10617                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10618         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10619         ctsio->kern_sg_entries = 0;
10620         ctsio->kern_rel_offset = 0;
10621         ctsio->kern_data_len = min(data_len, alloc_len);
10622         ctsio->kern_total_len = ctsio->kern_data_len;
10623
10624         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10625         if (format == 0) {
10626                 scsi_ulto2b(0x12, hdr->data_length);
10627                 hdr->first = 1;
10628                 hdr->last = 1;
10629                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10630                 descr->addr_ctl = 0x14;
10631                 descr->track_number = 1;
10632                 if (msf)
10633                         ctl_ultomsf(0, descr->track_start);
10634                 else
10635                         scsi_ulto4b(0, descr->track_start);
10636                 descr++;
10637                 descr->addr_ctl = 0x14;
10638                 descr->track_number = 0xaa;
10639                 if (msf)
10640                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10641                 else
10642                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10643         } else {
10644                 scsi_ulto2b(0x0a, hdr->data_length);
10645                 hdr->first = 1;
10646                 hdr->last = 1;
10647                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10648                 descr->addr_ctl = 0x14;
10649                 descr->track_number = 1;
10650                 if (msf)
10651                         ctl_ultomsf(0, descr->track_start);
10652                 else
10653                         scsi_ulto4b(0, descr->track_start);
10654         }
10655
10656         ctl_set_success(ctsio);
10657         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10658         ctsio->be_move_done = ctl_config_move_done;
10659         ctl_datamove((union ctl_io *)ctsio);
10660         return (CTL_RETVAL_COMPLETE);
10661 }
10662
10663 /*
10664  * For known CDB types, parse the LBA and length.
10665  */
10666 static int
10667 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10668 {
10669         if (io->io_hdr.io_type != CTL_IO_SCSI)
10670                 return (1);
10671
10672         switch (io->scsiio.cdb[0]) {
10673         case COMPARE_AND_WRITE: {
10674                 struct scsi_compare_and_write *cdb;
10675
10676                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10677
10678                 *lba = scsi_8btou64(cdb->addr);
10679                 *len = cdb->length;
10680                 break;
10681         }
10682         case READ_6:
10683         case WRITE_6: {
10684                 struct scsi_rw_6 *cdb;
10685
10686                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10687
10688                 *lba = scsi_3btoul(cdb->addr);
10689                 /* only 5 bits are valid in the most significant address byte */
10690                 *lba &= 0x1fffff;
10691                 *len = cdb->length;
10692                 break;
10693         }
10694         case READ_10:
10695         case WRITE_10: {
10696                 struct scsi_rw_10 *cdb;
10697
10698                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10699
10700                 *lba = scsi_4btoul(cdb->addr);
10701                 *len = scsi_2btoul(cdb->length);
10702                 break;
10703         }
10704         case WRITE_VERIFY_10: {
10705                 struct scsi_write_verify_10 *cdb;
10706
10707                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10708
10709                 *lba = scsi_4btoul(cdb->addr);
10710                 *len = scsi_2btoul(cdb->length);
10711                 break;
10712         }
10713         case READ_12:
10714         case WRITE_12: {
10715                 struct scsi_rw_12 *cdb;
10716
10717                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10718
10719                 *lba = scsi_4btoul(cdb->addr);
10720                 *len = scsi_4btoul(cdb->length);
10721                 break;
10722         }
10723         case WRITE_VERIFY_12: {
10724                 struct scsi_write_verify_12 *cdb;
10725
10726                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10727
10728                 *lba = scsi_4btoul(cdb->addr);
10729                 *len = scsi_4btoul(cdb->length);
10730                 break;
10731         }
10732         case READ_16:
10733         case WRITE_16: {
10734                 struct scsi_rw_16 *cdb;
10735
10736                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10737
10738                 *lba = scsi_8btou64(cdb->addr);
10739                 *len = scsi_4btoul(cdb->length);
10740                 break;
10741         }
10742         case WRITE_ATOMIC_16: {
10743                 struct scsi_write_atomic_16 *cdb;
10744
10745                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10746
10747                 *lba = scsi_8btou64(cdb->addr);
10748                 *len = scsi_2btoul(cdb->length);
10749                 break;
10750         }
10751         case WRITE_VERIFY_16: {
10752                 struct scsi_write_verify_16 *cdb;
10753
10754                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10755
10756                 *lba = scsi_8btou64(cdb->addr);
10757                 *len = scsi_4btoul(cdb->length);
10758                 break;
10759         }
10760         case WRITE_SAME_10: {
10761                 struct scsi_write_same_10 *cdb;
10762
10763                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10764
10765                 *lba = scsi_4btoul(cdb->addr);
10766                 *len = scsi_2btoul(cdb->length);
10767                 break;
10768         }
10769         case WRITE_SAME_16: {
10770                 struct scsi_write_same_16 *cdb;
10771
10772                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10773
10774                 *lba = scsi_8btou64(cdb->addr);
10775                 *len = scsi_4btoul(cdb->length);
10776                 break;
10777         }
10778         case VERIFY_10: {
10779                 struct scsi_verify_10 *cdb;
10780
10781                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10782
10783                 *lba = scsi_4btoul(cdb->addr);
10784                 *len = scsi_2btoul(cdb->length);
10785                 break;
10786         }
10787         case VERIFY_12: {
10788                 struct scsi_verify_12 *cdb;
10789
10790                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10791
10792                 *lba = scsi_4btoul(cdb->addr);
10793                 *len = scsi_4btoul(cdb->length);
10794                 break;
10795         }
10796         case VERIFY_16: {
10797                 struct scsi_verify_16 *cdb;
10798
10799                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10800
10801                 *lba = scsi_8btou64(cdb->addr);
10802                 *len = scsi_4btoul(cdb->length);
10803                 break;
10804         }
10805         case UNMAP: {
10806                 *lba = 0;
10807                 *len = UINT64_MAX;
10808                 break;
10809         }
10810         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10811                 struct scsi_get_lba_status *cdb;
10812
10813                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10814                 *lba = scsi_8btou64(cdb->addr);
10815                 *len = UINT32_MAX;
10816                 break;
10817         }
10818         default:
10819                 return (1);
10820                 break; /* NOTREACHED */
10821         }
10822
10823         return (0);
10824 }
10825
10826 static ctl_action
10827 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10828     bool seq)
10829 {
10830         uint64_t endlba1, endlba2;
10831
10832         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10833         endlba2 = lba2 + len2 - 1;
10834
10835         if ((endlba1 < lba2) || (endlba2 < lba1))
10836                 return (CTL_ACTION_PASS);
10837         else
10838                 return (CTL_ACTION_BLOCK);
10839 }
10840
10841 static int
10842 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10843 {
10844         struct ctl_ptr_len_flags *ptrlen;
10845         struct scsi_unmap_desc *buf, *end, *range;
10846         uint64_t lba;
10847         uint32_t len;
10848
10849         /* If not UNMAP -- go other way. */
10850         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10851             io->scsiio.cdb[0] != UNMAP)
10852                 return (CTL_ACTION_ERROR);
10853
10854         /* If UNMAP without data -- block and wait for data. */
10855         ptrlen = (struct ctl_ptr_len_flags *)
10856             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10857         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10858             ptrlen->ptr == NULL)
10859                 return (CTL_ACTION_BLOCK);
10860
10861         /* UNMAP with data -- check for collision. */
10862         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10863         end = buf + ptrlen->len / sizeof(*buf);
10864         for (range = buf; range < end; range++) {
10865                 lba = scsi_8btou64(range->lba);
10866                 len = scsi_4btoul(range->length);
10867                 if ((lba < lba2 + len2) && (lba + len > lba2))
10868                         return (CTL_ACTION_BLOCK);
10869         }
10870         return (CTL_ACTION_PASS);
10871 }
10872
10873 static ctl_action
10874 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10875 {
10876         uint64_t lba1, lba2;
10877         uint64_t len1, len2;
10878         int retval;
10879
10880         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10881                 return (CTL_ACTION_ERROR);
10882
10883         retval = ctl_extent_check_unmap(io1, lba2, len2);
10884         if (retval != CTL_ACTION_ERROR)
10885                 return (retval);
10886
10887         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10888                 return (CTL_ACTION_ERROR);
10889
10890         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10891                 seq = FALSE;
10892         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10893 }
10894
10895 static ctl_action
10896 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10897 {
10898         uint64_t lba1, lba2;
10899         uint64_t len1, len2;
10900
10901         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10902                 return (CTL_ACTION_PASS);
10903         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10904                 return (CTL_ACTION_ERROR);
10905         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10906                 return (CTL_ACTION_ERROR);
10907
10908         if (lba1 + len1 == lba2)
10909                 return (CTL_ACTION_BLOCK);
10910         return (CTL_ACTION_PASS);
10911 }
10912
10913 static ctl_action
10914 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10915     union ctl_io *ooa_io)
10916 {
10917         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10918         const ctl_serialize_action *serialize_row;
10919
10920         /*
10921          * Aborted commands are not going to be executed and may even
10922          * not report completion, so we don't care about their order.
10923          * Let them complete ASAP to clean the OOA queue.
10924          */
10925         if (pending_io->io_hdr.flags & CTL_FLAG_ABORT)
10926                 return (CTL_ACTION_SKIP);
10927
10928         /*
10929          * The initiator attempted multiple untagged commands at the same
10930          * time.  Can't do that.
10931          */
10932         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10933          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10934          && ((pending_io->io_hdr.nexus.targ_port ==
10935               ooa_io->io_hdr.nexus.targ_port)
10936           && (pending_io->io_hdr.nexus.initid ==
10937               ooa_io->io_hdr.nexus.initid))
10938          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10939               CTL_FLAG_STATUS_SENT)) == 0))
10940                 return (CTL_ACTION_OVERLAP);
10941
10942         /*
10943          * The initiator attempted to send multiple tagged commands with
10944          * the same ID.  (It's fine if different initiators have the same
10945          * tag ID.)
10946          *
10947          * Even if all of those conditions are true, we don't kill the I/O
10948          * if the command ahead of us has been aborted.  We won't end up
10949          * sending it to the FETD, and it's perfectly legal to resend a
10950          * command with the same tag number as long as the previous
10951          * instance of this tag number has been aborted somehow.
10952          */
10953         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10954          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10955          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10956          && ((pending_io->io_hdr.nexus.targ_port ==
10957               ooa_io->io_hdr.nexus.targ_port)
10958           && (pending_io->io_hdr.nexus.initid ==
10959               ooa_io->io_hdr.nexus.initid))
10960          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10961               CTL_FLAG_STATUS_SENT)) == 0))
10962                 return (CTL_ACTION_OVERLAP_TAG);
10963
10964         /*
10965          * If we get a head of queue tag, SAM-3 says that we should
10966          * immediately execute it.
10967          *
10968          * What happens if this command would normally block for some other
10969          * reason?  e.g. a request sense with a head of queue tag
10970          * immediately after a write.  Normally that would block, but this
10971          * will result in its getting executed immediately...
10972          *
10973          * We currently return "pass" instead of "skip", so we'll end up
10974          * going through the rest of the queue to check for overlapped tags.
10975          *
10976          * XXX KDM check for other types of blockage first??
10977          */
10978         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10979                 return (CTL_ACTION_PASS);
10980
10981         /*
10982          * Ordered tags have to block until all items ahead of them
10983          * have completed.  If we get called with an ordered tag, we always
10984          * block, if something else is ahead of us in the queue.
10985          */
10986         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10987                 return (CTL_ACTION_BLOCK);
10988
10989         /*
10990          * Simple tags get blocked until all head of queue and ordered tags
10991          * ahead of them have completed.  I'm lumping untagged commands in
10992          * with simple tags here.  XXX KDM is that the right thing to do?
10993          */
10994         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10995           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10996          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10997           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10998                 return (CTL_ACTION_BLOCK);
10999
11000         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11001         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
11002             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
11003              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
11004              pending_io->scsiio.cdb[1], pending_io));
11005         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11006         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
11007                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
11008         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
11009             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
11010              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
11011              ooa_io->scsiio.cdb[1], ooa_io));
11012
11013         serialize_row = ctl_serialize_table[ooa_entry->seridx];
11014
11015         switch (serialize_row[pending_entry->seridx]) {
11016         case CTL_SER_BLOCK:
11017                 return (CTL_ACTION_BLOCK);
11018         case CTL_SER_EXTENT:
11019                 return (ctl_extent_check(ooa_io, pending_io,
11020                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11021         case CTL_SER_EXTENTOPT:
11022                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
11023                     SCP_QUEUE_ALG_UNRESTRICTED)
11024                         return (ctl_extent_check(ooa_io, pending_io,
11025                             (lun->be_lun &&
11026                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11027                 return (CTL_ACTION_PASS);
11028         case CTL_SER_EXTENTSEQ:
11029                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
11030                         return (ctl_extent_check_seq(ooa_io, pending_io));
11031                 return (CTL_ACTION_PASS);
11032         case CTL_SER_PASS:
11033                 return (CTL_ACTION_PASS);
11034         case CTL_SER_BLOCKOPT:
11035                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
11036                     SCP_QUEUE_ALG_UNRESTRICTED)
11037                         return (CTL_ACTION_BLOCK);
11038                 return (CTL_ACTION_PASS);
11039         case CTL_SER_SKIP:
11040                 return (CTL_ACTION_SKIP);
11041         default:
11042                 panic("%s: Invalid serialization value %d for %d => %d",
11043                     __func__, serialize_row[pending_entry->seridx],
11044                     pending_entry->seridx, ooa_entry->seridx);
11045         }
11046
11047         return (CTL_ACTION_ERROR);
11048 }
11049
11050 /*
11051  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11052  * Assumptions:
11053  * - pending_io is generally either incoming, or on the blocked queue
11054  * - starting I/O is the I/O we want to start the check with.
11055  */
11056 static ctl_action
11057 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11058               union ctl_io **starting_io)
11059 {
11060         union ctl_io *ooa_io;
11061         ctl_action action;
11062
11063         mtx_assert(&lun->lun_lock, MA_OWNED);
11064
11065         /*
11066          * Run back along the OOA queue, starting with the current
11067          * blocked I/O and going through every I/O before it on the
11068          * queue.  If starting_io is NULL, we'll just end up returning
11069          * CTL_ACTION_PASS.
11070          */
11071         for (ooa_io = *starting_io; ooa_io != NULL;
11072              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11073              ooa_links)){
11074                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11075                 if (action != CTL_ACTION_PASS) {
11076                         *starting_io = ooa_io;
11077                         return (action);
11078                 }
11079         }
11080
11081         *starting_io = NULL;
11082         return (CTL_ACTION_PASS);
11083 }
11084
11085 /*
11086  * Try to unblock the specified I/O.
11087  *
11088  * skip parameter allows explicitly skip present blocker of the I/O,
11089  * starting from the previous one on OOA queue.  It can be used when
11090  * we know for sure that the blocker I/O does no longer count.
11091  */
11092 static void
11093 ctl_try_unblock_io(struct ctl_lun *lun, union ctl_io *io, bool skip)
11094 {
11095         struct ctl_softc *softc = lun->ctl_softc;
11096         union ctl_io *bio, *obio;
11097         const struct ctl_cmd_entry *entry;
11098         union ctl_ha_msg msg_info;
11099         ctl_action action;
11100
11101         mtx_assert(&lun->lun_lock, MA_OWNED);
11102
11103         if (io->io_hdr.blocker == NULL)
11104                 return;
11105
11106         obio = bio = io->io_hdr.blocker;
11107         if (skip)
11108                 bio = (union ctl_io *)TAILQ_PREV(&bio->io_hdr, ctl_ooaq,
11109                     ooa_links);
11110         action = ctl_check_ooa(lun, io, &bio);
11111         if (action == CTL_ACTION_BLOCK) {
11112                 /* Still blocked, but may be by different I/O now. */
11113                 if (bio != obio) {
11114                         TAILQ_REMOVE(&obio->io_hdr.blocked_queue,
11115                             &io->io_hdr, blocked_links);
11116                         TAILQ_INSERT_TAIL(&bio->io_hdr.blocked_queue,
11117                             &io->io_hdr, blocked_links);
11118                         io->io_hdr.blocker = bio;
11119                 }
11120                 return;
11121         }
11122
11123         /* No longer blocked, one way or another. */
11124         TAILQ_REMOVE(&obio->io_hdr.blocked_queue, &io->io_hdr, blocked_links);
11125         io->io_hdr.blocker = NULL;
11126
11127         switch (action) {
11128         case CTL_ACTION_OVERLAP:
11129                 ctl_set_overlapped_cmd(&io->scsiio);
11130                 goto error;
11131         case CTL_ACTION_OVERLAP_TAG:
11132                 ctl_set_overlapped_tag(&io->scsiio,
11133                     io->scsiio.tag_num & 0xff);
11134                 goto error;
11135         case CTL_ACTION_PASS:
11136         case CTL_ACTION_SKIP:
11137
11138                 /* Serializing commands from the other SC retire there. */
11139                 if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) &&
11140                     (softc->ha_mode != CTL_HA_MODE_XFER)) {
11141                         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11142                         msg_info.hdr.original_sc = io->io_hdr.remote_io;
11143                         msg_info.hdr.serializing_sc = io;
11144                         msg_info.hdr.msg_type = CTL_MSG_R2R;
11145                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11146                             sizeof(msg_info.hdr), M_NOWAIT);
11147                         break;
11148                 }
11149
11150                 /*
11151                  * Check this I/O for LUN state changes that may have happened
11152                  * while this command was blocked. The LUN state may have been
11153                  * changed by a command ahead of us in the queue.
11154                  */
11155                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
11156                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
11157                         ctl_done(io);
11158                         break;
11159                 }
11160
11161                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11162                 ctl_enqueue_rtr(io);
11163                 break;
11164         case CTL_ACTION_ERROR:
11165         default:
11166                 ctl_set_internal_failure(&io->scsiio,
11167                                          /*sks_valid*/ 0,
11168                                          /*retry_count*/ 0);
11169
11170 error:
11171                 /* Serializing commands from the other SC are done here. */
11172                 if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) &&
11173                     (softc->ha_mode != CTL_HA_MODE_XFER)) {
11174                         ctl_try_unblock_others(lun, io, TRUE);
11175                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
11176
11177                         ctl_copy_sense_data_back(io, &msg_info);
11178                         msg_info.hdr.original_sc = io->io_hdr.remote_io;
11179                         msg_info.hdr.serializing_sc = NULL;
11180                         msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
11181                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11182                             sizeof(msg_info.scsi), M_WAITOK);
11183                         ctl_free_io(io);
11184                         break;
11185                 }
11186
11187                 ctl_done(io);
11188                 break;
11189         }
11190 }
11191
11192 /*
11193  * Try to unblock I/Os blocked by the specified I/O.
11194  *
11195  * skip parameter allows explicitly skip the specified I/O as blocker,
11196  * starting from the previous one on the OOA queue.  It can be used when
11197  * we know for sure that the specified I/O does no longer count (done).
11198  * It has to be still on OOA queue though so that we know where to start.
11199  */
11200 static void
11201 ctl_try_unblock_others(struct ctl_lun *lun, union ctl_io *bio, bool skip)
11202 {
11203         union ctl_io *io, *next_io;
11204
11205         mtx_assert(&lun->lun_lock, MA_OWNED);
11206
11207         for (io = (union ctl_io *)TAILQ_FIRST(&bio->io_hdr.blocked_queue);
11208              io != NULL; io = next_io) {
11209                 next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, blocked_links);
11210
11211                 KASSERT(io->io_hdr.blocker != NULL,
11212                     ("I/O %p on blocked list without blocker", io));
11213                 ctl_try_unblock_io(lun, io, skip);
11214         }
11215         KASSERT(!skip || TAILQ_EMPTY(&bio->io_hdr.blocked_queue),
11216             ("blocked_queue is not empty after skipping %p", bio));
11217 }
11218
11219 /*
11220  * This routine (with one exception) checks LUN flags that can be set by
11221  * commands ahead of us in the OOA queue.  These flags have to be checked
11222  * when a command initially comes in, and when we pull a command off the
11223  * blocked queue and are preparing to execute it.  The reason we have to
11224  * check these flags for commands on the blocked queue is that the LUN
11225  * state may have been changed by a command ahead of us while we're on the
11226  * blocked queue.
11227  *
11228  * Ordering is somewhat important with these checks, so please pay
11229  * careful attention to the placement of any new checks.
11230  */
11231 static int
11232 ctl_scsiio_lun_check(struct ctl_lun *lun,
11233     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11234 {
11235         struct ctl_softc *softc = lun->ctl_softc;
11236         int retval;
11237         uint32_t residx;
11238
11239         retval = 0;
11240
11241         mtx_assert(&lun->lun_lock, MA_OWNED);
11242
11243         /*
11244          * If this shelf is a secondary shelf controller, we may have to
11245          * reject some commands disallowed by HA mode and link state.
11246          */
11247         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11248                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11249                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11250                         ctl_set_lun_unavail(ctsio);
11251                         retval = 1;
11252                         goto bailout;
11253                 }
11254                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11255                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11256                         ctl_set_lun_transit(ctsio);
11257                         retval = 1;
11258                         goto bailout;
11259                 }
11260                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11261                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11262                         ctl_set_lun_standby(ctsio);
11263                         retval = 1;
11264                         goto bailout;
11265                 }
11266
11267                 /* The rest of checks are only done on executing side */
11268                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11269                         goto bailout;
11270         }
11271
11272         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11273                 if (lun->be_lun &&
11274                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11275                         ctl_set_hw_write_protected(ctsio);
11276                         retval = 1;
11277                         goto bailout;
11278                 }
11279                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11280                         ctl_set_sense(ctsio, /*current_error*/ 1,
11281                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11282                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11283                         retval = 1;
11284                         goto bailout;
11285                 }
11286         }
11287
11288         /*
11289          * Check for a reservation conflict.  If this command isn't allowed
11290          * even on reserved LUNs, and if this initiator isn't the one who
11291          * reserved us, reject the command with a reservation conflict.
11292          */
11293         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11294         if ((lun->flags & CTL_LUN_RESERVED)
11295          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11296                 if (lun->res_idx != residx) {
11297                         ctl_set_reservation_conflict(ctsio);
11298                         retval = 1;
11299                         goto bailout;
11300                 }
11301         }
11302
11303         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11304             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11305                 /* No reservation or command is allowed. */;
11306         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11307             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11308              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11309              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11310                 /* The command is allowed for Write Exclusive resv. */;
11311         } else {
11312                 /*
11313                  * if we aren't registered or it's a res holder type
11314                  * reservation and this isn't the res holder then set a
11315                  * conflict.
11316                  */
11317                 if (ctl_get_prkey(lun, residx) == 0 ||
11318                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11319                         ctl_set_reservation_conflict(ctsio);
11320                         retval = 1;
11321                         goto bailout;
11322                 }
11323         }
11324
11325         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11326                 if (lun->flags & CTL_LUN_EJECTED)
11327                         ctl_set_lun_ejected(ctsio);
11328                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11329                         if (lun->flags & CTL_LUN_REMOVABLE)
11330                                 ctl_set_lun_no_media(ctsio);
11331                         else
11332                                 ctl_set_lun_int_reqd(ctsio);
11333                 } else if (lun->flags & CTL_LUN_STOPPED)
11334                         ctl_set_lun_stopped(ctsio);
11335                 else
11336                         goto bailout;
11337                 retval = 1;
11338                 goto bailout;
11339         }
11340
11341 bailout:
11342         return (retval);
11343 }
11344
11345 static void
11346 ctl_failover_io(union ctl_io *io, int have_lock)
11347 {
11348         ctl_set_busy(&io->scsiio);
11349         ctl_done(io);
11350 }
11351
11352 static void
11353 ctl_failover_lun(union ctl_io *rio)
11354 {
11355         struct ctl_softc *softc = CTL_SOFTC(rio);
11356         struct ctl_lun *lun;
11357         struct ctl_io_hdr *io, *next_io;
11358         uint32_t targ_lun;
11359
11360         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11361         CTL_DEBUG_PRINT(("FAILOVER for lun %u\n", targ_lun));
11362
11363         /* Find and lock the LUN. */
11364         mtx_lock(&softc->ctl_lock);
11365         if (targ_lun > ctl_max_luns ||
11366             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11367                 mtx_unlock(&softc->ctl_lock);
11368                 return;
11369         }
11370         mtx_lock(&lun->lun_lock);
11371         mtx_unlock(&softc->ctl_lock);
11372         if (lun->flags & CTL_LUN_DISABLED) {
11373                 mtx_unlock(&lun->lun_lock);
11374                 return;
11375         }
11376
11377         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11378                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11379                         /* We are master */
11380                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11381                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11382                                         io->flags |= CTL_FLAG_ABORT;
11383                                         io->flags |= CTL_FLAG_FAILOVER;
11384                                         ctl_try_unblock_io(lun,
11385                                             (union ctl_io *)io, FALSE);
11386                                 } else { /* This can be only due to DATAMOVE */
11387                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11388                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11389                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11390                                         io->port_status = 31340;
11391                                         ctl_enqueue_isc((union ctl_io *)io);
11392                                 }
11393                         } else
11394                         /* We are slave */
11395                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11396                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11397                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11398                                         io->flags |= CTL_FLAG_FAILOVER;
11399                                 } else {
11400                                         ctl_set_busy(&((union ctl_io *)io)->
11401                                             scsiio);
11402                                         ctl_done((union ctl_io *)io);
11403                                 }
11404                         }
11405                 }
11406         } else { /* SERIALIZE modes */
11407                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11408                         /* We are master */
11409                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11410                                 if (io->blocker != NULL) {
11411                                         TAILQ_REMOVE(&io->blocker->io_hdr.blocked_queue,
11412                                             io, blocked_links);
11413                                         io->blocker = NULL;
11414                                 }
11415                                 ctl_try_unblock_others(lun, (union ctl_io *)io,
11416                                     TRUE);
11417                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11418                                 ctl_free_io((union ctl_io *)io);
11419                         } else
11420                         /* We are slave */
11421                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11422                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11423                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11424                                         ctl_set_busy(&((union ctl_io *)io)->
11425                                             scsiio);
11426                                         ctl_done((union ctl_io *)io);
11427                                 }
11428                         }
11429                 }
11430         }
11431         mtx_unlock(&lun->lun_lock);
11432 }
11433
11434 static int
11435 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11436 {
11437         struct ctl_lun *lun;
11438         const struct ctl_cmd_entry *entry;
11439         union ctl_io *bio;
11440         uint32_t initidx, targ_lun;
11441         int retval = 0;
11442
11443         lun = NULL;
11444         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11445         if (targ_lun < ctl_max_luns)
11446                 lun = softc->ctl_luns[targ_lun];
11447         if (lun) {
11448                 /*
11449                  * If the LUN is invalid, pretend that it doesn't exist.
11450                  * It will go away as soon as all pending I/O has been
11451                  * completed.
11452                  */
11453                 mtx_lock(&lun->lun_lock);
11454                 if (lun->flags & CTL_LUN_DISABLED) {
11455                         mtx_unlock(&lun->lun_lock);
11456                         lun = NULL;
11457                 }
11458         }
11459         CTL_LUN(ctsio) = lun;
11460         if (lun) {
11461                 CTL_BACKEND_LUN(ctsio) = lun->be_lun;
11462
11463                 /*
11464                  * Every I/O goes into the OOA queue for a particular LUN,
11465                  * and stays there until completion.
11466                  */
11467 #ifdef CTL_TIME_IO
11468                 if (TAILQ_EMPTY(&lun->ooa_queue))
11469                         lun->idle_time += getsbinuptime() - lun->last_busy;
11470 #endif
11471                 TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11472         }
11473
11474         /* Get command entry and return error if it is unsuppotyed. */
11475         entry = ctl_validate_command(ctsio);
11476         if (entry == NULL) {
11477                 if (lun)
11478                         mtx_unlock(&lun->lun_lock);
11479                 return (retval);
11480         }
11481
11482         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11483         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11484
11485         /*
11486          * Check to see whether we can send this command to LUNs that don't
11487          * exist.  This should pretty much only be the case for inquiry
11488          * and request sense.  Further checks, below, really require having
11489          * a LUN, so we can't really check the command anymore.  Just put
11490          * it on the rtr queue.
11491          */
11492         if (lun == NULL) {
11493                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11494                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11495                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11496                         return (retval);
11497                 }
11498
11499                 ctl_set_unsupported_lun(ctsio);
11500                 ctl_done((union ctl_io *)ctsio);
11501                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11502                 return (retval);
11503         } else {
11504                 /*
11505                  * Make sure we support this particular command on this LUN.
11506                  * e.g., we don't support writes to the control LUN.
11507                  */
11508                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11509                         mtx_unlock(&lun->lun_lock);
11510                         ctl_set_invalid_opcode(ctsio);
11511                         ctl_done((union ctl_io *)ctsio);
11512                         return (retval);
11513                 }
11514         }
11515
11516         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11517
11518         /*
11519          * If we've got a request sense, it'll clear the contingent
11520          * allegiance condition.  Otherwise, if we have a CA condition for
11521          * this initiator, clear it, because it sent down a command other
11522          * than request sense.
11523          */
11524         if (ctsio->cdb[0] != REQUEST_SENSE) {
11525                 struct scsi_sense_data *ps;
11526
11527                 ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
11528                 if (ps != NULL)
11529                         ps[initidx % CTL_MAX_INIT_PER_PORT].error_code = 0;
11530         }
11531
11532         /*
11533          * If the command has this flag set, it handles its own unit
11534          * attention reporting, we shouldn't do anything.  Otherwise we
11535          * check for any pending unit attentions, and send them back to the
11536          * initiator.  We only do this when a command initially comes in,
11537          * not when we pull it off the blocked queue.
11538          *
11539          * According to SAM-3, section 5.3.2, the order that things get
11540          * presented back to the host is basically unit attentions caused
11541          * by some sort of reset event, busy status, reservation conflicts
11542          * or task set full, and finally any other status.
11543          *
11544          * One issue here is that some of the unit attentions we report
11545          * don't fall into the "reset" category (e.g. "reported luns data
11546          * has changed").  So reporting it here, before the reservation
11547          * check, may be technically wrong.  I guess the only thing to do
11548          * would be to check for and report the reset events here, and then
11549          * check for the other unit attention types after we check for a
11550          * reservation conflict.
11551          *
11552          * XXX KDM need to fix this
11553          */
11554         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11555                 ctl_ua_type ua_type;
11556                 u_int sense_len = 0;
11557
11558                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11559                     &sense_len, SSD_TYPE_NONE);
11560                 if (ua_type != CTL_UA_NONE) {
11561                         mtx_unlock(&lun->lun_lock);
11562                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11563                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11564                         ctsio->sense_len = sense_len;
11565                         ctl_done((union ctl_io *)ctsio);
11566                         return (retval);
11567                 }
11568         }
11569
11570
11571         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11572                 mtx_unlock(&lun->lun_lock);
11573                 ctl_done((union ctl_io *)ctsio);
11574                 return (retval);
11575         }
11576
11577         /*
11578          * XXX CHD this is where we want to send IO to other side if
11579          * this LUN is secondary on this SC. We will need to make a copy
11580          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11581          * the copy we send as FROM_OTHER.
11582          * We also need to stuff the address of the original IO so we can
11583          * find it easily. Something similar will need be done on the other
11584          * side so when we are done we can find the copy.
11585          */
11586         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11587             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11588             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11589                 union ctl_ha_msg msg_info;
11590                 int isc_retval;
11591
11592                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11593                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11594                 mtx_unlock(&lun->lun_lock);
11595
11596                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11597                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11598                 msg_info.hdr.serializing_sc = NULL;
11599                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11600                 msg_info.scsi.tag_num = ctsio->tag_num;
11601                 msg_info.scsi.tag_type = ctsio->tag_type;
11602                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11603                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11604
11605                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11606                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11607                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11608                         ctl_set_busy(ctsio);
11609                         ctl_done((union ctl_io *)ctsio);
11610                         return (retval);
11611                 }
11612                 return (retval);
11613         }
11614
11615         bio = (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq, ooa_links);
11616         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio, &bio)) {
11617         case CTL_ACTION_BLOCK:
11618                 ctsio->io_hdr.blocker = bio;
11619                 TAILQ_INSERT_TAIL(&bio->io_hdr.blocked_queue, &ctsio->io_hdr,
11620                                   blocked_links);
11621                 mtx_unlock(&lun->lun_lock);
11622                 return (retval);
11623         case CTL_ACTION_PASS:
11624         case CTL_ACTION_SKIP:
11625                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11626                 mtx_unlock(&lun->lun_lock);
11627                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11628                 break;
11629         case CTL_ACTION_OVERLAP:
11630                 mtx_unlock(&lun->lun_lock);
11631                 ctl_set_overlapped_cmd(ctsio);
11632                 ctl_done((union ctl_io *)ctsio);
11633                 break;
11634         case CTL_ACTION_OVERLAP_TAG:
11635                 mtx_unlock(&lun->lun_lock);
11636                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11637                 ctl_done((union ctl_io *)ctsio);
11638                 break;
11639         case CTL_ACTION_ERROR:
11640         default:
11641                 mtx_unlock(&lun->lun_lock);
11642                 ctl_set_internal_failure(ctsio,
11643                                          /*sks_valid*/ 0,
11644                                          /*retry_count*/ 0);
11645                 ctl_done((union ctl_io *)ctsio);
11646                 break;
11647         }
11648         return (retval);
11649 }
11650
11651 const struct ctl_cmd_entry *
11652 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11653 {
11654         const struct ctl_cmd_entry *entry;
11655         int service_action;
11656
11657         entry = &ctl_cmd_table[ctsio->cdb[0]];
11658         if (sa)
11659                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11660         if (entry->flags & CTL_CMD_FLAG_SA5) {
11661                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11662                 entry = &((const struct ctl_cmd_entry *)
11663                     entry->execute)[service_action];
11664         }
11665         return (entry);
11666 }
11667
11668 const struct ctl_cmd_entry *
11669 ctl_validate_command(struct ctl_scsiio *ctsio)
11670 {
11671         const struct ctl_cmd_entry *entry;
11672         int i, sa;
11673         uint8_t diff;
11674
11675         entry = ctl_get_cmd_entry(ctsio, &sa);
11676         if (entry->execute == NULL) {
11677                 if (sa)
11678                         ctl_set_invalid_field(ctsio,
11679                                               /*sks_valid*/ 1,
11680                                               /*command*/ 1,
11681                                               /*field*/ 1,
11682                                               /*bit_valid*/ 1,
11683                                               /*bit*/ 4);
11684                 else
11685                         ctl_set_invalid_opcode(ctsio);
11686                 ctl_done((union ctl_io *)ctsio);
11687                 return (NULL);
11688         }
11689         KASSERT(entry->length > 0,
11690             ("Not defined length for command 0x%02x/0x%02x",
11691              ctsio->cdb[0], ctsio->cdb[1]));
11692         for (i = 1; i < entry->length; i++) {
11693                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11694                 if (diff == 0)
11695                         continue;
11696                 ctl_set_invalid_field(ctsio,
11697                                       /*sks_valid*/ 1,
11698                                       /*command*/ 1,
11699                                       /*field*/ i,
11700                                       /*bit_valid*/ 1,
11701                                       /*bit*/ fls(diff) - 1);
11702                 ctl_done((union ctl_io *)ctsio);
11703                 return (NULL);
11704         }
11705         return (entry);
11706 }
11707
11708 static int
11709 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11710 {
11711
11712         switch (lun_type) {
11713         case T_DIRECT:
11714                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11715                         return (0);
11716                 break;
11717         case T_PROCESSOR:
11718                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11719                         return (0);
11720                 break;
11721         case T_CDROM:
11722                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11723                         return (0);
11724                 break;
11725         default:
11726                 return (0);
11727         }
11728         return (1);
11729 }
11730
11731 static int
11732 ctl_scsiio(struct ctl_scsiio *ctsio)
11733 {
11734         int retval;
11735         const struct ctl_cmd_entry *entry;
11736
11737         retval = CTL_RETVAL_COMPLETE;
11738
11739         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11740
11741         entry = ctl_get_cmd_entry(ctsio, NULL);
11742
11743         /*
11744          * If this I/O has been aborted, just send it straight to
11745          * ctl_done() without executing it.
11746          */
11747         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11748                 ctl_done((union ctl_io *)ctsio);
11749                 goto bailout;
11750         }
11751
11752         /*
11753          * All the checks should have been handled by ctl_scsiio_precheck().
11754          * We should be clear now to just execute the I/O.
11755          */
11756         retval = entry->execute(ctsio);
11757
11758 bailout:
11759         return (retval);
11760 }
11761
11762 static int
11763 ctl_target_reset(union ctl_io *io)
11764 {
11765         struct ctl_softc *softc = CTL_SOFTC(io);
11766         struct ctl_port *port = CTL_PORT(io);
11767         struct ctl_lun *lun;
11768         uint32_t initidx;
11769         ctl_ua_type ua_type;
11770
11771         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11772                 union ctl_ha_msg msg_info;
11773
11774                 msg_info.hdr.nexus = io->io_hdr.nexus;
11775                 msg_info.task.task_action = io->taskio.task_action;
11776                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11777                 msg_info.hdr.original_sc = NULL;
11778                 msg_info.hdr.serializing_sc = NULL;
11779                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11780                     sizeof(msg_info.task), M_WAITOK);
11781         }
11782
11783         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11784         if (io->taskio.task_action == CTL_TASK_TARGET_RESET)
11785                 ua_type = CTL_UA_TARG_RESET;
11786         else
11787                 ua_type = CTL_UA_BUS_RESET;
11788         mtx_lock(&softc->ctl_lock);
11789         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11790                 if (port != NULL &&
11791                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
11792                         continue;
11793                 ctl_do_lun_reset(lun, initidx, ua_type);
11794         }
11795         mtx_unlock(&softc->ctl_lock);
11796         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11797         return (0);
11798 }
11799
11800 /*
11801  * The LUN should always be set.  The I/O is optional, and is used to
11802  * distinguish between I/Os sent by this initiator, and by other
11803  * initiators.  We set unit attention for initiators other than this one.
11804  * SAM-3 is vague on this point.  It does say that a unit attention should
11805  * be established for other initiators when a LUN is reset (see section
11806  * 5.7.3), but it doesn't specifically say that the unit attention should
11807  * be established for this particular initiator when a LUN is reset.  Here
11808  * is the relevant text, from SAM-3 rev 8:
11809  *
11810  * 5.7.2 When a SCSI initiator port aborts its own tasks
11811  *
11812  * When a SCSI initiator port causes its own task(s) to be aborted, no
11813  * notification that the task(s) have been aborted shall be returned to
11814  * the SCSI initiator port other than the completion response for the
11815  * command or task management function action that caused the task(s) to
11816  * be aborted and notification(s) associated with related effects of the
11817  * action (e.g., a reset unit attention condition).
11818  *
11819  * XXX KDM for now, we're setting unit attention for all initiators.
11820  */
11821 static void
11822 ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua_type)
11823 {
11824         union ctl_io *xio;
11825         int i;
11826
11827         mtx_lock(&lun->lun_lock);
11828         /* Abort tasks. */
11829         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11830              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11831                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11832                 ctl_try_unblock_io(lun, xio, FALSE);
11833         }
11834         /* Clear CA. */
11835         for (i = 0; i < ctl_max_ports; i++) {
11836                 free(lun->pending_sense[i], M_CTL);
11837                 lun->pending_sense[i] = NULL;
11838         }
11839         /* Clear reservation. */
11840         lun->flags &= ~CTL_LUN_RESERVED;
11841         /* Clear prevent media removal. */
11842         if (lun->prevent) {
11843                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
11844                         ctl_clear_mask(lun->prevent, i);
11845                 lun->prevent_count = 0;
11846         }
11847         /* Clear TPC status */
11848         ctl_tpc_lun_clear(lun, -1);
11849         /* Establish UA. */
11850 #if 0
11851         ctl_est_ua_all(lun, initidx, ua_type);
11852 #else
11853         ctl_est_ua_all(lun, -1, ua_type);
11854 #endif
11855         mtx_unlock(&lun->lun_lock);
11856 }
11857
11858 static int
11859 ctl_lun_reset(union ctl_io *io)
11860 {
11861         struct ctl_softc *softc = CTL_SOFTC(io);
11862         struct ctl_lun *lun;
11863         uint32_t targ_lun, initidx;
11864
11865         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11866         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11867         mtx_lock(&softc->ctl_lock);
11868         if (targ_lun >= ctl_max_luns ||
11869             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11870                 mtx_unlock(&softc->ctl_lock);
11871                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11872                 return (1);
11873         }
11874         ctl_do_lun_reset(lun, initidx, CTL_UA_LUN_RESET);
11875         mtx_unlock(&softc->ctl_lock);
11876         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11877
11878         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11879                 union ctl_ha_msg msg_info;
11880
11881                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11882                 msg_info.hdr.nexus = io->io_hdr.nexus;
11883                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
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         return (0);
11890 }
11891
11892 static void
11893 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11894     int other_sc)
11895 {
11896         union ctl_io *xio;
11897
11898         mtx_assert(&lun->lun_lock, MA_OWNED);
11899
11900         /*
11901          * Run through the OOA queue and attempt to find the given I/O.
11902          * The target port, initiator ID, tag type and tag number have to
11903          * match the values that we got from the initiator.  If we have an
11904          * untagged command to abort, simply abort the first untagged command
11905          * we come to.  We only allow one untagged command at a time of course.
11906          */
11907         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11908              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11909
11910                 if ((targ_port == UINT32_MAX ||
11911                      targ_port == xio->io_hdr.nexus.targ_port) &&
11912                     (init_id == UINT32_MAX ||
11913                      init_id == xio->io_hdr.nexus.initid)) {
11914                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11915                             init_id != xio->io_hdr.nexus.initid)
11916                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11917                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11918                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11919                                 union ctl_ha_msg msg_info;
11920
11921                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11922                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11923                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11924                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11925                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11926                                 msg_info.hdr.original_sc = NULL;
11927                                 msg_info.hdr.serializing_sc = NULL;
11928                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11929                                     sizeof(msg_info.task), M_NOWAIT);
11930                         }
11931                         ctl_try_unblock_io(lun, xio, FALSE);
11932                 }
11933         }
11934 }
11935
11936 static int
11937 ctl_abort_task_set(union ctl_io *io)
11938 {
11939         struct ctl_softc *softc = CTL_SOFTC(io);
11940         struct ctl_lun *lun;
11941         uint32_t targ_lun;
11942
11943         /*
11944          * Look up the LUN.
11945          */
11946         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11947         mtx_lock(&softc->ctl_lock);
11948         if (targ_lun >= ctl_max_luns ||
11949             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11950                 mtx_unlock(&softc->ctl_lock);
11951                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11952                 return (1);
11953         }
11954
11955         mtx_lock(&lun->lun_lock);
11956         mtx_unlock(&softc->ctl_lock);
11957         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11958                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11959                     io->io_hdr.nexus.initid,
11960                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11961         } else { /* CTL_TASK_CLEAR_TASK_SET */
11962                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11963                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11964         }
11965         mtx_unlock(&lun->lun_lock);
11966         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11967         return (0);
11968 }
11969
11970 static void
11971 ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
11972     ctl_ua_type ua_type)
11973 {
11974         struct ctl_lun *lun;
11975         struct scsi_sense_data *ps;
11976         uint32_t p, i;
11977
11978         p = initidx / CTL_MAX_INIT_PER_PORT;
11979         i = initidx % CTL_MAX_INIT_PER_PORT;
11980         mtx_lock(&softc->ctl_lock);
11981         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11982                 mtx_lock(&lun->lun_lock);
11983                 /* Abort tasks. */
11984                 ctl_abort_tasks_lun(lun, p, i, 1);
11985                 /* Clear CA. */
11986                 ps = lun->pending_sense[p];
11987                 if (ps != NULL)
11988                         ps[i].error_code = 0;
11989                 /* Clear reservation. */
11990                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11991                         lun->flags &= ~CTL_LUN_RESERVED;
11992                 /* Clear prevent media removal. */
11993                 if (lun->prevent && ctl_is_set(lun->prevent, initidx)) {
11994                         ctl_clear_mask(lun->prevent, initidx);
11995                         lun->prevent_count--;
11996                 }
11997                 /* Clear TPC status */
11998                 ctl_tpc_lun_clear(lun, initidx);
11999                 /* Establish UA. */
12000                 ctl_est_ua(lun, initidx, ua_type);
12001                 mtx_unlock(&lun->lun_lock);
12002         }
12003         mtx_unlock(&softc->ctl_lock);
12004 }
12005
12006 static int
12007 ctl_i_t_nexus_reset(union ctl_io *io)
12008 {
12009         struct ctl_softc *softc = CTL_SOFTC(io);
12010         uint32_t initidx;
12011
12012         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12013                 union ctl_ha_msg msg_info;
12014
12015                 msg_info.hdr.nexus = io->io_hdr.nexus;
12016                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
12017                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12018                 msg_info.hdr.original_sc = NULL;
12019                 msg_info.hdr.serializing_sc = NULL;
12020                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12021                     sizeof(msg_info.task), M_WAITOK);
12022         }
12023
12024         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12025         ctl_i_t_nexus_loss(softc, initidx, CTL_UA_I_T_NEXUS_LOSS);
12026         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12027         return (0);
12028 }
12029
12030 static int
12031 ctl_abort_task(union ctl_io *io)
12032 {
12033         struct ctl_softc *softc = CTL_SOFTC(io);
12034         union ctl_io *xio;
12035         struct ctl_lun *lun;
12036         uint32_t targ_lun;
12037
12038         /*
12039          * Look up the LUN.
12040          */
12041         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12042         mtx_lock(&softc->ctl_lock);
12043         if (targ_lun >= ctl_max_luns ||
12044             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12045                 mtx_unlock(&softc->ctl_lock);
12046                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12047                 return (1);
12048         }
12049
12050         mtx_lock(&lun->lun_lock);
12051         mtx_unlock(&softc->ctl_lock);
12052         /*
12053          * Run through the OOA queue and attempt to find the given I/O.
12054          * The target port, initiator ID, tag type and tag number have to
12055          * match the values that we got from the initiator.  If we have an
12056          * untagged command to abort, simply abort the first untagged command
12057          * we come to.  We only allow one untagged command at a time of course.
12058          */
12059         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12060              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12061
12062                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12063                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12064                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12065                         continue;
12066
12067                 /*
12068                  * If the abort says that the task is untagged, the
12069                  * task in the queue must be untagged.  Otherwise,
12070                  * we just check to see whether the tag numbers
12071                  * match.  This is because the QLogic firmware
12072                  * doesn't pass back the tag type in an abort
12073                  * request.
12074                  */
12075 #if 0
12076                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12077                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12078                  || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12079 #else
12080                 /*
12081                  * XXX KDM we've got problems with FC, because it
12082                  * doesn't send down a tag type with aborts.  So we
12083                  * can only really go by the tag number...
12084                  * This may cause problems with parallel SCSI.
12085                  * Need to figure that out!!
12086                  */
12087                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
12088 #endif
12089                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12090                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12091                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12092                                 union ctl_ha_msg msg_info;
12093
12094                                 msg_info.hdr.nexus = io->io_hdr.nexus;
12095                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12096                                 msg_info.task.tag_num = io->taskio.tag_num;
12097                                 msg_info.task.tag_type = io->taskio.tag_type;
12098                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12099                                 msg_info.hdr.original_sc = NULL;
12100                                 msg_info.hdr.serializing_sc = NULL;
12101                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12102                                     sizeof(msg_info.task), M_NOWAIT);
12103                         }
12104                         ctl_try_unblock_io(lun, xio, FALSE);
12105                 }
12106         }
12107         mtx_unlock(&lun->lun_lock);
12108         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12109         return (0);
12110 }
12111
12112 static int
12113 ctl_query_task(union ctl_io *io, int task_set)
12114 {
12115         struct ctl_softc *softc = CTL_SOFTC(io);
12116         union ctl_io *xio;
12117         struct ctl_lun *lun;
12118         int found = 0;
12119         uint32_t targ_lun;
12120
12121         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12122         mtx_lock(&softc->ctl_lock);
12123         if (targ_lun >= ctl_max_luns ||
12124             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12125                 mtx_unlock(&softc->ctl_lock);
12126                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12127                 return (1);
12128         }
12129         mtx_lock(&lun->lun_lock);
12130         mtx_unlock(&softc->ctl_lock);
12131         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12132              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12133
12134                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12135                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12136                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12137                         continue;
12138
12139                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12140                         found = 1;
12141                         break;
12142                 }
12143         }
12144         mtx_unlock(&lun->lun_lock);
12145         if (found)
12146                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12147         else
12148                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12149         return (0);
12150 }
12151
12152 static int
12153 ctl_query_async_event(union ctl_io *io)
12154 {
12155         struct ctl_softc *softc = CTL_SOFTC(io);
12156         struct ctl_lun *lun;
12157         ctl_ua_type ua;
12158         uint32_t targ_lun, initidx;
12159
12160         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12161         mtx_lock(&softc->ctl_lock);
12162         if (targ_lun >= ctl_max_luns ||
12163             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12164                 mtx_unlock(&softc->ctl_lock);
12165                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12166                 return (1);
12167         }
12168         mtx_lock(&lun->lun_lock);
12169         mtx_unlock(&softc->ctl_lock);
12170         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12171         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12172         mtx_unlock(&lun->lun_lock);
12173         if (ua != CTL_UA_NONE)
12174                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12175         else
12176                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12177         return (0);
12178 }
12179
12180 static void
12181 ctl_run_task(union ctl_io *io)
12182 {
12183         int retval = 1;
12184
12185         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12186         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12187             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12188         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12189         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12190         switch (io->taskio.task_action) {
12191         case CTL_TASK_ABORT_TASK:
12192                 retval = ctl_abort_task(io);
12193                 break;
12194         case CTL_TASK_ABORT_TASK_SET:
12195         case CTL_TASK_CLEAR_TASK_SET:
12196                 retval = ctl_abort_task_set(io);
12197                 break;
12198         case CTL_TASK_CLEAR_ACA:
12199                 break;
12200         case CTL_TASK_I_T_NEXUS_RESET:
12201                 retval = ctl_i_t_nexus_reset(io);
12202                 break;
12203         case CTL_TASK_LUN_RESET:
12204                 retval = ctl_lun_reset(io);
12205                 break;
12206         case CTL_TASK_TARGET_RESET:
12207         case CTL_TASK_BUS_RESET:
12208                 retval = ctl_target_reset(io);
12209                 break;
12210         case CTL_TASK_PORT_LOGIN:
12211                 break;
12212         case CTL_TASK_PORT_LOGOUT:
12213                 break;
12214         case CTL_TASK_QUERY_TASK:
12215                 retval = ctl_query_task(io, 0);
12216                 break;
12217         case CTL_TASK_QUERY_TASK_SET:
12218                 retval = ctl_query_task(io, 1);
12219                 break;
12220         case CTL_TASK_QUERY_ASYNC_EVENT:
12221                 retval = ctl_query_async_event(io);
12222                 break;
12223         default:
12224                 printf("%s: got unknown task management event %d\n",
12225                        __func__, io->taskio.task_action);
12226                 break;
12227         }
12228         if (retval == 0)
12229                 io->io_hdr.status = CTL_SUCCESS;
12230         else
12231                 io->io_hdr.status = CTL_ERROR;
12232         ctl_done(io);
12233 }
12234
12235 /*
12236  * For HA operation.  Handle commands that come in from the other
12237  * controller.
12238  */
12239 static void
12240 ctl_handle_isc(union ctl_io *io)
12241 {
12242         struct ctl_softc *softc = CTL_SOFTC(io);
12243         struct ctl_lun *lun;
12244         const struct ctl_cmd_entry *entry;
12245         uint32_t targ_lun;
12246
12247         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12248         switch (io->io_hdr.msg_type) {
12249         case CTL_MSG_SERIALIZE:
12250                 ctl_serialize_other_sc_cmd(&io->scsiio);
12251                 break;
12252         case CTL_MSG_R2R:               /* Only used in SER_ONLY mode. */
12253                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12254                 if (targ_lun >= ctl_max_luns ||
12255                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12256                         ctl_done(io);
12257                         break;
12258                 }
12259                 mtx_lock(&lun->lun_lock);
12260                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
12261                         mtx_unlock(&lun->lun_lock);
12262                         ctl_done(io);
12263                         break;
12264                 }
12265                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12266                 mtx_unlock(&lun->lun_lock);
12267                 ctl_enqueue_rtr(io);
12268                 break;
12269         case CTL_MSG_FINISH_IO:
12270                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12271                         ctl_done(io);
12272                         break;
12273                 }
12274                 if (targ_lun >= ctl_max_luns ||
12275                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12276                         ctl_free_io(io);
12277                         break;
12278                 }
12279                 mtx_lock(&lun->lun_lock);
12280                 ctl_try_unblock_others(lun, io, TRUE);
12281                 TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12282                 mtx_unlock(&lun->lun_lock);
12283                 ctl_free_io(io);
12284                 break;
12285         case CTL_MSG_PERS_ACTION:
12286                 ctl_hndl_per_res_out_on_other_sc(io);
12287                 ctl_free_io(io);
12288                 break;
12289         case CTL_MSG_BAD_JUJU:
12290                 ctl_done(io);
12291                 break;
12292         case CTL_MSG_DATAMOVE:          /* Only used in XFER mode */
12293                 ctl_datamove_remote(io);
12294                 break;
12295         case CTL_MSG_DATAMOVE_DONE:     /* Only used in XFER mode */
12296                 io->scsiio.be_move_done(io);
12297                 break;
12298         case CTL_MSG_FAILOVER:
12299                 ctl_failover_lun(io);
12300                 ctl_free_io(io);
12301                 break;
12302         default:
12303                 printf("%s: Invalid message type %d\n",
12304                        __func__, io->io_hdr.msg_type);
12305                 ctl_free_io(io);
12306                 break;
12307         }
12308
12309 }
12310
12311
12312 /*
12313  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12314  * there is no match.
12315  */
12316 static ctl_lun_error_pattern
12317 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12318 {
12319         const struct ctl_cmd_entry *entry;
12320         ctl_lun_error_pattern filtered_pattern, pattern;
12321
12322         pattern = desc->error_pattern;
12323
12324         /*
12325          * XXX KDM we need more data passed into this function to match a
12326          * custom pattern, and we actually need to implement custom pattern
12327          * matching.
12328          */
12329         if (pattern & CTL_LUN_PAT_CMD)
12330                 return (CTL_LUN_PAT_CMD);
12331
12332         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12333                 return (CTL_LUN_PAT_ANY);
12334
12335         entry = ctl_get_cmd_entry(ctsio, NULL);
12336
12337         filtered_pattern = entry->pattern & pattern;
12338
12339         /*
12340          * If the user requested specific flags in the pattern (e.g.
12341          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12342          * flags.
12343          *
12344          * If the user did not specify any flags, it doesn't matter whether
12345          * or not the command supports the flags.
12346          */
12347         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12348              (pattern & ~CTL_LUN_PAT_MASK))
12349                 return (CTL_LUN_PAT_NONE);
12350
12351         /*
12352          * If the user asked for a range check, see if the requested LBA
12353          * range overlaps with this command's LBA range.
12354          */
12355         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12356                 uint64_t lba1;
12357                 uint64_t len1;
12358                 ctl_action action;
12359                 int retval;
12360
12361                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12362                 if (retval != 0)
12363                         return (CTL_LUN_PAT_NONE);
12364
12365                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12366                                               desc->lba_range.len, FALSE);
12367                 /*
12368                  * A "pass" means that the LBA ranges don't overlap, so
12369                  * this doesn't match the user's range criteria.
12370                  */
12371                 if (action == CTL_ACTION_PASS)
12372                         return (CTL_LUN_PAT_NONE);
12373         }
12374
12375         return (filtered_pattern);
12376 }
12377
12378 static void
12379 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12380 {
12381         struct ctl_error_desc *desc, *desc2;
12382
12383         mtx_assert(&lun->lun_lock, MA_OWNED);
12384
12385         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12386                 ctl_lun_error_pattern pattern;
12387                 /*
12388                  * Check to see whether this particular command matches
12389                  * the pattern in the descriptor.
12390                  */
12391                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12392                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12393                         continue;
12394
12395                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12396                 case CTL_LUN_INJ_ABORTED:
12397                         ctl_set_aborted(&io->scsiio);
12398                         break;
12399                 case CTL_LUN_INJ_MEDIUM_ERR:
12400                         ctl_set_medium_error(&io->scsiio,
12401                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12402                              CTL_FLAG_DATA_OUT);
12403                         break;
12404                 case CTL_LUN_INJ_UA:
12405                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12406                          * OCCURRED */
12407                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12408                         break;
12409                 case CTL_LUN_INJ_CUSTOM:
12410                         /*
12411                          * We're assuming the user knows what he is doing.
12412                          * Just copy the sense information without doing
12413                          * checks.
12414                          */
12415                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12416                               MIN(sizeof(desc->custom_sense),
12417                                   sizeof(io->scsiio.sense_data)));
12418                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12419                         io->scsiio.sense_len = SSD_FULL_SIZE;
12420                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12421                         break;
12422                 case CTL_LUN_INJ_NONE:
12423                 default:
12424                         /*
12425                          * If this is an error injection type we don't know
12426                          * about, clear the continuous flag (if it is set)
12427                          * so it will get deleted below.
12428                          */
12429                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12430                         break;
12431                 }
12432                 /*
12433                  * By default, each error injection action is a one-shot
12434                  */
12435                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12436                         continue;
12437
12438                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12439
12440                 free(desc, M_CTL);
12441         }
12442 }
12443
12444 #ifdef CTL_IO_DELAY
12445 static void
12446 ctl_datamove_timer_wakeup(void *arg)
12447 {
12448         union ctl_io *io;
12449
12450         io = (union ctl_io *)arg;
12451
12452         ctl_datamove(io);
12453 }
12454 #endif /* CTL_IO_DELAY */
12455
12456 void
12457 ctl_datamove(union ctl_io *io)
12458 {
12459         void (*fe_datamove)(union ctl_io *io);
12460
12461         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12462
12463         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12464
12465         /* No data transferred yet.  Frontend must update this when done. */
12466         io->scsiio.kern_data_resid = io->scsiio.kern_data_len;
12467
12468 #ifdef CTL_TIME_IO
12469         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12470                 char str[256];
12471                 char path_str[64];
12472                 struct sbuf sb;
12473
12474                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12475                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12476
12477                 sbuf_cat(&sb, path_str);
12478                 switch (io->io_hdr.io_type) {
12479                 case CTL_IO_SCSI:
12480                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12481                         sbuf_printf(&sb, "\n");
12482                         sbuf_cat(&sb, path_str);
12483                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12484                                     io->scsiio.tag_num, io->scsiio.tag_type);
12485                         break;
12486                 case CTL_IO_TASK:
12487                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12488                                     "Tag Type: %d\n", io->taskio.task_action,
12489                                     io->taskio.tag_num, io->taskio.tag_type);
12490                         break;
12491                 default:
12492                         panic("%s: Invalid CTL I/O type %d\n",
12493                             __func__, io->io_hdr.io_type);
12494                 }
12495                 sbuf_cat(&sb, path_str);
12496                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12497                             (intmax_t)time_uptime - io->io_hdr.start_time);
12498                 sbuf_finish(&sb);
12499                 printf("%s", sbuf_data(&sb));
12500         }
12501 #endif /* CTL_TIME_IO */
12502
12503 #ifdef CTL_IO_DELAY
12504         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12505                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12506         } else {
12507                 struct ctl_lun *lun;
12508
12509                 lun = CTL_LUN(io);
12510                 if ((lun != NULL)
12511                  && (lun->delay_info.datamove_delay > 0)) {
12512
12513                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12514                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12515                         callout_reset(&io->io_hdr.delay_callout,
12516                                       lun->delay_info.datamove_delay * hz,
12517                                       ctl_datamove_timer_wakeup, io);
12518                         if (lun->delay_info.datamove_type ==
12519                             CTL_DELAY_TYPE_ONESHOT)
12520                                 lun->delay_info.datamove_delay = 0;
12521                         return;
12522                 }
12523         }
12524 #endif
12525
12526         /*
12527          * This command has been aborted.  Set the port status, so we fail
12528          * the data move.
12529          */
12530         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12531                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12532                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12533                        io->io_hdr.nexus.targ_port,
12534                        io->io_hdr.nexus.targ_lun);
12535                 io->io_hdr.port_status = 31337;
12536                 /*
12537                  * Note that the backend, in this case, will get the
12538                  * callback in its context.  In other cases it may get
12539                  * called in the frontend's interrupt thread context.
12540                  */
12541                 io->scsiio.be_move_done(io);
12542                 return;
12543         }
12544
12545         /* Don't confuse frontend with zero length data move. */
12546         if (io->scsiio.kern_data_len == 0) {
12547                 io->scsiio.be_move_done(io);
12548                 return;
12549         }
12550
12551         fe_datamove = CTL_PORT(io)->fe_datamove;
12552         fe_datamove(io);
12553 }
12554
12555 static void
12556 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12557 {
12558         union ctl_ha_msg msg;
12559 #ifdef CTL_TIME_IO
12560         struct bintime cur_bt;
12561 #endif
12562
12563         memset(&msg, 0, sizeof(msg));
12564         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12565         msg.hdr.original_sc = io;
12566         msg.hdr.serializing_sc = io->io_hdr.remote_io;
12567         msg.hdr.nexus = io->io_hdr.nexus;
12568         msg.hdr.status = io->io_hdr.status;
12569         msg.scsi.kern_data_resid = io->scsiio.kern_data_resid;
12570         msg.scsi.tag_num = io->scsiio.tag_num;
12571         msg.scsi.tag_type = io->scsiio.tag_type;
12572         msg.scsi.scsi_status = io->scsiio.scsi_status;
12573         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12574                io->scsiio.sense_len);
12575         msg.scsi.sense_len = io->scsiio.sense_len;
12576         msg.scsi.port_status = io->io_hdr.port_status;
12577         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12578         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12579                 ctl_failover_io(io, /*have_lock*/ have_lock);
12580                 return;
12581         }
12582         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12583             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12584             msg.scsi.sense_len, M_WAITOK);
12585
12586 #ifdef CTL_TIME_IO
12587         getbinuptime(&cur_bt);
12588         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12589         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12590 #endif
12591         io->io_hdr.num_dmas++;
12592 }
12593
12594 /*
12595  * The DMA to the remote side is done, now we need to tell the other side
12596  * we're done so it can continue with its data movement.
12597  */
12598 static void
12599 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12600 {
12601         union ctl_io *io;
12602         uint32_t i;
12603
12604         io = rq->context;
12605
12606         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12607                 printf("%s: ISC DMA write failed with error %d", __func__,
12608                        rq->ret);
12609                 ctl_set_internal_failure(&io->scsiio,
12610                                          /*sks_valid*/ 1,
12611                                          /*retry_count*/ rq->ret);
12612         }
12613
12614         ctl_dt_req_free(rq);
12615
12616         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12617                 free(CTL_LSGLT(io)[i].addr, M_CTL);
12618         free(CTL_RSGL(io), M_CTL);
12619         CTL_RSGL(io) = NULL;
12620         CTL_LSGL(io) = NULL;
12621
12622         /*
12623          * The data is in local and remote memory, so now we need to send
12624          * status (good or back) back to the other side.
12625          */
12626         ctl_send_datamove_done(io, /*have_lock*/ 0);
12627 }
12628
12629 /*
12630  * We've moved the data from the host/controller into local memory.  Now we
12631  * need to push it over to the remote controller's memory.
12632  */
12633 static int
12634 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12635 {
12636         int retval;
12637
12638         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12639                                           ctl_datamove_remote_write_cb);
12640         return (retval);
12641 }
12642
12643 static void
12644 ctl_datamove_remote_write(union ctl_io *io)
12645 {
12646         int retval;
12647         void (*fe_datamove)(union ctl_io *io);
12648
12649         /*
12650          * - Get the data from the host/HBA into local memory.
12651          * - DMA memory from the local controller to the remote controller.
12652          * - Send status back to the remote controller.
12653          */
12654
12655         retval = ctl_datamove_remote_sgl_setup(io);
12656         if (retval != 0)
12657                 return;
12658
12659         /* Switch the pointer over so the FETD knows what to do */
12660         io->scsiio.kern_data_ptr = (uint8_t *)CTL_LSGL(io);
12661
12662         /*
12663          * Use a custom move done callback, since we need to send completion
12664          * back to the other controller, not to the backend on this side.
12665          */
12666         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12667
12668         fe_datamove = CTL_PORT(io)->fe_datamove;
12669         fe_datamove(io);
12670 }
12671
12672 static int
12673 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12674 {
12675         uint32_t i;
12676
12677         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12678                 free(CTL_LSGLT(io)[i].addr, M_CTL);
12679         free(CTL_RSGL(io), M_CTL);
12680         CTL_RSGL(io) = NULL;
12681         CTL_LSGL(io) = NULL;
12682
12683         /*
12684          * The read is done, now we need to send status (good or bad) back
12685          * to the other side.
12686          */
12687         ctl_send_datamove_done(io, /*have_lock*/ 0);
12688
12689         return (0);
12690 }
12691
12692 static void
12693 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12694 {
12695         union ctl_io *io;
12696         void (*fe_datamove)(union ctl_io *io);
12697
12698         io = rq->context;
12699
12700         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12701                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12702                        rq->ret);
12703                 ctl_set_internal_failure(&io->scsiio,
12704                                          /*sks_valid*/ 1,
12705                                          /*retry_count*/ rq->ret);
12706         }
12707
12708         ctl_dt_req_free(rq);
12709
12710         /* Switch the pointer over so the FETD knows what to do */
12711         io->scsiio.kern_data_ptr = (uint8_t *)CTL_LSGL(io);
12712
12713         /*
12714          * Use a custom move done callback, since we need to send completion
12715          * back to the other controller, not to the backend on this side.
12716          */
12717         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12718
12719         /* XXX KDM add checks like the ones in ctl_datamove? */
12720
12721         fe_datamove = CTL_PORT(io)->fe_datamove;
12722         fe_datamove(io);
12723 }
12724
12725 static int
12726 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12727 {
12728         struct ctl_sg_entry *local_sglist;
12729         uint32_t len_to_go;
12730         int retval;
12731         int i;
12732
12733         retval = 0;
12734         local_sglist = CTL_LSGL(io);
12735         len_to_go = io->scsiio.kern_data_len;
12736
12737         /*
12738          * The difficult thing here is that the size of the various
12739          * S/G segments may be different than the size from the
12740          * remote controller.  That'll make it harder when DMAing
12741          * the data back to the other side.
12742          */
12743         for (i = 0; len_to_go > 0; i++) {
12744                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12745                 local_sglist[i].addr =
12746                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12747
12748                 len_to_go -= local_sglist[i].len;
12749         }
12750         /*
12751          * Reset the number of S/G entries accordingly.  The original
12752          * number of S/G entries is available in rem_sg_entries.
12753          */
12754         io->scsiio.kern_sg_entries = i;
12755
12756         return (retval);
12757 }
12758
12759 static int
12760 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12761                          ctl_ha_dt_cb callback)
12762 {
12763         struct ctl_ha_dt_req *rq;
12764         struct ctl_sg_entry *remote_sglist, *local_sglist;
12765         uint32_t local_used, remote_used, total_used;
12766         int i, j, isc_ret;
12767
12768         rq = ctl_dt_req_alloc();
12769
12770         /*
12771          * If we failed to allocate the request, and if the DMA didn't fail
12772          * anyway, set busy status.  This is just a resource allocation
12773          * failure.
12774          */
12775         if ((rq == NULL)
12776          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12777              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12778                 ctl_set_busy(&io->scsiio);
12779
12780         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12781             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12782
12783                 if (rq != NULL)
12784                         ctl_dt_req_free(rq);
12785
12786                 /*
12787                  * The data move failed.  We need to return status back
12788                  * to the other controller.  No point in trying to DMA
12789                  * data to the remote controller.
12790                  */
12791
12792                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12793
12794                 return (1);
12795         }
12796
12797         local_sglist = CTL_LSGL(io);
12798         remote_sglist = CTL_RSGL(io);
12799         local_used = 0;
12800         remote_used = 0;
12801         total_used = 0;
12802
12803         /*
12804          * Pull/push the data over the wire from/to the other controller.
12805          * This takes into account the possibility that the local and
12806          * remote sglists may not be identical in terms of the size of
12807          * the elements and the number of elements.
12808          *
12809          * One fundamental assumption here is that the length allocated for
12810          * both the local and remote sglists is identical.  Otherwise, we've
12811          * essentially got a coding error of some sort.
12812          */
12813         isc_ret = CTL_HA_STATUS_SUCCESS;
12814         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12815                 uint32_t cur_len;
12816                 uint8_t *tmp_ptr;
12817
12818                 rq->command = command;
12819                 rq->context = io;
12820
12821                 /*
12822                  * Both pointers should be aligned.  But it is possible
12823                  * that the allocation length is not.  They should both
12824                  * also have enough slack left over at the end, though,
12825                  * to round up to the next 8 byte boundary.
12826                  */
12827                 cur_len = MIN(local_sglist[i].len - local_used,
12828                               remote_sglist[j].len - remote_used);
12829                 rq->size = cur_len;
12830
12831                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12832                 tmp_ptr += local_used;
12833
12834 #if 0
12835                 /* Use physical addresses when talking to ISC hardware */
12836                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12837                         /* XXX KDM use busdma */
12838                         rq->local = vtophys(tmp_ptr);
12839                 } else
12840                         rq->local = tmp_ptr;
12841 #else
12842                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12843                     ("HA does not support BUS_ADDR"));
12844                 rq->local = tmp_ptr;
12845 #endif
12846
12847                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12848                 tmp_ptr += remote_used;
12849                 rq->remote = tmp_ptr;
12850
12851                 rq->callback = NULL;
12852
12853                 local_used += cur_len;
12854                 if (local_used >= local_sglist[i].len) {
12855                         i++;
12856                         local_used = 0;
12857                 }
12858
12859                 remote_used += cur_len;
12860                 if (remote_used >= remote_sglist[j].len) {
12861                         j++;
12862                         remote_used = 0;
12863                 }
12864                 total_used += cur_len;
12865
12866                 if (total_used >= io->scsiio.kern_data_len)
12867                         rq->callback = callback;
12868
12869                 isc_ret = ctl_dt_single(rq);
12870                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
12871                         break;
12872         }
12873         if (isc_ret != CTL_HA_STATUS_WAIT) {
12874                 rq->ret = isc_ret;
12875                 callback(rq);
12876         }
12877
12878         return (0);
12879 }
12880
12881 static void
12882 ctl_datamove_remote_read(union ctl_io *io)
12883 {
12884         int retval;
12885         uint32_t i;
12886
12887         /*
12888          * This will send an error to the other controller in the case of a
12889          * failure.
12890          */
12891         retval = ctl_datamove_remote_sgl_setup(io);
12892         if (retval != 0)
12893                 return;
12894
12895         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12896                                           ctl_datamove_remote_read_cb);
12897         if (retval != 0) {
12898                 /*
12899                  * Make sure we free memory if there was an error..  The
12900                  * ctl_datamove_remote_xfer() function will send the
12901                  * datamove done message, or call the callback with an
12902                  * error if there is a problem.
12903                  */
12904                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12905                         free(CTL_LSGLT(io)[i].addr, M_CTL);
12906                 free(CTL_RSGL(io), M_CTL);
12907                 CTL_RSGL(io) = NULL;
12908                 CTL_LSGL(io) = NULL;
12909         }
12910 }
12911
12912 /*
12913  * Process a datamove request from the other controller.  This is used for
12914  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12915  * first.  Once that is complete, the data gets DMAed into the remote
12916  * controller's memory.  For reads, we DMA from the remote controller's
12917  * memory into our memory first, and then move it out to the FETD.
12918  */
12919 static void
12920 ctl_datamove_remote(union ctl_io *io)
12921 {
12922
12923         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12924
12925         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12926                 ctl_failover_io(io, /*have_lock*/ 0);
12927                 return;
12928         }
12929
12930         /*
12931          * Note that we look for an aborted I/O here, but don't do some of
12932          * the other checks that ctl_datamove() normally does.
12933          * We don't need to run the datamove delay code, since that should
12934          * have been done if need be on the other controller.
12935          */
12936         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12937                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
12938                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12939                        io->io_hdr.nexus.targ_port,
12940                        io->io_hdr.nexus.targ_lun);
12941                 io->io_hdr.port_status = 31338;
12942                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12943                 return;
12944         }
12945
12946         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
12947                 ctl_datamove_remote_write(io);
12948         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
12949                 ctl_datamove_remote_read(io);
12950         else {
12951                 io->io_hdr.port_status = 31339;
12952                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12953         }
12954 }
12955
12956 static void
12957 ctl_process_done(union ctl_io *io)
12958 {
12959         struct ctl_softc *softc = CTL_SOFTC(io);
12960         struct ctl_port *port = CTL_PORT(io);
12961         struct ctl_lun *lun = CTL_LUN(io);
12962         void (*fe_done)(union ctl_io *io);
12963         union ctl_ha_msg msg;
12964
12965         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12966         fe_done = port->fe_done;
12967
12968 #ifdef CTL_TIME_IO
12969         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12970                 char str[256];
12971                 char path_str[64];
12972                 struct sbuf sb;
12973
12974                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12975                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12976
12977                 sbuf_cat(&sb, path_str);
12978                 switch (io->io_hdr.io_type) {
12979                 case CTL_IO_SCSI:
12980                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12981                         sbuf_printf(&sb, "\n");
12982                         sbuf_cat(&sb, path_str);
12983                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12984                                     io->scsiio.tag_num, io->scsiio.tag_type);
12985                         break;
12986                 case CTL_IO_TASK:
12987                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12988                                     "Tag Type: %d\n", io->taskio.task_action,
12989                                     io->taskio.tag_num, io->taskio.tag_type);
12990                         break;
12991                 default:
12992                         panic("%s: Invalid CTL I/O type %d\n",
12993                             __func__, io->io_hdr.io_type);
12994                 }
12995                 sbuf_cat(&sb, path_str);
12996                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12997                             (intmax_t)time_uptime - io->io_hdr.start_time);
12998                 sbuf_finish(&sb);
12999                 printf("%s", sbuf_data(&sb));
13000         }
13001 #endif /* CTL_TIME_IO */
13002
13003         switch (io->io_hdr.io_type) {
13004         case CTL_IO_SCSI:
13005                 break;
13006         case CTL_IO_TASK:
13007                 if (ctl_debug & CTL_DEBUG_INFO)
13008                         ctl_io_error_print(io, NULL);
13009                 fe_done(io);
13010                 return;
13011         default:
13012                 panic("%s: Invalid CTL I/O type %d\n",
13013                     __func__, io->io_hdr.io_type);
13014         }
13015
13016         if (lun == NULL) {
13017                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13018                                  io->io_hdr.nexus.targ_mapped_lun));
13019                 goto bailout;
13020         }
13021
13022         mtx_lock(&lun->lun_lock);
13023
13024         /*
13025          * Check to see if we have any informational exception and status
13026          * of this command can be modified to report it in form of either
13027          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
13028          */
13029         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
13030             io->io_hdr.status == CTL_SUCCESS &&
13031             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
13032                 uint8_t mrie = lun->MODE_IE.mrie;
13033                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
13034                     (lun->MODE_VER.byte3 & SMS_VER_PER));
13035                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
13036                      mrie == SIEP_MRIE_REC_UNCOND ||
13037                      mrie == SIEP_MRIE_NO_SENSE) &&
13038                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
13039                      CTL_CMD_FLAG_NO_SENSE) == 0) {
13040                         ctl_set_sense(&io->scsiio,
13041                               /*current_error*/ 1,
13042                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
13043                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
13044                               /*asc*/ lun->ie_asc,
13045                               /*ascq*/ lun->ie_ascq,
13046                               SSD_ELEM_NONE);
13047                         lun->ie_reported = 1;
13048                 }
13049         } else if (lun->ie_reported < 0)
13050                 lun->ie_reported = 0;
13051
13052         /*
13053          * Check to see if we have any errors to inject here.  We only
13054          * inject errors for commands that don't already have errors set.
13055          */
13056         if (!STAILQ_EMPTY(&lun->error_list) &&
13057             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13058             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13059                 ctl_inject_error(lun, io);
13060
13061         /*
13062          * XXX KDM how do we treat commands that aren't completed
13063          * successfully?
13064          *
13065          * XXX KDM should we also track I/O latency?
13066          */
13067         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13068             io->io_hdr.io_type == CTL_IO_SCSI) {
13069                 int type;
13070 #ifdef CTL_TIME_IO
13071                 struct bintime bt;
13072
13073                 getbinuptime(&bt);
13074                 bintime_sub(&bt, &io->io_hdr.start_bt);
13075 #endif
13076                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13077                     CTL_FLAG_DATA_IN)
13078                         type = CTL_STATS_READ;
13079                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13080                     CTL_FLAG_DATA_OUT)
13081                         type = CTL_STATS_WRITE;
13082                 else
13083                         type = CTL_STATS_NO_IO;
13084
13085                 lun->stats.bytes[type] += io->scsiio.kern_total_len;
13086                 lun->stats.operations[type] ++;
13087                 lun->stats.dmas[type] += io->io_hdr.num_dmas;
13088 #ifdef CTL_TIME_IO
13089                 bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt);
13090                 bintime_add(&lun->stats.time[type], &bt);
13091 #endif
13092
13093                 mtx_lock(&port->port_lock);
13094                 port->stats.bytes[type] += io->scsiio.kern_total_len;
13095                 port->stats.operations[type] ++;
13096                 port->stats.dmas[type] += io->io_hdr.num_dmas;
13097 #ifdef CTL_TIME_IO
13098                 bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt);
13099                 bintime_add(&port->stats.time[type], &bt);
13100 #endif
13101                 mtx_unlock(&port->port_lock);
13102         }
13103
13104         /*
13105          * Run through the blocked queue of this I/O and see if anything
13106          * can be unblocked, now that this I/O is done and will be removed.
13107          * We need to do it before removal to have OOA position to start.
13108          */
13109         ctl_try_unblock_others(lun, io, TRUE);
13110
13111         /*
13112          * Remove this from the OOA queue.
13113          */
13114         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13115 #ifdef CTL_TIME_IO
13116         if (TAILQ_EMPTY(&lun->ooa_queue))
13117                 lun->last_busy = getsbinuptime();
13118 #endif
13119
13120         /*
13121          * If the LUN has been invalidated, free it if there is nothing
13122          * left on its OOA queue.
13123          */
13124         if ((lun->flags & CTL_LUN_INVALID)
13125          && TAILQ_EMPTY(&lun->ooa_queue)) {
13126                 mtx_unlock(&lun->lun_lock);
13127                 ctl_free_lun(lun);
13128         } else
13129                 mtx_unlock(&lun->lun_lock);
13130
13131 bailout:
13132
13133         /*
13134          * If this command has been aborted, make sure we set the status
13135          * properly.  The FETD is responsible for freeing the I/O and doing
13136          * whatever it needs to do to clean up its state.
13137          */
13138         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13139                 ctl_set_task_aborted(&io->scsiio);
13140
13141         /*
13142          * If enabled, print command error status.
13143          */
13144         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13145             (ctl_debug & CTL_DEBUG_INFO) != 0)
13146                 ctl_io_error_print(io, NULL);
13147
13148         /*
13149          * Tell the FETD or the other shelf controller we're done with this
13150          * command.  Note that only SCSI commands get to this point.  Task
13151          * management commands are completed above.
13152          */
13153         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13154             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13155                 memset(&msg, 0, sizeof(msg));
13156                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13157                 msg.hdr.serializing_sc = io->io_hdr.remote_io;
13158                 msg.hdr.nexus = io->io_hdr.nexus;
13159                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13160                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13161                     M_WAITOK);
13162         }
13163
13164         fe_done(io);
13165 }
13166
13167 /*
13168  * Front end should call this if it doesn't do autosense.  When the request
13169  * sense comes back in from the initiator, we'll dequeue this and send it.
13170  */
13171 int
13172 ctl_queue_sense(union ctl_io *io)
13173 {
13174         struct ctl_softc *softc = CTL_SOFTC(io);
13175         struct ctl_port *port = CTL_PORT(io);
13176         struct ctl_lun *lun;
13177         struct scsi_sense_data *ps;
13178         uint32_t initidx, p, targ_lun;
13179
13180         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13181
13182         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13183
13184         /*
13185          * LUN lookup will likely move to the ctl_work_thread() once we
13186          * have our new queueing infrastructure (that doesn't put things on
13187          * a per-LUN queue initially).  That is so that we can handle
13188          * things like an INQUIRY to a LUN that we don't have enabled.  We
13189          * can't deal with that right now.
13190          * If we don't have a LUN for this, just toss the sense information.
13191          */
13192         mtx_lock(&softc->ctl_lock);
13193         if (targ_lun >= ctl_max_luns ||
13194             (lun = softc->ctl_luns[targ_lun]) == NULL) {
13195                 mtx_unlock(&softc->ctl_lock);
13196                 goto bailout;
13197         }
13198         mtx_lock(&lun->lun_lock);
13199         mtx_unlock(&softc->ctl_lock);
13200
13201         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13202         p = initidx / CTL_MAX_INIT_PER_PORT;
13203         if (lun->pending_sense[p] == NULL) {
13204                 lun->pending_sense[p] = malloc(sizeof(*ps) * CTL_MAX_INIT_PER_PORT,
13205                     M_CTL, M_NOWAIT | M_ZERO);
13206         }
13207         if ((ps = lun->pending_sense[p]) != NULL) {
13208                 ps += initidx % CTL_MAX_INIT_PER_PORT;
13209                 memset(ps, 0, sizeof(*ps));
13210                 memcpy(ps, &io->scsiio.sense_data, io->scsiio.sense_len);
13211         }
13212         mtx_unlock(&lun->lun_lock);
13213
13214 bailout:
13215         ctl_free_io(io);
13216         return (CTL_RETVAL_COMPLETE);
13217 }
13218
13219 /*
13220  * Primary command inlet from frontend ports.  All SCSI and task I/O
13221  * requests must go through this function.
13222  */
13223 int
13224 ctl_queue(union ctl_io *io)
13225 {
13226         struct ctl_port *port = CTL_PORT(io);
13227
13228         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13229
13230 #ifdef CTL_TIME_IO
13231         io->io_hdr.start_time = time_uptime;
13232         getbinuptime(&io->io_hdr.start_bt);
13233 #endif /* CTL_TIME_IO */
13234
13235         /* Map FE-specific LUN ID into global one. */
13236         io->io_hdr.nexus.targ_mapped_lun =
13237             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13238
13239         switch (io->io_hdr.io_type) {
13240         case CTL_IO_SCSI:
13241         case CTL_IO_TASK:
13242                 if (ctl_debug & CTL_DEBUG_CDB)
13243                         ctl_io_print(io);
13244                 ctl_enqueue_incoming(io);
13245                 break;
13246         default:
13247                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13248                 return (EINVAL);
13249         }
13250
13251         return (CTL_RETVAL_COMPLETE);
13252 }
13253
13254 #ifdef CTL_IO_DELAY
13255 static void
13256 ctl_done_timer_wakeup(void *arg)
13257 {
13258         union ctl_io *io;
13259
13260         io = (union ctl_io *)arg;
13261         ctl_done(io);
13262 }
13263 #endif /* CTL_IO_DELAY */
13264
13265 void
13266 ctl_serseq_done(union ctl_io *io)
13267 {
13268         struct ctl_lun *lun = CTL_LUN(io);;
13269
13270         if (lun->be_lun == NULL ||
13271             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13272                 return;
13273         mtx_lock(&lun->lun_lock);
13274         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13275         ctl_try_unblock_others(lun, io, FALSE);
13276         mtx_unlock(&lun->lun_lock);
13277 }
13278
13279 void
13280 ctl_done(union ctl_io *io)
13281 {
13282
13283         /*
13284          * Enable this to catch duplicate completion issues.
13285          */
13286 #if 0
13287         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13288                 printf("%s: type %d msg %d cdb %x iptl: "
13289                        "%u:%u:%u tag 0x%04x "
13290                        "flag %#x status %x\n",
13291                         __func__,
13292                         io->io_hdr.io_type,
13293                         io->io_hdr.msg_type,
13294                         io->scsiio.cdb[0],
13295                         io->io_hdr.nexus.initid,
13296                         io->io_hdr.nexus.targ_port,
13297                         io->io_hdr.nexus.targ_lun,
13298                         (io->io_hdr.io_type ==
13299                         CTL_IO_TASK) ?
13300                         io->taskio.tag_num :
13301                         io->scsiio.tag_num,
13302                         io->io_hdr.flags,
13303                         io->io_hdr.status);
13304         } else
13305                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13306 #endif
13307
13308         /*
13309          * This is an internal copy of an I/O, and should not go through
13310          * the normal done processing logic.
13311          */
13312         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13313                 return;
13314
13315 #ifdef CTL_IO_DELAY
13316         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13317                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13318         } else {
13319                 struct ctl_lun *lun = CTL_LUN(io);
13320
13321                 if ((lun != NULL)
13322                  && (lun->delay_info.done_delay > 0)) {
13323
13324                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13325                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13326                         callout_reset(&io->io_hdr.delay_callout,
13327                                       lun->delay_info.done_delay * hz,
13328                                       ctl_done_timer_wakeup, io);
13329                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13330                                 lun->delay_info.done_delay = 0;
13331                         return;
13332                 }
13333         }
13334 #endif /* CTL_IO_DELAY */
13335
13336         ctl_enqueue_done(io);
13337 }
13338
13339 static void
13340 ctl_work_thread(void *arg)
13341 {
13342         struct ctl_thread *thr = (struct ctl_thread *)arg;
13343         struct ctl_softc *softc = thr->ctl_softc;
13344         union ctl_io *io;
13345         int retval;
13346
13347         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13348         thread_lock(curthread);
13349         sched_prio(curthread, PUSER - 1);
13350         thread_unlock(curthread);
13351
13352         while (!softc->shutdown) {
13353                 /*
13354                  * We handle the queues in this order:
13355                  * - ISC
13356                  * - done queue (to free up resources, unblock other commands)
13357                  * - incoming queue
13358                  * - RtR queue
13359                  *
13360                  * If those queues are empty, we break out of the loop and
13361                  * go to sleep.
13362                  */
13363                 mtx_lock(&thr->queue_lock);
13364                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13365                 if (io != NULL) {
13366                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13367                         mtx_unlock(&thr->queue_lock);
13368                         ctl_handle_isc(io);
13369                         continue;
13370                 }
13371                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13372                 if (io != NULL) {
13373                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13374                         /* clear any blocked commands, call fe_done */
13375                         mtx_unlock(&thr->queue_lock);
13376                         ctl_process_done(io);
13377                         continue;
13378                 }
13379                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13380                 if (io != NULL) {
13381                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13382                         mtx_unlock(&thr->queue_lock);
13383                         if (io->io_hdr.io_type == CTL_IO_TASK)
13384                                 ctl_run_task(io);
13385                         else
13386                                 ctl_scsiio_precheck(softc, &io->scsiio);
13387                         continue;
13388                 }
13389                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13390                 if (io != NULL) {
13391                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13392                         mtx_unlock(&thr->queue_lock);
13393                         retval = ctl_scsiio(&io->scsiio);
13394                         if (retval != CTL_RETVAL_COMPLETE)
13395                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13396                         continue;
13397                 }
13398
13399                 /* Sleep until we have something to do. */
13400                 mtx_sleep(thr, &thr->queue_lock, PDROP, "-", 0);
13401         }
13402         thr->thread = NULL;
13403         kthread_exit();
13404 }
13405
13406 static void
13407 ctl_lun_thread(void *arg)
13408 {
13409         struct ctl_softc *softc = (struct ctl_softc *)arg;
13410         struct ctl_be_lun *be_lun;
13411
13412         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13413         thread_lock(curthread);
13414         sched_prio(curthread, PUSER - 1);
13415         thread_unlock(curthread);
13416
13417         while (!softc->shutdown) {
13418                 mtx_lock(&softc->ctl_lock);
13419                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13420                 if (be_lun != NULL) {
13421                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13422                         mtx_unlock(&softc->ctl_lock);
13423                         ctl_create_lun(be_lun);
13424                         continue;
13425                 }
13426
13427                 /* Sleep until we have something to do. */
13428                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13429                     PDROP, "-", 0);
13430         }
13431         softc->lun_thread = NULL;
13432         kthread_exit();
13433 }
13434
13435 static void
13436 ctl_thresh_thread(void *arg)
13437 {
13438         struct ctl_softc *softc = (struct ctl_softc *)arg;
13439         struct ctl_lun *lun;
13440         struct ctl_logical_block_provisioning_page *page;
13441         const char *attr;
13442         union ctl_ha_msg msg;
13443         uint64_t thres, val;
13444         int i, e, set;
13445
13446         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13447         thread_lock(curthread);
13448         sched_prio(curthread, PUSER - 1);
13449         thread_unlock(curthread);
13450
13451         while (!softc->shutdown) {
13452                 mtx_lock(&softc->ctl_lock);
13453                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13454                         if ((lun->flags & CTL_LUN_DISABLED) ||
13455                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13456                             lun->backend->lun_attr == NULL)
13457                                 continue;
13458                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13459                             softc->ha_mode == CTL_HA_MODE_XFER)
13460                                 continue;
13461                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13462                                 continue;
13463                         e = 0;
13464                         page = &lun->MODE_LBP;
13465                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13466                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13467                                         continue;
13468                                 thres = scsi_4btoul(page->descr[i].count);
13469                                 thres <<= CTL_LBP_EXPONENT;
13470                                 switch (page->descr[i].resource) {
13471                                 case 0x01:
13472                                         attr = "blocksavail";
13473                                         break;
13474                                 case 0x02:
13475                                         attr = "blocksused";
13476                                         break;
13477                                 case 0xf1:
13478                                         attr = "poolblocksavail";
13479                                         break;
13480                                 case 0xf2:
13481                                         attr = "poolblocksused";
13482                                         break;
13483                                 default:
13484                                         continue;
13485                                 }
13486                                 mtx_unlock(&softc->ctl_lock); // XXX
13487                                 val = lun->backend->lun_attr(
13488                                     lun->be_lun->be_lun, attr);
13489                                 mtx_lock(&softc->ctl_lock);
13490                                 if (val == UINT64_MAX)
13491                                         continue;
13492                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13493                                     == SLBPPD_ARMING_INC)
13494                                         e = (val >= thres);
13495                                 else
13496                                         e = (val <= thres);
13497                                 if (e)
13498                                         break;
13499                         }
13500                         mtx_lock(&lun->lun_lock);
13501                         if (e) {
13502                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13503                                     (uint8_t *)page, lun->ua_tpt_info);
13504                                 if (lun->lasttpt == 0 ||
13505                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13506                                         lun->lasttpt = time_uptime;
13507                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13508                                         set = 1;
13509                                 } else
13510                                         set = 0;
13511                         } else {
13512                                 lun->lasttpt = 0;
13513                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13514                                 set = -1;
13515                         }
13516                         mtx_unlock(&lun->lun_lock);
13517                         if (set != 0 &&
13518                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13519                                 /* Send msg to other side. */
13520                                 bzero(&msg.ua, sizeof(msg.ua));
13521                                 msg.hdr.msg_type = CTL_MSG_UA;
13522                                 msg.hdr.nexus.initid = -1;
13523                                 msg.hdr.nexus.targ_port = -1;
13524                                 msg.hdr.nexus.targ_lun = lun->lun;
13525                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13526                                 msg.ua.ua_all = 1;
13527                                 msg.ua.ua_set = (set > 0);
13528                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13529                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13530                                 mtx_unlock(&softc->ctl_lock); // XXX
13531                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13532                                     sizeof(msg.ua), M_WAITOK);
13533                                 mtx_lock(&softc->ctl_lock);
13534                         }
13535                 }
13536                 mtx_sleep(&softc->thresh_thread, &softc->ctl_lock,
13537                     PDROP, "-", CTL_LBP_PERIOD * hz);
13538         }
13539         softc->thresh_thread = NULL;
13540         kthread_exit();
13541 }
13542
13543 static void
13544 ctl_enqueue_incoming(union ctl_io *io)
13545 {
13546         struct ctl_softc *softc = CTL_SOFTC(io);
13547         struct ctl_thread *thr;
13548         u_int idx;
13549
13550         idx = (io->io_hdr.nexus.targ_port * 127 +
13551                io->io_hdr.nexus.initid) % worker_threads;
13552         thr = &softc->threads[idx];
13553         mtx_lock(&thr->queue_lock);
13554         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13555         mtx_unlock(&thr->queue_lock);
13556         wakeup(thr);
13557 }
13558
13559 static void
13560 ctl_enqueue_rtr(union ctl_io *io)
13561 {
13562         struct ctl_softc *softc = CTL_SOFTC(io);
13563         struct ctl_thread *thr;
13564
13565         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13566         mtx_lock(&thr->queue_lock);
13567         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13568         mtx_unlock(&thr->queue_lock);
13569         wakeup(thr);
13570 }
13571
13572 static void
13573 ctl_enqueue_done(union ctl_io *io)
13574 {
13575         struct ctl_softc *softc = CTL_SOFTC(io);
13576         struct ctl_thread *thr;
13577
13578         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13579         mtx_lock(&thr->queue_lock);
13580         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13581         mtx_unlock(&thr->queue_lock);
13582         wakeup(thr);
13583 }
13584
13585 static void
13586 ctl_enqueue_isc(union ctl_io *io)
13587 {
13588         struct ctl_softc *softc = CTL_SOFTC(io);
13589         struct ctl_thread *thr;
13590
13591         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13592         mtx_lock(&thr->queue_lock);
13593         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13594         mtx_unlock(&thr->queue_lock);
13595         wakeup(thr);
13596 }
13597
13598 /*
13599  *  vim: ts=8
13600  */