sys/cam/ctl/ctl.c

   1 /*-
   2  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
   3  * Copyright (c) 2012 The FreeBSD Foundation
   4  * Copyright (c) 2014-2017 Alexander Motin <mav@FreeBSD.org>
   5  * All rights reserved.
   6  *
   7  * Portions of this software were developed by Edward Tomasz Napierala
   8  * under sponsorship from the FreeBSD Foundation.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions, and the following disclaimer,
  15  *    without modification.
  16  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
  17  *    substantially similar to the "NO WARRANTY" disclaimer below
  18  *    ("Disclaimer") and any redistribution must be conditioned upon
  19  *    including a substantially similar Disclaimer requirement for further
  20  *    binary redistribution.
  21  *
  22  * NO WARRANTY
  23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
  26  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  27  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  31  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
  32  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  33  * POSSIBILITY OF SUCH DAMAGES.
  34  *
  35  * $Id$
  36  */
  37 /*
  38  * CAM Target Layer, a SCSI device emulation subsystem.
  39  *
  40  * Author: Ken Merry <ken@FreeBSD.org>
  41  */
  42
  43 #include <sys/cdefs.h>
  44 __FBSDID("$FreeBSD$");
  45
  46 #include <sys/param.h>
  47 #include <sys/systm.h>
  48 #include <sys/ctype.h>
  49 #include <sys/kernel.h>
  50 #include <sys/types.h>
  51 #include <sys/kthread.h>
  52 #include <sys/bio.h>
  53 #include <sys/fcntl.h>
  54 #include <sys/lock.h>
  55 #include <sys/module.h>
  56 #include <sys/mutex.h>
  57 #include <sys/condvar.h>
  58 #include <sys/malloc.h>
  59 #include <sys/conf.h>
  60 #include <sys/ioccom.h>
  61 #include <sys/queue.h>
  62 #include <sys/sbuf.h>
  63 #include <sys/smp.h>
  64 #include <sys/endian.h>
  65 #include <sys/sysctl.h>
  66 #include <vm/uma.h>
  67
  68 #include <cam/cam.h>
  69 #include <cam/scsi/scsi_all.h>
  70 #include <cam/scsi/scsi_cd.h>
  71 #include <cam/scsi/scsi_da.h>
  72 #include <cam/ctl/ctl_io.h>
  73 #include <cam/ctl/ctl.h>
  74 #include <cam/ctl/ctl_frontend.h>
  75 #include <cam/ctl/ctl_util.h>
  76 #include <cam/ctl/ctl_backend.h>
  77 #include <cam/ctl/ctl_ioctl.h>
  78 #include <cam/ctl/ctl_ha.h>
  79 #include <cam/ctl/ctl_private.h>
  80 #include <cam/ctl/ctl_debug.h>
  81 #include <cam/ctl/ctl_scsi_all.h>
  82 #include <cam/ctl/ctl_error.h>
  83
  84 struct ctl_softc *control_softc = NULL;
  85
  86 /*
  87  * Template mode pages.
  88  */
  89
  90 /*
  91  * Note that these are default values only.  The actual values will be
  92  * filled in when the user does a mode sense.
  93  */
  94 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
  95         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
  96         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
  97         /*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
  98         /*read_retry_count*/0,
  99         /*correction_span*/0,
 100         /*head_offset_count*/0,
 101         /*data_strobe_offset_cnt*/0,
 102         /*byte8*/SMS_RWER_LBPERE,
 103         /*write_retry_count*/0,
 104         /*reserved2*/0,
 105         /*recovery_time_limit*/{0, 0},
 106 };
 107
 108 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
 109         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 110         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 111         /*byte3*/SMS_RWER_PER,
 112         /*read_retry_count*/0,
 113         /*correction_span*/0,
 114         /*head_offset_count*/0,
 115         /*data_strobe_offset_cnt*/0,
 116         /*byte8*/SMS_RWER_LBPERE,
 117         /*write_retry_count*/0,
 118         /*reserved2*/0,
 119         /*recovery_time_limit*/{0, 0},
 120 };
 121
 122 const static struct scsi_format_page format_page_default = {
 123         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 124         /*page_length*/sizeof(struct scsi_format_page) - 2,
 125         /*tracks_per_zone*/ {0, 0},
 126         /*alt_sectors_per_zone*/ {0, 0},
 127         /*alt_tracks_per_zone*/ {0, 0},
 128         /*alt_tracks_per_lun*/ {0, 0},
 129         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
 130                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
 131         /*bytes_per_sector*/ {0, 0},
 132         /*interleave*/ {0, 0},
 133         /*track_skew*/ {0, 0},
 134         /*cylinder_skew*/ {0, 0},
 135         /*flags*/ SFP_HSEC,
 136         /*reserved*/ {0, 0, 0}
 137 };
 138
 139 const static struct scsi_format_page format_page_changeable = {
 140         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 141         /*page_length*/sizeof(struct scsi_format_page) - 2,
 142         /*tracks_per_zone*/ {0, 0},
 143         /*alt_sectors_per_zone*/ {0, 0},
 144         /*alt_tracks_per_zone*/ {0, 0},
 145         /*alt_tracks_per_lun*/ {0, 0},
 146         /*sectors_per_track*/ {0, 0},
 147         /*bytes_per_sector*/ {0, 0},
 148         /*interleave*/ {0, 0},
 149         /*track_skew*/ {0, 0},
 150         /*cylinder_skew*/ {0, 0},
 151         /*flags*/ 0,
 152         /*reserved*/ {0, 0, 0}
 153 };
 154
 155 const static struct scsi_rigid_disk_page rigid_disk_page_default = {
 156         /*page_code*/SMS_RIGID_DISK_PAGE,
 157         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 158         /*cylinders*/ {0, 0, 0},
 159         /*heads*/ CTL_DEFAULT_HEADS,
 160         /*start_write_precomp*/ {0, 0, 0},
 161         /*start_reduced_current*/ {0, 0, 0},
 162         /*step_rate*/ {0, 0},
 163         /*landing_zone_cylinder*/ {0, 0, 0},
 164         /*rpl*/ SRDP_RPL_DISABLED,
 165         /*rotational_offset*/ 0,
 166         /*reserved1*/ 0,
 167         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
 168                            CTL_DEFAULT_ROTATION_RATE & 0xff},
 169         /*reserved2*/ {0, 0}
 170 };
 171
 172 const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
 173         /*page_code*/SMS_RIGID_DISK_PAGE,
 174         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 175         /*cylinders*/ {0, 0, 0},
 176         /*heads*/ 0,
 177         /*start_write_precomp*/ {0, 0, 0},
 178         /*start_reduced_current*/ {0, 0, 0},
 179         /*step_rate*/ {0, 0},
 180         /*landing_zone_cylinder*/ {0, 0, 0},
 181         /*rpl*/ 0,
 182         /*rotational_offset*/ 0,
 183         /*reserved1*/ 0,
 184         /*rotation_rate*/ {0, 0},
 185         /*reserved2*/ {0, 0}
 186 };
 187
 188 const static struct scsi_da_verify_recovery_page verify_er_page_default = {
 189         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 190         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 191         /*byte3*/0,
 192         /*read_retry_count*/0,
 193         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 194         /*recovery_time_limit*/{0, 0},
 195 };
 196
 197 const static struct scsi_da_verify_recovery_page verify_er_page_changeable = {
 198         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 199         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 200         /*byte3*/SMS_VER_PER,
 201         /*read_retry_count*/0,
 202         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 203         /*recovery_time_limit*/{0, 0},
 204 };
 205
 206 const static struct scsi_caching_page caching_page_default = {
 207         /*page_code*/SMS_CACHING_PAGE,
 208         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 209         /*flags1*/ SCP_DISC | SCP_WCE,
 210         /*ret_priority*/ 0,
 211         /*disable_pf_transfer_len*/ {0xff, 0xff},
 212         /*min_prefetch*/ {0, 0},
 213         /*max_prefetch*/ {0xff, 0xff},
 214         /*max_pf_ceiling*/ {0xff, 0xff},
 215         /*flags2*/ 0,
 216         /*cache_segments*/ 0,
 217         /*cache_seg_size*/ {0, 0},
 218         /*reserved*/ 0,
 219         /*non_cache_seg_size*/ {0, 0, 0}
 220 };
 221
 222 const static struct scsi_caching_page caching_page_changeable = {
 223         /*page_code*/SMS_CACHING_PAGE,
 224         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 225         /*flags1*/ SCP_WCE | SCP_RCD,
 226         /*ret_priority*/ 0,
 227         /*disable_pf_transfer_len*/ {0, 0},
 228         /*min_prefetch*/ {0, 0},
 229         /*max_prefetch*/ {0, 0},
 230         /*max_pf_ceiling*/ {0, 0},
 231         /*flags2*/ 0,
 232         /*cache_segments*/ 0,
 233         /*cache_seg_size*/ {0, 0},
 234         /*reserved*/ 0,
 235         /*non_cache_seg_size*/ {0, 0, 0}
 236 };
 237
 238 const static struct scsi_control_page control_page_default = {
 239         /*page_code*/SMS_CONTROL_MODE_PAGE,
 240         /*page_length*/sizeof(struct scsi_control_page) - 2,
 241         /*rlec*/0,
 242         /*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
 243         /*eca_and_aen*/0,
 244         /*flags4*/SCP_TAS,
 245         /*aen_holdoff_period*/{0, 0},
 246         /*busy_timeout_period*/{0, 0},
 247         /*extended_selftest_completion_time*/{0, 0}
 248 };
 249
 250 const static struct scsi_control_page control_page_changeable = {
 251         /*page_code*/SMS_CONTROL_MODE_PAGE,
 252         /*page_length*/sizeof(struct scsi_control_page) - 2,
 253         /*rlec*/SCP_DSENSE,
 254         /*queue_flags*/SCP_QUEUE_ALG_MASK | SCP_NUAR,
 255         /*eca_and_aen*/SCP_SWP,
 256         /*flags4*/0,
 257         /*aen_holdoff_period*/{0, 0},
 258         /*busy_timeout_period*/{0, 0},
 259         /*extended_selftest_completion_time*/{0, 0}
 260 };
 261
 262 #define CTL_CEM_LEN     (sizeof(struct scsi_control_ext_page) - 4)
 263
 264 const static struct scsi_control_ext_page control_ext_page_default = {
 265         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 266         /*subpage_code*/0x01,
 267         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 268         /*flags*/0,
 269         /*prio*/0,
 270         /*max_sense*/0
 271 };
 272
 273 const static struct scsi_control_ext_page control_ext_page_changeable = {
 274         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 275         /*subpage_code*/0x01,
 276         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 277         /*flags*/0,
 278         /*prio*/0,
 279         /*max_sense*/0xff
 280 };
 281
 282 const static struct scsi_info_exceptions_page ie_page_default = {
 283         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 284         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 285         /*info_flags*/SIEP_FLAGS_EWASC,
 286         /*mrie*/SIEP_MRIE_NO,
 287         /*interval_timer*/{0, 0, 0, 0},
 288         /*report_count*/{0, 0, 0, 1}
 289 };
 290
 291 const static struct scsi_info_exceptions_page ie_page_changeable = {
 292         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 293         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 294         /*info_flags*/SIEP_FLAGS_EWASC | SIEP_FLAGS_DEXCPT | SIEP_FLAGS_TEST |
 295             SIEP_FLAGS_LOGERR,
 296         /*mrie*/0x0f,
 297         /*interval_timer*/{0xff, 0xff, 0xff, 0xff},
 298         /*report_count*/{0xff, 0xff, 0xff, 0xff}
 299 };
 300
 301 #define CTL_LBPM_LEN    (sizeof(struct ctl_logical_block_provisioning_page) - 4)
 302
 303 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
 304         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 305         /*subpage_code*/0x02,
 306         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 307         /*flags*/0,
 308         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 309         /*descr*/{}},
 310         {{/*flags*/0,
 311           /*resource*/0x01,
 312           /*reserved*/{0, 0},
 313           /*count*/{0, 0, 0, 0}},
 314          {/*flags*/0,
 315           /*resource*/0x02,
 316           /*reserved*/{0, 0},
 317           /*count*/{0, 0, 0, 0}},
 318          {/*flags*/0,
 319           /*resource*/0xf1,
 320           /*reserved*/{0, 0},
 321           /*count*/{0, 0, 0, 0}},
 322          {/*flags*/0,
 323           /*resource*/0xf2,
 324           /*reserved*/{0, 0},
 325           /*count*/{0, 0, 0, 0}}
 326         }
 327 };
 328
 329 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
 330         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 331         /*subpage_code*/0x02,
 332         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 333         /*flags*/SLBPP_SITUA,
 334         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 335         /*descr*/{}},
 336         {{/*flags*/0,
 337           /*resource*/0,
 338           /*reserved*/{0, 0},
 339           /*count*/{0, 0, 0, 0}},
 340          {/*flags*/0,
 341           /*resource*/0,
 342           /*reserved*/{0, 0},
 343           /*count*/{0, 0, 0, 0}},
 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         }
 353 };
 354
 355 const static struct scsi_cddvd_capabilities_page cddvd_page_default = {
 356         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 357         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 358         /*caps1*/0x3f,
 359         /*caps2*/0x00,
 360         /*caps3*/0xf0,
 361         /*caps4*/0x00,
 362         /*caps5*/0x29,
 363         /*caps6*/0x00,
 364         /*obsolete*/{0, 0},
 365         /*nvol_levels*/{0, 0},
 366         /*buffer_size*/{8, 0},
 367         /*obsolete2*/{0, 0},
 368         /*reserved*/0,
 369         /*digital*/0,
 370         /*obsolete3*/0,
 371         /*copy_management*/0,
 372         /*reserved2*/0,
 373         /*rotation_control*/0,
 374         /*cur_write_speed*/0,
 375         /*num_speed_descr*/0,
 376 };
 377
 378 const static struct scsi_cddvd_capabilities_page cddvd_page_changeable = {
 379         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 380         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 381         /*caps1*/0,
 382         /*caps2*/0,
 383         /*caps3*/0,
 384         /*caps4*/0,
 385         /*caps5*/0,
 386         /*caps6*/0,
 387         /*obsolete*/{0, 0},
 388         /*nvol_levels*/{0, 0},
 389         /*buffer_size*/{0, 0},
 390         /*obsolete2*/{0, 0},
 391         /*reserved*/0,
 392         /*digital*/0,
 393         /*obsolete3*/0,
 394         /*copy_management*/0,
 395         /*reserved2*/0,
 396         /*rotation_control*/0,
 397         /*cur_write_speed*/0,
 398         /*num_speed_descr*/0,
 399 };
 400
 401 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
 402 static int worker_threads = -1;
 403 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 404     &worker_threads, 1, "Number of worker threads");
 405 static int ctl_debug = CTL_DEBUG_NONE;
 406 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
 407     &ctl_debug, 0, "Enabled debug flags");
 408 static int ctl_lun_map_size = 1024;
 409 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, lun_map_size, CTLFLAG_RWTUN,
 410     &ctl_lun_map_size, 0, "Size of per-port LUN map (max LUN + 1)");
 411 #ifdef  CTL_TIME_IO
 412 static int ctl_time_io_secs = CTL_TIME_IO_DEFAULT_SECS;
 413 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, time_io_secs, CTLFLAG_RWTUN,
 414     &ctl_time_io_secs, 0, "Log requests taking more seconds");
 415 #endif
 416
 417 /*
 418  * Maximum number of LUNs we support.  MUST be a power of 2.
 419  */
 420 #define CTL_DEFAULT_MAX_LUNS    1024
 421 static int ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
 422 TUNABLE_INT("kern.cam.ctl.max_luns", &ctl_max_luns);
 423 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_luns, CTLFLAG_RDTUN,
 424     &ctl_max_luns, CTL_DEFAULT_MAX_LUNS, "Maximum number of LUNs");
 425
 426 /*
 427  * Maximum number of ports registered at one time.
 428  */
 429 #define CTL_DEFAULT_MAX_PORTS           256
 430 static int ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
 431 TUNABLE_INT("kern.cam.ctl.max_ports", &ctl_max_ports);
 432 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_ports, CTLFLAG_RDTUN,
 433     &ctl_max_ports, CTL_DEFAULT_MAX_LUNS, "Maximum number of ports");
 434
 435 /*
 436  * Maximum number of initiators we support.
 437  */
 438 #define CTL_MAX_INITIATORS      (CTL_MAX_INIT_PER_PORT * ctl_max_ports)
 439
 440 /*
 441  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 442  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 443  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
 444  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
 445  */
 446 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
 447
 448 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 449                                   int param);
 450 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 451 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
 452 static int ctl_init(void);
 453 static int ctl_shutdown(void);
 454 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 455 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 456 static void ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 457 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 458                               struct ctl_ooa *ooa_hdr,
 459                               struct ctl_ooa_entry *kern_entries);
 460 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 461                      struct thread *td);
 462 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 463                          struct ctl_be_lun *be_lun);
 464 static int ctl_free_lun(struct ctl_lun *lun);
 465 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 466
 467 static int ctl_do_mode_select(union ctl_io *io);
 468 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 469                            uint64_t res_key, uint64_t sa_res_key,
 470                            uint8_t type, uint32_t residx,
 471                            struct ctl_scsiio *ctsio,
 472                            struct scsi_per_res_out *cdb,
 473                            struct scsi_per_res_out_parms* param);
 474 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 475                                   union ctl_ha_msg *msg);
 476 static void ctl_hndl_per_res_out_on_other_sc(union ctl_io *io);
 477 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 478 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 479 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 480 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 481 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 482 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 483                                          int alloc_len);
 484 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 485                                          int alloc_len);
 486 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 487 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 488 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 489 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 490 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 491 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
 492     bool seq);
 493 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
 494 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 495     union ctl_io *pending_io, union ctl_io *ooa_io);
 496 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 497                                 union ctl_io *starting_io);
 498 static int ctl_check_blocked(struct ctl_lun *lun);
 499 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
 500                                 const struct ctl_cmd_entry *entry,
 501                                 struct ctl_scsiio *ctsio);
 502 static void ctl_failover_lun(union ctl_io *io);
 503 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 504                                struct ctl_scsiio *ctsio);
 505 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 506
 507 static int ctl_target_reset(union ctl_io *io);
 508 static void ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx,
 509                          ctl_ua_type ua_type);
 510 static int ctl_lun_reset(union ctl_io *io);
 511 static int ctl_abort_task(union ctl_io *io);
 512 static int ctl_abort_task_set(union ctl_io *io);
 513 static int ctl_query_task(union ctl_io *io, int task_set);
 514 static void ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
 515                               ctl_ua_type ua_type);
 516 static int ctl_i_t_nexus_reset(union ctl_io *io);
 517 static int ctl_query_async_event(union ctl_io *io);
 518 static void ctl_run_task(union ctl_io *io);
 519 #ifdef CTL_IO_DELAY
 520 static void ctl_datamove_timer_wakeup(void *arg);
 521 static void ctl_done_timer_wakeup(void *arg);
 522 #endif /* CTL_IO_DELAY */
 523
 524 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 525 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 526 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 527 static void ctl_datamove_remote_write(union ctl_io *io);
 528 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 529 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 530 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 531 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 532                                     ctl_ha_dt_cb callback);
 533 static void ctl_datamove_remote_read(union ctl_io *io);
 534 static void ctl_datamove_remote(union ctl_io *io);
 535 static void ctl_process_done(union ctl_io *io);
 536 static void ctl_lun_thread(void *arg);
 537 static void ctl_thresh_thread(void *arg);
 538 static void ctl_work_thread(void *arg);
 539 static void ctl_enqueue_incoming(union ctl_io *io);
 540 static void ctl_enqueue_rtr(union ctl_io *io);
 541 static void ctl_enqueue_done(union ctl_io *io);
 542 static void ctl_enqueue_isc(union ctl_io *io);
 543 static const struct ctl_cmd_entry *
 544     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 545 static const struct ctl_cmd_entry *
 546     ctl_validate_command(struct ctl_scsiio *ctsio);
 547 static int ctl_cmd_applicable(uint8_t lun_type,
 548     const struct ctl_cmd_entry *entry);
 549 static int ctl_ha_init(void);
 550 static int ctl_ha_shutdown(void);
 551
 552 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
 553 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
 554 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
 555 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
 556
 557 /*
 558  * Load the serialization table.  This isn't very pretty, but is probably
 559  * the easiest way to do it.
 560  */
 561 #include "ctl_ser_table.c"
 562
 563 /*
 564  * We only need to define open, close and ioctl routines for this driver.
 565  */
 566 static struct cdevsw ctl_cdevsw = {
 567         .d_version =    D_VERSION,
 568         .d_flags =      0,
 569         .d_open =       ctl_open,
 570         .d_close =      ctl_close,
 571         .d_ioctl =      ctl_ioctl,
 572         .d_name =       "ctl",
 573 };
 574
 575
 576 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 577
 578 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 579
 580 static moduledata_t ctl_moduledata = {
 581         "ctl",
 582         ctl_module_event_handler,
 583         NULL
 584 };
 585
 586 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 587 MODULE_VERSION(ctl, 1);
 588
 589 static struct ctl_frontend ha_frontend =
 590 {
 591         .name = "ha",
 592         .init = ctl_ha_init,
 593         .shutdown = ctl_ha_shutdown,
 594 };
 595
 596 static int
 597 ctl_ha_init(void)
 598 {
 599         struct ctl_softc *softc = control_softc;
 600
 601         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
 602                             &softc->othersc_pool) != 0)
 603                 return (ENOMEM);
 604         if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
 605                 ctl_pool_free(softc->othersc_pool);
 606                 return (EIO);
 607         }
 608         if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
 609             != CTL_HA_STATUS_SUCCESS) {
 610                 ctl_ha_msg_destroy(softc);
 611                 ctl_pool_free(softc->othersc_pool);
 612                 return (EIO);
 613         }
 614         return (0);
 615 };
 616
 617 static int
 618 ctl_ha_shutdown(void)
 619 {
 620         struct ctl_softc *softc = control_softc;
 621         struct ctl_port *port;
 622
 623         ctl_ha_msg_shutdown(softc);
 624         if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL) != CTL_HA_STATUS_SUCCESS)
 625                 return (EIO);
 626         if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
 627                 return (EIO);
 628         ctl_pool_free(softc->othersc_pool);
 629         while ((port = STAILQ_FIRST(&ha_frontend.port_list)) != NULL) {
 630                 ctl_port_deregister(port);
 631                 free(port->port_name, M_CTL);
 632                 free(port, M_CTL);
 633         }
 634         return (0);
 635 };
 636
 637 static void
 638 ctl_ha_datamove(union ctl_io *io)
 639 {
 640         struct ctl_lun *lun = CTL_LUN(io);
 641         struct ctl_sg_entry *sgl;
 642         union ctl_ha_msg msg;
 643         uint32_t sg_entries_sent;
 644         int do_sg_copy, i, j;
 645
 646         memset(&msg.dt, 0, sizeof(msg.dt));
 647         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
 648         msg.hdr.original_sc = io->io_hdr.original_sc;
 649         msg.hdr.serializing_sc = io;
 650         msg.hdr.nexus = io->io_hdr.nexus;
 651         msg.hdr.status = io->io_hdr.status;
 652         msg.dt.flags = io->io_hdr.flags;
 653
 654         /*
 655          * We convert everything into a S/G list here.  We can't
 656          * pass by reference, only by value between controllers.
 657          * So we can't pass a pointer to the S/G list, only as many
 658          * S/G entries as we can fit in here.  If it's possible for
 659          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
 660          * then we need to break this up into multiple transfers.
 661          */
 662         if (io->scsiio.kern_sg_entries == 0) {
 663                 msg.dt.kern_sg_entries = 1;
 664 #if 0
 665                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 666                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 667                 } else {
 668                         /* XXX KDM use busdma here! */
 669                         msg.dt.sg_list[0].addr =
 670                             (void *)vtophys(io->scsiio.kern_data_ptr);
 671                 }
 672 #else
 673                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
 674                     ("HA does not support BUS_ADDR"));
 675                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 676 #endif
 677                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
 678                 do_sg_copy = 0;
 679         } else {
 680                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
 681                 do_sg_copy = 1;
 682         }
 683
 684         msg.dt.kern_data_len = io->scsiio.kern_data_len;
 685         msg.dt.kern_total_len = io->scsiio.kern_total_len;
 686         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
 687         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
 688         msg.dt.sg_sequence = 0;
 689
 690         /*
 691          * Loop until we've sent all of the S/G entries.  On the
 692          * other end, we'll recompose these S/G entries into one
 693          * contiguous list before processing.
 694          */
 695         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
 696             msg.dt.sg_sequence++) {
 697                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
 698                     sizeof(msg.dt.sg_list[0])),
 699                     msg.dt.kern_sg_entries - sg_entries_sent);
 700                 if (do_sg_copy != 0) {
 701                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
 702                         for (i = sg_entries_sent, j = 0;
 703                              i < msg.dt.cur_sg_entries; i++, j++) {
 704 #if 0
 705                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 706                                         msg.dt.sg_list[j].addr = sgl[i].addr;
 707                                 } else {
 708                                         /* XXX KDM use busdma here! */
 709                                         msg.dt.sg_list[j].addr =
 710                                             (void *)vtophys(sgl[i].addr);
 711                                 }
 712 #else
 713                                 KASSERT((io->io_hdr.flags &
 714                                     CTL_FLAG_BUS_ADDR) == 0,
 715                                     ("HA does not support BUS_ADDR"));
 716                                 msg.dt.sg_list[j].addr = sgl[i].addr;
 717 #endif
 718                                 msg.dt.sg_list[j].len = sgl[i].len;
 719                         }
 720                 }
 721
 722                 sg_entries_sent += msg.dt.cur_sg_entries;
 723                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
 724                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 725                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
 726                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
 727                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
 728                         io->io_hdr.port_status = 31341;
 729                         io->scsiio.be_move_done(io);
 730                         return;
 731                 }
 732                 msg.dt.sent_sg_entries = sg_entries_sent;
 733         }
 734
 735         /*
 736          * Officially handover the request from us to peer.
 737          * If failover has just happened, then we must return error.
 738          * If failover happen just after, then it is not our problem.
 739          */
 740         if (lun)
 741                 mtx_lock(&lun->lun_lock);
 742         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
 743                 if (lun)
 744                         mtx_unlock(&lun->lun_lock);
 745                 io->io_hdr.port_status = 31342;
 746                 io->scsiio.be_move_done(io);
 747                 return;
 748         }
 749         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
 750         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
 751         if (lun)
 752                 mtx_unlock(&lun->lun_lock);
 753 }
 754
 755 static void
 756 ctl_ha_done(union ctl_io *io)
 757 {
 758         union ctl_ha_msg msg;
 759
 760         if (io->io_hdr.io_type == CTL_IO_SCSI) {
 761                 memset(&msg, 0, sizeof(msg));
 762                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
 763                 msg.hdr.original_sc = io->io_hdr.original_sc;
 764                 msg.hdr.nexus = io->io_hdr.nexus;
 765                 msg.hdr.status = io->io_hdr.status;
 766                 msg.scsi.scsi_status = io->scsiio.scsi_status;
 767                 msg.scsi.tag_num = io->scsiio.tag_num;
 768                 msg.scsi.tag_type = io->scsiio.tag_type;
 769                 msg.scsi.sense_len = io->scsiio.sense_len;
 770                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
 771                     io->scsiio.sense_len);
 772                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 773                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
 774                     msg.scsi.sense_len, M_WAITOK);
 775         }
 776         ctl_free_io(io);
 777 }
 778
 779 static void
 780 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 781                             union ctl_ha_msg *msg_info)
 782 {
 783         struct ctl_scsiio *ctsio;
 784
 785         if (msg_info->hdr.original_sc == NULL) {
 786                 printf("%s: original_sc == NULL!\n", __func__);
 787                 /* XXX KDM now what? */
 788                 return;
 789         }
 790
 791         ctsio = &msg_info->hdr.original_sc->scsiio;
 792         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 793         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 794         ctsio->io_hdr.status = msg_info->hdr.status;
 795         ctsio->scsi_status = msg_info->scsi.scsi_status;
 796         ctsio->sense_len = msg_info->scsi.sense_len;
 797         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 798                msg_info->scsi.sense_len);
 799         ctl_enqueue_isc((union ctl_io *)ctsio);
 800 }
 801
 802 static void
 803 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 804                                 union ctl_ha_msg *msg_info)
 805 {
 806         struct ctl_scsiio *ctsio;
 807
 808         if (msg_info->hdr.serializing_sc == NULL) {
 809                 printf("%s: serializing_sc == NULL!\n", __func__);
 810                 /* XXX KDM now what? */
 811                 return;
 812         }
 813
 814         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 815         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 816         ctl_enqueue_isc((union ctl_io *)ctsio);
 817 }
 818
 819 void
 820 ctl_isc_announce_lun(struct ctl_lun *lun)
 821 {
 822         struct ctl_softc *softc = lun->ctl_softc;
 823         union ctl_ha_msg *msg;
 824         struct ctl_ha_msg_lun_pr_key pr_key;
 825         int i, k;
 826
 827         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 828                 return;
 829         mtx_lock(&lun->lun_lock);
 830         i = sizeof(msg->lun);
 831         if (lun->lun_devid)
 832                 i += lun->lun_devid->len;
 833         i += sizeof(pr_key) * lun->pr_key_count;
 834 alloc:
 835         mtx_unlock(&lun->lun_lock);
 836         msg = malloc(i, M_CTL, M_WAITOK);
 837         mtx_lock(&lun->lun_lock);
 838         k = sizeof(msg->lun);
 839         if (lun->lun_devid)
 840                 k += lun->lun_devid->len;
 841         k += sizeof(pr_key) * lun->pr_key_count;
 842         if (i < k) {
 843                 free(msg, M_CTL);
 844                 i = k;
 845                 goto alloc;
 846         }
 847         bzero(&msg->lun, sizeof(msg->lun));
 848         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 849         msg->hdr.nexus.targ_lun = lun->lun;
 850         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 851         msg->lun.flags = lun->flags;
 852         msg->lun.pr_generation = lun->pr_generation;
 853         msg->lun.pr_res_idx = lun->pr_res_idx;
 854         msg->lun.pr_res_type = lun->pr_res_type;
 855         msg->lun.pr_key_count = lun->pr_key_count;
 856         i = 0;
 857         if (lun->lun_devid) {
 858                 msg->lun.lun_devid_len = lun->lun_devid->len;
 859                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 860                     msg->lun.lun_devid_len);
 861                 i += msg->lun.lun_devid_len;
 862         }
 863         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 864                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 865                         continue;
 866                 pr_key.pr_iid = k;
 867                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 868                 i += sizeof(pr_key);
 869         }
 870         mtx_unlock(&lun->lun_lock);
 871         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 872             M_WAITOK);
 873         free(msg, M_CTL);
 874
 875         if (lun->flags & CTL_LUN_PRIMARY_SC) {
 876                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 877                         ctl_isc_announce_mode(lun, -1,
 878                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
 879                             lun->mode_pages.index[i].subpage);
 880                 }
 881         }
 882 }
 883
 884 void
 885 ctl_isc_announce_port(struct ctl_port *port)
 886 {
 887         struct ctl_softc *softc = port->ctl_softc;
 888         union ctl_ha_msg *msg;
 889         int i;
 890
 891         if (port->targ_port < softc->port_min ||
 892             port->targ_port >= softc->port_max ||
 893             softc->ha_link != CTL_HA_LINK_ONLINE)
 894                 return;
 895         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 896         if (port->lun_map)
 897                 i += port->lun_map_size * sizeof(uint32_t);
 898         if (port->port_devid)
 899                 i += port->port_devid->len;
 900         if (port->target_devid)
 901                 i += port->target_devid->len;
 902         if (port->init_devid)
 903                 i += port->init_devid->len;
 904         msg = malloc(i, M_CTL, M_WAITOK);
 905         bzero(&msg->port, sizeof(msg->port));
 906         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 907         msg->hdr.nexus.targ_port = port->targ_port;
 908         msg->port.port_type = port->port_type;
 909         msg->port.physical_port = port->physical_port;
 910         msg->port.virtual_port = port->virtual_port;
 911         msg->port.status = port->status;
 912         i = 0;
 913         msg->port.name_len = sprintf(&msg->port.data[i],
 914             "%d:%s", softc->ha_id, port->port_name) + 1;
 915         i += msg->port.name_len;
 916         if (port->lun_map) {
 917                 msg->port.lun_map_len = port->lun_map_size * sizeof(uint32_t);
 918                 memcpy(&msg->port.data[i], port->lun_map,
 919                     msg->port.lun_map_len);
 920                 i += msg->port.lun_map_len;
 921         }
 922         if (port->port_devid) {
 923                 msg->port.port_devid_len = port->port_devid->len;
 924                 memcpy(&msg->port.data[i], port->port_devid->data,
 925                     msg->port.port_devid_len);
 926                 i += msg->port.port_devid_len;
 927         }
 928         if (port->target_devid) {
 929                 msg->port.target_devid_len = port->target_devid->len;
 930                 memcpy(&msg->port.data[i], port->target_devid->data,
 931                     msg->port.target_devid_len);
 932                 i += msg->port.target_devid_len;
 933         }
 934         if (port->init_devid) {
 935                 msg->port.init_devid_len = port->init_devid->len;
 936                 memcpy(&msg->port.data[i], port->init_devid->data,
 937                     msg->port.init_devid_len);
 938                 i += msg->port.init_devid_len;
 939         }
 940         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 941             M_WAITOK);
 942         free(msg, M_CTL);
 943 }
 944
 945 void
 946 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 947 {
 948         struct ctl_softc *softc = port->ctl_softc;
 949         union ctl_ha_msg *msg;
 950         int i, l;
 951
 952         if (port->targ_port < softc->port_min ||
 953             port->targ_port >= softc->port_max ||
 954             softc->ha_link != CTL_HA_LINK_ONLINE)
 955                 return;
 956         mtx_lock(&softc->ctl_lock);
 957         i = sizeof(msg->iid);
 958         l = 0;
 959         if (port->wwpn_iid[iid].name)
 960                 l = strlen(port->wwpn_iid[iid].name) + 1;
 961         i += l;
 962         msg = malloc(i, M_CTL, M_NOWAIT);
 963         if (msg == NULL) {
 964                 mtx_unlock(&softc->ctl_lock);
 965                 return;
 966         }
 967         bzero(&msg->iid, sizeof(msg->iid));
 968         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 969         msg->hdr.nexus.targ_port = port->targ_port;
 970         msg->hdr.nexus.initid = iid;
 971         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 972         msg->iid.name_len = l;
 973         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 974         if (port->wwpn_iid[iid].name)
 975                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 976         mtx_unlock(&softc->ctl_lock);
 977         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 978         free(msg, M_CTL);
 979 }
 980
 981 void
 982 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
 983     uint8_t page, uint8_t subpage)
 984 {
 985         struct ctl_softc *softc = lun->ctl_softc;
 986         union ctl_ha_msg msg;
 987         u_int i;
 988
 989         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 990                 return;
 991         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 992                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
 993                     page && lun->mode_pages.index[i].subpage == subpage)
 994                         break;
 995         }
 996         if (i == CTL_NUM_MODE_PAGES)
 997                 return;
 998
 999         /* Don't try to replicate pages not present on this device. */
1000         if (lun->mode_pages.index[i].page_data == NULL)
1001                 return;
1002
1003         bzero(&msg.mode, sizeof(msg.mode));
1004         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
1005         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
1006         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
1007         msg.hdr.nexus.targ_lun = lun->lun;
1008         msg.hdr.nexus.targ_mapped_lun = lun->lun;
1009         msg.mode.page_code = page;
1010         msg.mode.subpage = subpage;
1011         msg.mode.page_len = lun->mode_pages.index[i].page_len;
1012         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
1013             msg.mode.page_len);
1014         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
1015             M_WAITOK);
1016 }
1017
1018 static void
1019 ctl_isc_ha_link_up(struct ctl_softc *softc)
1020 {
1021         struct ctl_port *port;
1022         struct ctl_lun *lun;
1023         union ctl_ha_msg msg;
1024         int i;
1025
1026         /* Announce this node parameters to peer for validation. */
1027         msg.login.msg_type = CTL_MSG_LOGIN;
1028         msg.login.version = CTL_HA_VERSION;
1029         msg.login.ha_mode = softc->ha_mode;
1030         msg.login.ha_id = softc->ha_id;
1031         msg.login.max_luns = ctl_max_luns;
1032         msg.login.max_ports = ctl_max_ports;
1033         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
1034         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
1035             M_WAITOK);
1036
1037         STAILQ_FOREACH(port, &softc->port_list, links) {
1038                 ctl_isc_announce_port(port);
1039                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1040                         if (port->wwpn_iid[i].in_use)
1041                                 ctl_isc_announce_iid(port, i);
1042                 }
1043         }
1044         STAILQ_FOREACH(lun, &softc->lun_list, links)
1045                 ctl_isc_announce_lun(lun);
1046 }
1047
1048 static void
1049 ctl_isc_ha_link_down(struct ctl_softc *softc)
1050 {
1051         struct ctl_port *port;
1052         struct ctl_lun *lun;
1053         union ctl_io *io;
1054         int i;
1055
1056         mtx_lock(&softc->ctl_lock);
1057         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1058                 mtx_lock(&lun->lun_lock);
1059                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
1060                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1061                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1062                 }
1063                 mtx_unlock(&lun->lun_lock);
1064
1065                 mtx_unlock(&softc->ctl_lock);
1066                 io = ctl_alloc_io(softc->othersc_pool);
1067                 mtx_lock(&softc->ctl_lock);
1068                 ctl_zero_io(io);
1069                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1070                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1071                 ctl_enqueue_isc(io);
1072         }
1073
1074         STAILQ_FOREACH(port, &softc->port_list, links) {
1075                 if (port->targ_port >= softc->port_min &&
1076                     port->targ_port < softc->port_max)
1077                         continue;
1078                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1079                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1080                         port->wwpn_iid[i].in_use = 0;
1081                         free(port->wwpn_iid[i].name, M_CTL);
1082                         port->wwpn_iid[i].name = NULL;
1083                 }
1084         }
1085         mtx_unlock(&softc->ctl_lock);
1086 }
1087
1088 static void
1089 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1090 {
1091         struct ctl_lun *lun;
1092         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
1093
1094         mtx_lock(&softc->ctl_lock);
1095         if (msg->hdr.nexus.targ_mapped_lun >= ctl_max_luns ||
1096             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) == NULL) {
1097                 mtx_unlock(&softc->ctl_lock);
1098                 return;
1099         }
1100         mtx_lock(&lun->lun_lock);
1101         mtx_unlock(&softc->ctl_lock);
1102         if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES && msg->ua.ua_set)
1103                 memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1104         if (msg->ua.ua_all) {
1105                 if (msg->ua.ua_set)
1106                         ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1107                 else
1108                         ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1109         } else {
1110                 if (msg->ua.ua_set)
1111                         ctl_est_ua(lun, iid, msg->ua.ua_type);
1112                 else
1113                         ctl_clr_ua(lun, iid, msg->ua.ua_type);
1114         }
1115         mtx_unlock(&lun->lun_lock);
1116 }
1117
1118 static void
1119 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1120 {
1121         struct ctl_lun *lun;
1122         struct ctl_ha_msg_lun_pr_key pr_key;
1123         int i, k;
1124         ctl_lun_flags oflags;
1125         uint32_t targ_lun;
1126
1127         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1128         mtx_lock(&softc->ctl_lock);
1129         if (targ_lun >= ctl_max_luns ||
1130             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1131                 mtx_unlock(&softc->ctl_lock);
1132                 return;
1133         }
1134         mtx_lock(&lun->lun_lock);
1135         mtx_unlock(&softc->ctl_lock);
1136         if (lun->flags & CTL_LUN_DISABLED) {
1137                 mtx_unlock(&lun->lun_lock);
1138                 return;
1139         }
1140         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1141         if (msg->lun.lun_devid_len != i || (i > 0 &&
1142             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1143                 mtx_unlock(&lun->lun_lock);
1144                 printf("%s: Received conflicting HA LUN %d\n",
1145                     __func__, targ_lun);
1146                 return;
1147         } else {
1148                 /* Record whether peer is primary. */
1149                 oflags = lun->flags;
1150                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1151                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1152                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1153                 else
1154                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1155                 if (oflags != lun->flags)
1156                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1157
1158                 /* If peer is primary and we are not -- use data */
1159                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1160                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1161                         lun->pr_generation = msg->lun.pr_generation;
1162                         lun->pr_res_idx = msg->lun.pr_res_idx;
1163                         lun->pr_res_type = msg->lun.pr_res_type;
1164                         lun->pr_key_count = msg->lun.pr_key_count;
1165                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1166                                 ctl_clr_prkey(lun, k);
1167                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1168                                 memcpy(&pr_key, &msg->lun.data[i],
1169                                     sizeof(pr_key));
1170                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1171                                 ctl_set_prkey(lun, pr_key.pr_iid,
1172                                     pr_key.pr_key);
1173                                 i += sizeof(pr_key);
1174                         }
1175                 }
1176
1177                 mtx_unlock(&lun->lun_lock);
1178                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1179                     __func__, targ_lun,
1180                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1181                     "primary" : "secondary"));
1182
1183                 /* If we are primary but peer doesn't know -- notify */
1184                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1185                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1186                         ctl_isc_announce_lun(lun);
1187         }
1188 }
1189
1190 static void
1191 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1192 {
1193         struct ctl_port *port;
1194         struct ctl_lun *lun;
1195         int i, new;
1196
1197         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1198         if (port == NULL) {
1199                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1200                     msg->hdr.nexus.targ_port));
1201                 new = 1;
1202                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1203                 port->frontend = &ha_frontend;
1204                 port->targ_port = msg->hdr.nexus.targ_port;
1205                 port->fe_datamove = ctl_ha_datamove;
1206                 port->fe_done = ctl_ha_done;
1207         } else if (port->frontend == &ha_frontend) {
1208                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1209                     msg->hdr.nexus.targ_port));
1210                 new = 0;
1211         } else {
1212                 printf("%s: Received conflicting HA port %d\n",
1213                     __func__, msg->hdr.nexus.targ_port);
1214                 return;
1215         }
1216         port->port_type = msg->port.port_type;
1217         port->physical_port = msg->port.physical_port;
1218         port->virtual_port = msg->port.virtual_port;
1219         port->status = msg->port.status;
1220         i = 0;
1221         free(port->port_name, M_CTL);
1222         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1223             M_CTL);
1224         i += msg->port.name_len;
1225         if (msg->port.lun_map_len != 0) {
1226                 if (port->lun_map == NULL ||
1227                     port->lun_map_size * sizeof(uint32_t) <
1228                     msg->port.lun_map_len) {
1229                         port->lun_map_size = 0;
1230                         free(port->lun_map, M_CTL);
1231                         port->lun_map = malloc(msg->port.lun_map_len,
1232                             M_CTL, M_WAITOK);
1233                 }
1234                 memcpy(port->lun_map, &msg->port.data[i], msg->port.lun_map_len);
1235                 port->lun_map_size = msg->port.lun_map_len / sizeof(uint32_t);
1236                 i += msg->port.lun_map_len;
1237         } else {
1238                 port->lun_map_size = 0;
1239                 free(port->lun_map, M_CTL);
1240                 port->lun_map = NULL;
1241         }
1242         if (msg->port.port_devid_len != 0) {
1243                 if (port->port_devid == NULL ||
1244                     port->port_devid->len < msg->port.port_devid_len) {
1245                         free(port->port_devid, M_CTL);
1246                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1247                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1248                 }
1249                 memcpy(port->port_devid->data, &msg->port.data[i],
1250                     msg->port.port_devid_len);
1251                 port->port_devid->len = msg->port.port_devid_len;
1252                 i += msg->port.port_devid_len;
1253         } else {
1254                 free(port->port_devid, M_CTL);
1255                 port->port_devid = NULL;
1256         }
1257         if (msg->port.target_devid_len != 0) {
1258                 if (port->target_devid == NULL ||
1259                     port->target_devid->len < msg->port.target_devid_len) {
1260                         free(port->target_devid, M_CTL);
1261                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1262                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1263                 }
1264                 memcpy(port->target_devid->data, &msg->port.data[i],
1265                     msg->port.target_devid_len);
1266                 port->target_devid->len = msg->port.target_devid_len;
1267                 i += msg->port.target_devid_len;
1268         } else {
1269                 free(port->target_devid, M_CTL);
1270                 port->target_devid = NULL;
1271         }
1272         if (msg->port.init_devid_len != 0) {
1273                 if (port->init_devid == NULL ||
1274                     port->init_devid->len < msg->port.init_devid_len) {
1275                         free(port->init_devid, M_CTL);
1276                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1277                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1278                 }
1279                 memcpy(port->init_devid->data, &msg->port.data[i],
1280                     msg->port.init_devid_len);
1281                 port->init_devid->len = msg->port.init_devid_len;
1282                 i += msg->port.init_devid_len;
1283         } else {
1284                 free(port->init_devid, M_CTL);
1285                 port->init_devid = NULL;
1286         }
1287         if (new) {
1288                 if (ctl_port_register(port) != 0) {
1289                         printf("%s: ctl_port_register() failed with error\n",
1290                             __func__);
1291                 }
1292         }
1293         mtx_lock(&softc->ctl_lock);
1294         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1295                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
1296                         continue;
1297                 mtx_lock(&lun->lun_lock);
1298                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1299                 mtx_unlock(&lun->lun_lock);
1300         }
1301         mtx_unlock(&softc->ctl_lock);
1302 }
1303
1304 static void
1305 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1306 {
1307         struct ctl_port *port;
1308         int iid;
1309
1310         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1311         if (port == NULL) {
1312                 printf("%s: Received IID for unknown port %d\n",
1313                     __func__, msg->hdr.nexus.targ_port);
1314                 return;
1315         }
1316         iid = msg->hdr.nexus.initid;
1317         if (port->wwpn_iid[iid].in_use != 0 &&
1318             msg->iid.in_use == 0)
1319                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
1320         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1321         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1322         free(port->wwpn_iid[iid].name, M_CTL);
1323         if (msg->iid.name_len) {
1324                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1325                     msg->iid.name_len, M_CTL);
1326         } else
1327                 port->wwpn_iid[iid].name = NULL;
1328 }
1329
1330 static void
1331 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1332 {
1333
1334         if (msg->login.version != CTL_HA_VERSION) {
1335                 printf("CTL HA peers have different versions %d != %d\n",
1336                     msg->login.version, CTL_HA_VERSION);
1337                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1338                 return;
1339         }
1340         if (msg->login.ha_mode != softc->ha_mode) {
1341                 printf("CTL HA peers have different ha_mode %d != %d\n",
1342                     msg->login.ha_mode, softc->ha_mode);
1343                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1344                 return;
1345         }
1346         if (msg->login.ha_id == softc->ha_id) {
1347                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1348                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1349                 return;
1350         }
1351         if (msg->login.max_luns != ctl_max_luns ||
1352             msg->login.max_ports != ctl_max_ports ||
1353             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1354                 printf("CTL HA peers have different limits\n");
1355                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1356                 return;
1357         }
1358 }
1359
1360 static void
1361 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1362 {
1363         struct ctl_lun *lun;
1364         u_int i;
1365         uint32_t initidx, targ_lun;
1366
1367         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1368         mtx_lock(&softc->ctl_lock);
1369         if (targ_lun >= ctl_max_luns ||
1370             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1371                 mtx_unlock(&softc->ctl_lock);
1372                 return;
1373         }
1374         mtx_lock(&lun->lun_lock);
1375         mtx_unlock(&softc->ctl_lock);
1376         if (lun->flags & CTL_LUN_DISABLED) {
1377                 mtx_unlock(&lun->lun_lock);
1378                 return;
1379         }
1380         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1381                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1382                     msg->mode.page_code &&
1383                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1384                         break;
1385         }
1386         if (i == CTL_NUM_MODE_PAGES) {
1387                 mtx_unlock(&lun->lun_lock);
1388                 return;
1389         }
1390         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1391             lun->mode_pages.index[i].page_len);
1392         initidx = ctl_get_initindex(&msg->hdr.nexus);
1393         if (initidx != -1)
1394                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1395         mtx_unlock(&lun->lun_lock);
1396 }
1397
1398 /*
1399  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1400  * subsystem come in here.
1401  */
1402 static void
1403 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1404 {
1405         struct ctl_softc *softc = control_softc;
1406         union ctl_io *io;
1407         struct ctl_prio *presio;
1408         ctl_ha_status isc_status;
1409
1410         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1411         if (event == CTL_HA_EVT_MSG_RECV) {
1412                 union ctl_ha_msg *msg, msgbuf;
1413
1414                 if (param > sizeof(msgbuf))
1415                         msg = malloc(param, M_CTL, M_WAITOK);
1416                 else
1417                         msg = &msgbuf;
1418                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1419                     M_WAITOK);
1420                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1421                         printf("%s: Error receiving message: %d\n",
1422                             __func__, isc_status);
1423                         if (msg != &msgbuf)
1424                                 free(msg, M_CTL);
1425                         return;
1426                 }
1427
1428                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1429                 switch (msg->hdr.msg_type) {
1430                 case CTL_MSG_SERIALIZE:
1431                         io = ctl_alloc_io(softc->othersc_pool);
1432                         ctl_zero_io(io);
1433                         // populate ctsio from msg
1434                         io->io_hdr.io_type = CTL_IO_SCSI;
1435                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1436                         io->io_hdr.original_sc = msg->hdr.original_sc;
1437                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1438                                             CTL_FLAG_IO_ACTIVE;
1439                         /*
1440                          * If we're in serialization-only mode, we don't
1441                          * want to go through full done processing.  Thus
1442                          * the COPY flag.
1443                          *
1444                          * XXX KDM add another flag that is more specific.
1445                          */
1446                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1447                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1448                         io->io_hdr.nexus = msg->hdr.nexus;
1449 #if 0
1450                         printf("port %u, iid %u, lun %u\n",
1451                                io->io_hdr.nexus.targ_port,
1452                                io->io_hdr.nexus.initid,
1453                                io->io_hdr.nexus.targ_lun);
1454 #endif
1455                         io->scsiio.tag_num = msg->scsi.tag_num;
1456                         io->scsiio.tag_type = msg->scsi.tag_type;
1457 #ifdef CTL_TIME_IO
1458                         io->io_hdr.start_time = time_uptime;
1459                         getbinuptime(&io->io_hdr.start_bt);
1460 #endif /* CTL_TIME_IO */
1461                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1462                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1463                                CTL_MAX_CDBLEN);
1464                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1465                                 const struct ctl_cmd_entry *entry;
1466
1467                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1468                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1469                                 io->io_hdr.flags |=
1470                                         entry->flags & CTL_FLAG_DATA_MASK;
1471                         }
1472                         ctl_enqueue_isc(io);
1473                         break;
1474
1475                 /* Performed on the Originating SC, XFER mode only */
1476                 case CTL_MSG_DATAMOVE: {
1477                         struct ctl_sg_entry *sgl;
1478                         int i, j;
1479
1480                         io = msg->hdr.original_sc;
1481                         if (io == NULL) {
1482                                 printf("%s: original_sc == NULL!\n", __func__);
1483                                 /* XXX KDM do something here */
1484                                 break;
1485                         }
1486                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1487                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1488                         /*
1489                          * Keep track of this, we need to send it back over
1490                          * when the datamove is complete.
1491                          */
1492                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1493                         if (msg->hdr.status == CTL_SUCCESS)
1494                                 io->io_hdr.status = msg->hdr.status;
1495
1496                         if (msg->dt.sg_sequence == 0) {
1497 #ifdef CTL_TIME_IO
1498                                 getbinuptime(&io->io_hdr.dma_start_bt);
1499 #endif
1500                                 i = msg->dt.kern_sg_entries +
1501                                     msg->dt.kern_data_len /
1502                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1503                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1504                                     M_WAITOK | M_ZERO);
1505                                 io->io_hdr.remote_sglist = sgl;
1506                                 io->io_hdr.local_sglist =
1507                                     &sgl[msg->dt.kern_sg_entries];
1508
1509                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1510
1511                                 io->scsiio.kern_sg_entries =
1512                                         msg->dt.kern_sg_entries;
1513                                 io->scsiio.rem_sg_entries =
1514                                         msg->dt.kern_sg_entries;
1515                                 io->scsiio.kern_data_len =
1516                                         msg->dt.kern_data_len;
1517                                 io->scsiio.kern_total_len =
1518                                         msg->dt.kern_total_len;
1519                                 io->scsiio.kern_data_resid =
1520                                         msg->dt.kern_data_resid;
1521                                 io->scsiio.kern_rel_offset =
1522                                         msg->dt.kern_rel_offset;
1523                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1524                                 io->io_hdr.flags |= msg->dt.flags &
1525                                     CTL_FLAG_BUS_ADDR;
1526                         } else
1527                                 sgl = (struct ctl_sg_entry *)
1528                                         io->scsiio.kern_data_ptr;
1529
1530                         for (i = msg->dt.sent_sg_entries, j = 0;
1531                              i < (msg->dt.sent_sg_entries +
1532                              msg->dt.cur_sg_entries); i++, j++) {
1533                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1534                                 sgl[i].len = msg->dt.sg_list[j].len;
1535
1536 #if 0
1537                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1538                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1539 #endif
1540                         }
1541
1542                         /*
1543                          * If this is the last piece of the I/O, we've got
1544                          * the full S/G list.  Queue processing in the thread.
1545                          * Otherwise wait for the next piece.
1546                          */
1547                         if (msg->dt.sg_last != 0)
1548                                 ctl_enqueue_isc(io);
1549                         break;
1550                 }
1551                 /* Performed on the Serializing (primary) SC, XFER mode only */
1552                 case CTL_MSG_DATAMOVE_DONE: {
1553                         if (msg->hdr.serializing_sc == NULL) {
1554                                 printf("%s: serializing_sc == NULL!\n",
1555                                        __func__);
1556                                 /* XXX KDM now what? */
1557                                 break;
1558                         }
1559                         /*
1560                          * We grab the sense information here in case
1561                          * there was a failure, so we can return status
1562                          * back to the initiator.
1563                          */
1564                         io = msg->hdr.serializing_sc;
1565                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1566                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1567                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1568                         io->io_hdr.port_status = msg->scsi.port_status;
1569                         io->scsiio.kern_data_resid = msg->scsi.kern_data_resid;
1570                         if (msg->hdr.status != CTL_STATUS_NONE) {
1571                                 io->io_hdr.status = msg->hdr.status;
1572                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1573                                 io->scsiio.sense_len = msg->scsi.sense_len;
1574                                 memcpy(&io->scsiio.sense_data,
1575                                     &msg->scsi.sense_data,
1576                                     msg->scsi.sense_len);
1577                                 if (msg->hdr.status == CTL_SUCCESS)
1578                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1579                         }
1580                         ctl_enqueue_isc(io);
1581                         break;
1582                 }
1583
1584                 /* Preformed on Originating SC, SER_ONLY mode */
1585                 case CTL_MSG_R2R:
1586                         io = msg->hdr.original_sc;
1587                         if (io == NULL) {
1588                                 printf("%s: original_sc == NULL!\n",
1589                                     __func__);
1590                                 break;
1591                         }
1592                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1593                         io->io_hdr.msg_type = CTL_MSG_R2R;
1594                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1595                         ctl_enqueue_isc(io);
1596                         break;
1597
1598                 /*
1599                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1600                  * mode.
1601                  * Performed on the Originating (i.e. secondary) SC in XFER
1602                  * mode
1603                  */
1604                 case CTL_MSG_FINISH_IO:
1605                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1606                                 ctl_isc_handler_finish_xfer(softc, msg);
1607                         else
1608                                 ctl_isc_handler_finish_ser_only(softc, msg);
1609                         break;
1610
1611                 /* Preformed on Originating SC */
1612                 case CTL_MSG_BAD_JUJU:
1613                         io = msg->hdr.original_sc;
1614                         if (io == NULL) {
1615                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1616                                        __func__);
1617                                 break;
1618                         }
1619                         ctl_copy_sense_data(msg, io);
1620                         /*
1621                          * IO should have already been cleaned up on other
1622                          * SC so clear this flag so we won't send a message
1623                          * back to finish the IO there.
1624                          */
1625                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1626                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1627
1628                         /* io = msg->hdr.serializing_sc; */
1629                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1630                         ctl_enqueue_isc(io);
1631                         break;
1632
1633                 /* Handle resets sent from the other side */
1634                 case CTL_MSG_MANAGE_TASKS: {
1635                         struct ctl_taskio *taskio;
1636                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1637                             softc->othersc_pool);
1638                         ctl_zero_io((union ctl_io *)taskio);
1639                         taskio->io_hdr.io_type = CTL_IO_TASK;
1640                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1641                         taskio->io_hdr.nexus = msg->hdr.nexus;
1642                         taskio->task_action = msg->task.task_action;
1643                         taskio->tag_num = msg->task.tag_num;
1644                         taskio->tag_type = msg->task.tag_type;
1645 #ifdef CTL_TIME_IO
1646                         taskio->io_hdr.start_time = time_uptime;
1647                         getbinuptime(&taskio->io_hdr.start_bt);
1648 #endif /* CTL_TIME_IO */
1649                         ctl_run_task((union ctl_io *)taskio);
1650                         break;
1651                 }
1652                 /* Persistent Reserve action which needs attention */
1653                 case CTL_MSG_PERS_ACTION:
1654                         presio = (struct ctl_prio *)ctl_alloc_io(
1655                             softc->othersc_pool);
1656                         ctl_zero_io((union ctl_io *)presio);
1657                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1658                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1659                         presio->io_hdr.nexus = msg->hdr.nexus;
1660                         presio->pr_msg = msg->pr;
1661                         ctl_enqueue_isc((union ctl_io *)presio);
1662                         break;
1663                 case CTL_MSG_UA:
1664                         ctl_isc_ua(softc, msg, param);
1665                         break;
1666                 case CTL_MSG_PORT_SYNC:
1667                         ctl_isc_port_sync(softc, msg, param);
1668                         break;
1669                 case CTL_MSG_LUN_SYNC:
1670                         ctl_isc_lun_sync(softc, msg, param);
1671                         break;
1672                 case CTL_MSG_IID_SYNC:
1673                         ctl_isc_iid_sync(softc, msg, param);
1674                         break;
1675                 case CTL_MSG_LOGIN:
1676                         ctl_isc_login(softc, msg, param);
1677                         break;
1678                 case CTL_MSG_MODE_SYNC:
1679                         ctl_isc_mode_sync(softc, msg, param);
1680                         break;
1681                 default:
1682                         printf("Received HA message of unknown type %d\n",
1683                             msg->hdr.msg_type);
1684                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1685                         break;
1686                 }
1687                 if (msg != &msgbuf)
1688                         free(msg, M_CTL);
1689         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1690                 printf("CTL: HA link status changed from %d to %d\n",
1691                     softc->ha_link, param);
1692                 if (param == softc->ha_link)
1693                         return;
1694                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1695                         softc->ha_link = param;
1696                         ctl_isc_ha_link_down(softc);
1697                 } else {
1698                         softc->ha_link = param;
1699                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1700                                 ctl_isc_ha_link_up(softc);
1701                 }
1702                 return;
1703         } else {
1704                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1705                 return;
1706         }
1707 }
1708
1709 static void
1710 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1711 {
1712
1713         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1714             src->scsi.sense_len);
1715         dest->scsiio.scsi_status = src->scsi.scsi_status;
1716         dest->scsiio.sense_len = src->scsi.sense_len;
1717         dest->io_hdr.status = src->hdr.status;
1718 }
1719
1720 static void
1721 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1722 {
1723
1724         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1725             src->scsiio.sense_len);
1726         dest->scsi.scsi_status = src->scsiio.scsi_status;
1727         dest->scsi.sense_len = src->scsiio.sense_len;
1728         dest->hdr.status = src->io_hdr.status;
1729 }
1730
1731 void
1732 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1733 {
1734         struct ctl_softc *softc = lun->ctl_softc;
1735         ctl_ua_type *pu;
1736
1737         if (initidx < softc->init_min || initidx >= softc->init_max)
1738                 return;
1739         mtx_assert(&lun->lun_lock, MA_OWNED);
1740         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1741         if (pu == NULL)
1742                 return;
1743         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1744 }
1745
1746 void
1747 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1748 {
1749         int i;
1750
1751         mtx_assert(&lun->lun_lock, MA_OWNED);
1752         if (lun->pending_ua[port] == NULL)
1753                 return;
1754         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1755                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1756                         continue;
1757                 lun->pending_ua[port][i] |= ua;
1758         }
1759 }
1760
1761 void
1762 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1763 {
1764         struct ctl_softc *softc = lun->ctl_softc;
1765         int i;
1766
1767         mtx_assert(&lun->lun_lock, MA_OWNED);
1768         for (i = softc->port_min; i < softc->port_max; i++)
1769                 ctl_est_ua_port(lun, i, except, ua);
1770 }
1771
1772 void
1773 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1774 {
1775         struct ctl_softc *softc = lun->ctl_softc;
1776         ctl_ua_type *pu;
1777
1778         if (initidx < softc->init_min || initidx >= softc->init_max)
1779                 return;
1780         mtx_assert(&lun->lun_lock, MA_OWNED);
1781         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1782         if (pu == NULL)
1783                 return;
1784         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1785 }
1786
1787 void
1788 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1789 {
1790         struct ctl_softc *softc = lun->ctl_softc;
1791         int i, j;
1792
1793         mtx_assert(&lun->lun_lock, MA_OWNED);
1794         for (i = softc->port_min; i < softc->port_max; i++) {
1795                 if (lun->pending_ua[i] == NULL)
1796                         continue;
1797                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1798                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1799                                 continue;
1800                         lun->pending_ua[i][j] &= ~ua;
1801                 }
1802         }
1803 }
1804
1805 void
1806 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1807     ctl_ua_type ua_type)
1808 {
1809         struct ctl_lun *lun;
1810
1811         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1812         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1813                 mtx_lock(&lun->lun_lock);
1814                 ctl_clr_ua(lun, initidx, ua_type);
1815                 mtx_unlock(&lun->lun_lock);
1816         }
1817 }
1818
1819 static int
1820 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1821 {
1822         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1823         struct ctl_lun *lun;
1824         struct ctl_lun_req ireq;
1825         int error, value;
1826
1827         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1828         error = sysctl_handle_int(oidp, &value, 0, req);
1829         if ((error != 0) || (req->newptr == NULL))
1830                 return (error);
1831
1832         mtx_lock(&softc->ctl_lock);
1833         if (value == 0)
1834                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1835         else
1836                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1837         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1838                 mtx_unlock(&softc->ctl_lock);
1839                 bzero(&ireq, sizeof(ireq));
1840                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1841                 ireq.reqdata.modify.lun_id = lun->lun;
1842                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1843                     curthread);
1844                 if (ireq.status != CTL_LUN_OK) {
1845                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1846                             __func__, ireq.status, ireq.error_str);
1847                 }
1848                 mtx_lock(&softc->ctl_lock);
1849         }
1850         mtx_unlock(&softc->ctl_lock);
1851         return (0);
1852 }
1853
1854 static int
1855 ctl_init(void)
1856 {
1857         struct make_dev_args args;
1858         struct ctl_softc *softc;
1859         int i, error;
1860
1861         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1862                                M_WAITOK | M_ZERO);
1863
1864         make_dev_args_init(&args);
1865         args.mda_devsw = &ctl_cdevsw;
1866         args.mda_uid = UID_ROOT;
1867         args.mda_gid = GID_OPERATOR;
1868         args.mda_mode = 0600;
1869         args.mda_si_drv1 = softc;
1870         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1871         if (error != 0) {
1872                 free(softc, M_DEVBUF);
1873                 control_softc = NULL;
1874                 return (error);
1875         }
1876
1877         sysctl_ctx_init(&softc->sysctl_ctx);
1878         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1879                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1880                 CTLFLAG_RD, 0, "CAM Target Layer");
1881
1882         if (softc->sysctl_tree == NULL) {
1883                 printf("%s: unable to allocate sysctl tree\n", __func__);
1884                 destroy_dev(softc->dev);
1885                 free(softc, M_DEVBUF);
1886                 control_softc = NULL;
1887                 return (ENOMEM);
1888         }
1889
1890         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1891         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1892             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1893         softc->flags = 0;
1894
1895         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1896             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1897             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1898
1899         if (ctl_max_luns <= 0 || powerof2(ctl_max_luns) == 0) {
1900                 printf("Bad value %d for kern.cam.ctl.max_luns, must be a power of two, using %d\n",
1901                     ctl_max_luns, CTL_DEFAULT_MAX_LUNS);
1902                 ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
1903         }
1904         softc->ctl_luns = malloc(sizeof(struct ctl_lun *) * ctl_max_luns,
1905             M_DEVBUF, M_WAITOK | M_ZERO);
1906         softc->ctl_lun_mask = malloc(sizeof(uint32_t) *
1907             ((ctl_max_luns + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1908         if (ctl_max_ports <= 0 || powerof2(ctl_max_ports) == 0) {
1909                 printf("Bad value %d for kern.cam.ctl.max_ports, must be a power of two, using %d\n",
1910                     ctl_max_ports, CTL_DEFAULT_MAX_PORTS);
1911                 ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
1912         }
1913         softc->ctl_port_mask = malloc(sizeof(uint32_t) *
1914           ((ctl_max_ports + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1915         softc->ctl_ports = malloc(sizeof(struct ctl_port *) * ctl_max_ports,
1916              M_DEVBUF, M_WAITOK | M_ZERO);
1917
1918
1919         /*
1920          * In Copan's HA scheme, the "master" and "slave" roles are
1921          * figured out through the slot the controller is in.  Although it
1922          * is an active/active system, someone has to be in charge.
1923          */
1924         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1925             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1926             "HA head ID (0 - no HA)");
1927         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1928                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1929                 softc->is_single = 1;
1930                 softc->port_cnt = ctl_max_ports;
1931                 softc->port_min = 0;
1932         } else {
1933                 softc->port_cnt = ctl_max_ports / NUM_HA_SHELVES;
1934                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1935         }
1936         softc->port_max = softc->port_min + softc->port_cnt;
1937         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1938         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1939
1940         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1941             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1942             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1943
1944         STAILQ_INIT(&softc->lun_list);
1945         STAILQ_INIT(&softc->pending_lun_queue);
1946         STAILQ_INIT(&softc->fe_list);
1947         STAILQ_INIT(&softc->port_list);
1948         STAILQ_INIT(&softc->be_list);
1949         ctl_tpc_init(softc);
1950
1951         if (worker_threads <= 0)
1952                 worker_threads = max(1, mp_ncpus / 4);
1953         if (worker_threads > CTL_MAX_THREADS)
1954                 worker_threads = CTL_MAX_THREADS;
1955
1956         for (i = 0; i < worker_threads; i++) {
1957                 struct ctl_thread *thr = &softc->threads[i];
1958
1959                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1960                 thr->ctl_softc = softc;
1961                 STAILQ_INIT(&thr->incoming_queue);
1962                 STAILQ_INIT(&thr->rtr_queue);
1963                 STAILQ_INIT(&thr->done_queue);
1964                 STAILQ_INIT(&thr->isc_queue);
1965
1966                 error = kproc_kthread_add(ctl_work_thread, thr,
1967                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1968                 if (error != 0) {
1969                         printf("error creating CTL work thread!\n");
1970                         return (error);
1971                 }
1972         }
1973         error = kproc_kthread_add(ctl_lun_thread, softc,
1974             &softc->ctl_proc, &softc->lun_thread, 0, 0, "ctl", "lun");
1975         if (error != 0) {
1976                 printf("error creating CTL lun thread!\n");
1977                 return (error);
1978         }
1979         error = kproc_kthread_add(ctl_thresh_thread, softc,
1980             &softc->ctl_proc, &softc->thresh_thread, 0, 0, "ctl", "thresh");
1981         if (error != 0) {
1982                 printf("error creating CTL threshold thread!\n");
1983                 return (error);
1984         }
1985
1986         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1987             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1988             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1989
1990         if (softc->is_single == 0) {
1991                 if (ctl_frontend_register(&ha_frontend) != 0)
1992                         softc->is_single = 1;
1993         }
1994         return (0);
1995 }
1996
1997 static int
1998 ctl_shutdown(void)
1999 {
2000         struct ctl_softc *softc = control_softc;
2001         int i;
2002
2003         if (softc->is_single == 0)
2004                 ctl_frontend_deregister(&ha_frontend);
2005
2006         destroy_dev(softc->dev);
2007
2008         /* Shutdown CTL threads. */
2009         softc->shutdown = 1;
2010         for (i = 0; i < worker_threads; i++) {
2011                 struct ctl_thread *thr = &softc->threads[i];
2012                 while (thr->thread != NULL) {
2013                         wakeup(thr);
2014                         if (thr->thread != NULL)
2015                                 pause("CTL thr shutdown", 1);
2016                 }
2017                 mtx_destroy(&thr->queue_lock);
2018         }
2019         while (softc->lun_thread != NULL) {
2020                 wakeup(&softc->pending_lun_queue);
2021                 if (softc->lun_thread != NULL)
2022                         pause("CTL thr shutdown", 1);
2023         }
2024         while (softc->thresh_thread != NULL) {
2025                 wakeup(softc->thresh_thread);
2026                 if (softc->thresh_thread != NULL)
2027                         pause("CTL thr shutdown", 1);
2028         }
2029
2030         ctl_tpc_shutdown(softc);
2031         uma_zdestroy(softc->io_zone);
2032         mtx_destroy(&softc->ctl_lock);
2033
2034         free(softc->ctl_luns, M_DEVBUF);
2035         free(softc->ctl_lun_mask, M_DEVBUF);
2036         free(softc->ctl_port_mask, M_DEVBUF);
2037         free(softc->ctl_ports, M_DEVBUF);
2038
2039         sysctl_ctx_free(&softc->sysctl_ctx);
2040
2041         free(softc, M_DEVBUF);
2042         control_softc = NULL;
2043         return (0);
2044 }
2045
2046 static int
2047 ctl_module_event_handler(module_t mod, int what, void *arg)
2048 {
2049
2050         switch (what) {
2051         case MOD_LOAD:
2052                 return (ctl_init());
2053         case MOD_UNLOAD:
2054                 return (ctl_shutdown());
2055         default:
2056                 return (EOPNOTSUPP);
2057         }
2058 }
2059
2060 /*
2061  * XXX KDM should we do some access checks here?  Bump a reference count to
2062  * prevent a CTL module from being unloaded while someone has it open?
2063  */
2064 static int
2065 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
2066 {
2067         return (0);
2068 }
2069
2070 static int
2071 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2072 {
2073         return (0);
2074 }
2075
2076 /*
2077  * Remove an initiator by port number and initiator ID.
2078  * Returns 0 for success, -1 for failure.
2079  */
2080 int
2081 ctl_remove_initiator(struct ctl_port *port, int iid)
2082 {
2083         struct ctl_softc *softc = port->ctl_softc;
2084         int last;
2085
2086         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2087
2088         if (iid > CTL_MAX_INIT_PER_PORT) {
2089                 printf("%s: initiator ID %u > maximun %u!\n",
2090                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2091                 return (-1);
2092         }
2093
2094         mtx_lock(&softc->ctl_lock);
2095         last = (--port->wwpn_iid[iid].in_use == 0);
2096         port->wwpn_iid[iid].last_use = time_uptime;
2097         mtx_unlock(&softc->ctl_lock);
2098         if (last)
2099                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
2100         ctl_isc_announce_iid(port, iid);
2101
2102         return (0);
2103 }
2104
2105 /*
2106  * Add an initiator to the initiator map.
2107  * Returns iid for success, < 0 for failure.
2108  */
2109 int
2110 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2111 {
2112         struct ctl_softc *softc = port->ctl_softc;
2113         time_t best_time;
2114         int i, best;
2115
2116         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2117
2118         if (iid >= CTL_MAX_INIT_PER_PORT) {
2119                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2120                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2121                 free(name, M_CTL);
2122                 return (-1);
2123         }
2124
2125         mtx_lock(&softc->ctl_lock);
2126
2127         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2128                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2129                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2130                                 iid = i;
2131                                 break;
2132                         }
2133                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2134                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2135                                 iid = i;
2136                                 break;
2137                         }
2138                 }
2139         }
2140
2141         if (iid < 0) {
2142                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2143                         if (port->wwpn_iid[i].in_use == 0 &&
2144                             port->wwpn_iid[i].wwpn == 0 &&
2145                             port->wwpn_iid[i].name == NULL) {
2146                                 iid = i;
2147                                 break;
2148                         }
2149                 }
2150         }
2151
2152         if (iid < 0) {
2153                 best = -1;
2154                 best_time = INT32_MAX;
2155                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2156                         if (port->wwpn_iid[i].in_use == 0) {
2157                                 if (port->wwpn_iid[i].last_use < best_time) {
2158                                         best = i;
2159                                         best_time = port->wwpn_iid[i].last_use;
2160                                 }
2161                         }
2162                 }
2163                 iid = best;
2164         }
2165
2166         if (iid < 0) {
2167                 mtx_unlock(&softc->ctl_lock);
2168                 free(name, M_CTL);
2169                 return (-2);
2170         }
2171
2172         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2173                 /*
2174                  * This is not an error yet.
2175                  */
2176                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2177 #if 0
2178                         printf("%s: port %d iid %u WWPN %#jx arrived"
2179                             " again\n", __func__, port->targ_port,
2180                             iid, (uintmax_t)wwpn);
2181 #endif
2182                         goto take;
2183                 }
2184                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2185                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2186 #if 0
2187                         printf("%s: port %d iid %u name '%s' arrived"
2188                             " again\n", __func__, port->targ_port,
2189                             iid, name);
2190 #endif
2191                         goto take;
2192                 }
2193
2194                 /*
2195                  * This is an error, but what do we do about it?  The
2196                  * driver is telling us we have a new WWPN for this
2197                  * initiator ID, so we pretty much need to use it.
2198                  */
2199                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2200                     " but WWPN %#jx '%s' is still at that address\n",
2201                     __func__, port->targ_port, iid, wwpn, name,
2202                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2203                     port->wwpn_iid[iid].name);
2204         }
2205 take:
2206         free(port->wwpn_iid[iid].name, M_CTL);
2207         port->wwpn_iid[iid].name = name;
2208         port->wwpn_iid[iid].wwpn = wwpn;
2209         port->wwpn_iid[iid].in_use++;
2210         mtx_unlock(&softc->ctl_lock);
2211         ctl_isc_announce_iid(port, iid);
2212
2213         return (iid);
2214 }
2215
2216 static int
2217 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2218 {
2219         int len;
2220
2221         switch (port->port_type) {
2222         case CTL_PORT_FC:
2223         {
2224                 struct scsi_transportid_fcp *id =
2225                     (struct scsi_transportid_fcp *)buf;
2226                 if (port->wwpn_iid[iid].wwpn == 0)
2227                         return (0);
2228                 memset(id, 0, sizeof(*id));
2229                 id->format_protocol = SCSI_PROTO_FC;
2230                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2231                 return (sizeof(*id));
2232         }
2233         case CTL_PORT_ISCSI:
2234         {
2235                 struct scsi_transportid_iscsi_port *id =
2236                     (struct scsi_transportid_iscsi_port *)buf;
2237                 if (port->wwpn_iid[iid].name == NULL)
2238                         return (0);
2239                 memset(id, 0, 256);
2240                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2241                     SCSI_PROTO_ISCSI;
2242                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2243                 len = roundup2(min(len, 252), 4);
2244                 scsi_ulto2b(len, id->additional_length);
2245                 return (sizeof(*id) + len);
2246         }
2247         case CTL_PORT_SAS:
2248         {
2249                 struct scsi_transportid_sas *id =
2250                     (struct scsi_transportid_sas *)buf;
2251                 if (port->wwpn_iid[iid].wwpn == 0)
2252                         return (0);
2253                 memset(id, 0, sizeof(*id));
2254                 id->format_protocol = SCSI_PROTO_SAS;
2255                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2256                 return (sizeof(*id));
2257         }
2258         default:
2259         {
2260                 struct scsi_transportid_spi *id =
2261                     (struct scsi_transportid_spi *)buf;
2262                 memset(id, 0, sizeof(*id));
2263                 id->format_protocol = SCSI_PROTO_SPI;
2264                 scsi_ulto2b(iid, id->scsi_addr);
2265                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2266                 return (sizeof(*id));
2267         }
2268         }
2269 }
2270
2271 /*
2272  * Serialize a command that went down the "wrong" side, and so was sent to
2273  * this controller for execution.  The logic is a little different than the
2274  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2275  * sent back to the other side, but in the success case, we execute the
2276  * command on this side (XFER mode) or tell the other side to execute it
2277  * (SER_ONLY mode).
2278  */
2279 static void
2280 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2281 {
2282         struct ctl_softc *softc = CTL_SOFTC(ctsio);
2283         struct ctl_port *port = CTL_PORT(ctsio);
2284         union ctl_ha_msg msg_info;
2285         struct ctl_lun *lun;
2286         const struct ctl_cmd_entry *entry;
2287         uint32_t targ_lun;
2288
2289         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2290
2291         /* Make sure that we know about this port. */
2292         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2293                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2294                                          /*retry_count*/ 1);
2295                 goto badjuju;
2296         }
2297
2298         /* Make sure that we know about this LUN. */
2299         mtx_lock(&softc->ctl_lock);
2300         if (targ_lun >= ctl_max_luns ||
2301             (lun = softc->ctl_luns[targ_lun]) == NULL) {
2302                 mtx_unlock(&softc->ctl_lock);
2303
2304                 /*
2305                  * The other node would not send this request to us unless
2306                  * received announce that we are primary node for this LUN.
2307                  * If this LUN does not exist now, it is probably result of
2308                  * a race, so respond to initiator in the most opaque way.
2309                  */
2310                 ctl_set_busy(ctsio);
2311                 goto badjuju;
2312         }
2313         mtx_lock(&lun->lun_lock);
2314         mtx_unlock(&softc->ctl_lock);
2315
2316         /*
2317          * If the LUN is invalid, pretend that it doesn't exist.
2318          * It will go away as soon as all pending I/Os completed.
2319          */
2320         if (lun->flags & CTL_LUN_DISABLED) {
2321                 mtx_unlock(&lun->lun_lock);
2322                 ctl_set_busy(ctsio);
2323                 goto badjuju;
2324         }
2325
2326         entry = ctl_get_cmd_entry(ctsio, NULL);
2327         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2328                 mtx_unlock(&lun->lun_lock);
2329                 goto badjuju;
2330         }
2331
2332         CTL_LUN(ctsio) = lun;
2333         CTL_BACKEND_LUN(ctsio) = lun->be_lun;
2334
2335         /*
2336          * Every I/O goes into the OOA queue for a
2337          * particular LUN, and stays there until completion.
2338          */
2339 #ifdef CTL_TIME_IO
2340         if (TAILQ_EMPTY(&lun->ooa_queue))
2341                 lun->idle_time += getsbinuptime() - lun->last_busy;
2342 #endif
2343         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2344
2345         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2346                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2347                  ooa_links))) {
2348         case CTL_ACTION_BLOCK:
2349                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2350                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2351                                   blocked_links);
2352                 mtx_unlock(&lun->lun_lock);
2353                 break;
2354         case CTL_ACTION_PASS:
2355         case CTL_ACTION_SKIP:
2356                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2357                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2358                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2359                         mtx_unlock(&lun->lun_lock);
2360                 } else {
2361                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2362                         mtx_unlock(&lun->lun_lock);
2363
2364                         /* send msg back to other side */
2365                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2366                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2367                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2368                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2369                             sizeof(msg_info.hdr), M_WAITOK);
2370                 }
2371                 break;
2372         case CTL_ACTION_OVERLAP:
2373                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2374                 mtx_unlock(&lun->lun_lock);
2375                 ctl_set_overlapped_cmd(ctsio);
2376                 goto badjuju;
2377         case CTL_ACTION_OVERLAP_TAG:
2378                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2379                 mtx_unlock(&lun->lun_lock);
2380                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2381                 goto badjuju;
2382         case CTL_ACTION_ERROR:
2383         default:
2384                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2385                 mtx_unlock(&lun->lun_lock);
2386
2387                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2388                                          /*retry_count*/ 0);
2389 badjuju:
2390                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2391                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2392                 msg_info.hdr.serializing_sc = NULL;
2393                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2394                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2395                     sizeof(msg_info.scsi), M_WAITOK);
2396                 ctl_free_io((union ctl_io *)ctsio);
2397                 break;
2398         }
2399 }
2400
2401 /*
2402  * Returns 0 for success, errno for failure.
2403  */
2404 static void
2405 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2406                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2407 {
2408         union ctl_io *io;
2409
2410         mtx_lock(&lun->lun_lock);
2411         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2412              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2413              ooa_links)) {
2414                 struct ctl_ooa_entry *entry;
2415
2416                 /*
2417                  * If we've got more than we can fit, just count the
2418                  * remaining entries.
2419                  */
2420                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2421                         continue;
2422
2423                 entry = &kern_entries[*cur_fill_num];
2424
2425                 entry->tag_num = io->scsiio.tag_num;
2426                 entry->lun_num = lun->lun;
2427 #ifdef CTL_TIME_IO
2428                 entry->start_bt = io->io_hdr.start_bt;
2429 #endif
2430                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2431                 entry->cdb_len = io->scsiio.cdb_len;
2432                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2433                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2434
2435                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2436                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2437
2438                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2439                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2440
2441                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2442                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2443
2444                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2445                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2446         }
2447         mtx_unlock(&lun->lun_lock);
2448 }
2449
2450 static void *
2451 ctl_copyin_alloc(void *user_addr, unsigned int len, char *error_str,
2452                  size_t error_str_len)
2453 {
2454         void *kptr;
2455
2456         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2457
2458         if (copyin(user_addr, kptr, len) != 0) {
2459                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2460                          "from user address %p to kernel address %p", len,
2461                          user_addr, kptr);
2462                 free(kptr, M_CTL);
2463                 return (NULL);
2464         }
2465
2466         return (kptr);
2467 }
2468
2469 static void
2470 ctl_free_args(int num_args, struct ctl_be_arg *args)
2471 {
2472         int i;
2473
2474         if (args == NULL)
2475                 return;
2476
2477         for (i = 0; i < num_args; i++) {
2478                 free(args[i].kname, M_CTL);
2479                 free(args[i].kvalue, M_CTL);
2480         }
2481
2482         free(args, M_CTL);
2483 }
2484
2485 static struct ctl_be_arg *
2486 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2487                 char *error_str, size_t error_str_len)
2488 {
2489         struct ctl_be_arg *args;
2490         int i;
2491
2492         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2493                                 error_str, error_str_len);
2494
2495         if (args == NULL)
2496                 goto bailout;
2497
2498         for (i = 0; i < num_args; i++) {
2499                 args[i].kname = NULL;
2500                 args[i].kvalue = NULL;
2501         }
2502
2503         for (i = 0; i < num_args; i++) {
2504                 uint8_t *tmpptr;
2505
2506                 if (args[i].namelen == 0) {
2507                         snprintf(error_str, error_str_len, "Argument %d "
2508                                  "name length is zero", i);
2509                         goto bailout;
2510                 }
2511
2512                 args[i].kname = ctl_copyin_alloc(args[i].name,
2513                         args[i].namelen, error_str, error_str_len);
2514                 if (args[i].kname == NULL)
2515                         goto bailout;
2516
2517                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2518                         snprintf(error_str, error_str_len, "Argument %d "
2519                                  "name is not NUL-terminated", i);
2520                         goto bailout;
2521                 }
2522
2523                 if (args[i].flags & CTL_BEARG_RD) {
2524                         if (args[i].vallen == 0) {
2525                                 snprintf(error_str, error_str_len, "Argument %d "
2526                                          "value length is zero", i);
2527                                 goto bailout;
2528                         }
2529
2530                         tmpptr = ctl_copyin_alloc(args[i].value,
2531                                 args[i].vallen, error_str, error_str_len);
2532                         if (tmpptr == NULL)
2533                                 goto bailout;
2534
2535                         if ((args[i].flags & CTL_BEARG_ASCII)
2536                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2537                                 snprintf(error_str, error_str_len, "Argument "
2538                                     "%d value is not NUL-terminated", i);
2539                                 free(tmpptr, M_CTL);
2540                                 goto bailout;
2541                         }
2542                         args[i].kvalue = tmpptr;
2543                 } else {
2544                         args[i].kvalue = malloc(args[i].vallen,
2545                             M_CTL, M_WAITOK | M_ZERO);
2546                 }
2547         }
2548
2549         return (args);
2550 bailout:
2551
2552         ctl_free_args(num_args, args);
2553
2554         return (NULL);
2555 }
2556
2557 static void
2558 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2559 {
2560         int i;
2561
2562         for (i = 0; i < num_args; i++) {
2563                 if (args[i].flags & CTL_BEARG_WR)
2564                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2565         }
2566 }
2567
2568 /*
2569  * Escape characters that are illegal or not recommended in XML.
2570  */
2571 int
2572 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2573 {
2574         char *end = str + size;
2575         int retval;
2576
2577         retval = 0;
2578
2579         for (; *str && str < end; str++) {
2580                 switch (*str) {
2581                 case '&':
2582                         retval = sbuf_printf(sb, "&amp;");
2583                         break;
2584                 case '>':
2585                         retval = sbuf_printf(sb, "&gt;");
2586                         break;
2587                 case '<':
2588                         retval = sbuf_printf(sb, "&lt;");
2589                         break;
2590                 default:
2591                         retval = sbuf_putc(sb, *str);
2592                         break;
2593                 }
2594
2595                 if (retval != 0)
2596                         break;
2597
2598         }
2599
2600         return (retval);
2601 }
2602
2603 static void
2604 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2605 {
2606         struct scsi_vpd_id_descriptor *desc;
2607         int i;
2608
2609         if (id == NULL || id->len < 4)
2610                 return;
2611         desc = (struct scsi_vpd_id_descriptor *)id->data;
2612         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2613         case SVPD_ID_TYPE_T10:
2614                 sbuf_printf(sb, "t10.");
2615                 break;
2616         case SVPD_ID_TYPE_EUI64:
2617                 sbuf_printf(sb, "eui.");
2618                 break;
2619         case SVPD_ID_TYPE_NAA:
2620                 sbuf_printf(sb, "naa.");
2621                 break;
2622         case SVPD_ID_TYPE_SCSI_NAME:
2623                 break;
2624         }
2625         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2626         case SVPD_ID_CODESET_BINARY:
2627                 for (i = 0; i < desc->length; i++)
2628                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2629                 break;
2630         case SVPD_ID_CODESET_ASCII:
2631                 sbuf_printf(sb, "%.*s", (int)desc->length,
2632                     (char *)desc->identifier);
2633                 break;
2634         case SVPD_ID_CODESET_UTF8:
2635                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2636                 break;
2637         }
2638 }
2639
2640 static int
2641 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2642           struct thread *td)
2643 {
2644         struct ctl_softc *softc = dev->si_drv1;
2645         struct ctl_port *port;
2646         struct ctl_lun *lun;
2647         int retval;
2648
2649         retval = 0;
2650
2651         switch (cmd) {
2652         case CTL_IO:
2653                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2654                 break;
2655         case CTL_ENABLE_PORT:
2656         case CTL_DISABLE_PORT:
2657         case CTL_SET_PORT_WWNS: {
2658                 struct ctl_port *port;
2659                 struct ctl_port_entry *entry;
2660
2661                 entry = (struct ctl_port_entry *)addr;
2662
2663                 mtx_lock(&softc->ctl_lock);
2664                 STAILQ_FOREACH(port, &softc->port_list, links) {
2665                         int action, done;
2666
2667                         if (port->targ_port < softc->port_min ||
2668                             port->targ_port >= softc->port_max)
2669                                 continue;
2670
2671                         action = 0;
2672                         done = 0;
2673                         if ((entry->port_type == CTL_PORT_NONE)
2674                          && (entry->targ_port == port->targ_port)) {
2675                                 /*
2676                                  * If the user only wants to enable or
2677                                  * disable or set WWNs on a specific port,
2678                                  * do the operation and we're done.
2679                                  */
2680                                 action = 1;
2681                                 done = 1;
2682                         } else if (entry->port_type & port->port_type) {
2683                                 /*
2684                                  * Compare the user's type mask with the
2685                                  * particular frontend type to see if we
2686                                  * have a match.
2687                                  */
2688                                 action = 1;
2689                                 done = 0;
2690
2691                                 /*
2692                                  * Make sure the user isn't trying to set
2693                                  * WWNs on multiple ports at the same time.
2694                                  */
2695                                 if (cmd == CTL_SET_PORT_WWNS) {
2696                                         printf("%s: Can't set WWNs on "
2697                                                "multiple ports\n", __func__);
2698                                         retval = EINVAL;
2699                                         break;
2700                                 }
2701                         }
2702                         if (action == 0)
2703                                 continue;
2704
2705                         /*
2706                          * XXX KDM we have to drop the lock here, because
2707                          * the online/offline operations can potentially
2708                          * block.  We need to reference count the frontends
2709                          * so they can't go away,
2710                          */
2711                         if (cmd == CTL_ENABLE_PORT) {
2712                                 mtx_unlock(&softc->ctl_lock);
2713                                 ctl_port_online(port);
2714                                 mtx_lock(&softc->ctl_lock);
2715                         } else if (cmd == CTL_DISABLE_PORT) {
2716                                 mtx_unlock(&softc->ctl_lock);
2717                                 ctl_port_offline(port);
2718                                 mtx_lock(&softc->ctl_lock);
2719                         } else if (cmd == CTL_SET_PORT_WWNS) {
2720                                 ctl_port_set_wwns(port,
2721                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2722                                     1 : 0, entry->wwnn,
2723                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2724                                     1 : 0, entry->wwpn);
2725                         }
2726                         if (done != 0)
2727                                 break;
2728                 }
2729                 mtx_unlock(&softc->ctl_lock);
2730                 break;
2731         }
2732         case CTL_GET_OOA: {
2733                 struct ctl_ooa *ooa_hdr;
2734                 struct ctl_ooa_entry *entries;
2735                 uint32_t cur_fill_num;
2736
2737                 ooa_hdr = (struct ctl_ooa *)addr;
2738
2739                 if ((ooa_hdr->alloc_len == 0)
2740                  || (ooa_hdr->alloc_num == 0)) {
2741                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2742                                "must be non-zero\n", __func__,
2743                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2744                         retval = EINVAL;
2745                         break;
2746                 }
2747
2748                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2749                     sizeof(struct ctl_ooa_entry))) {
2750                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2751                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2752                                __func__, ooa_hdr->alloc_len,
2753                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2754                         retval = EINVAL;
2755                         break;
2756                 }
2757
2758                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2759                 if (entries == NULL) {
2760                         printf("%s: could not allocate %d bytes for OOA "
2761                                "dump\n", __func__, ooa_hdr->alloc_len);
2762                         retval = ENOMEM;
2763                         break;
2764                 }
2765
2766                 mtx_lock(&softc->ctl_lock);
2767                 if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 &&
2768                     (ooa_hdr->lun_num >= ctl_max_luns ||
2769                      softc->ctl_luns[ooa_hdr->lun_num] == NULL)) {
2770                         mtx_unlock(&softc->ctl_lock);
2771                         free(entries, M_CTL);
2772                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2773                                __func__, (uintmax_t)ooa_hdr->lun_num);
2774                         retval = EINVAL;
2775                         break;
2776                 }
2777
2778                 cur_fill_num = 0;
2779
2780                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2781                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2782                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2783                                     ooa_hdr, entries);
2784                         }
2785                 } else {
2786                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2787                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2788                             entries);
2789                 }
2790                 mtx_unlock(&softc->ctl_lock);
2791
2792                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2793                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2794                         sizeof(struct ctl_ooa_entry);
2795                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2796                 if (retval != 0) {
2797                         printf("%s: error copying out %d bytes for OOA dump\n",
2798                                __func__, ooa_hdr->fill_len);
2799                 }
2800
2801                 getbinuptime(&ooa_hdr->cur_bt);
2802
2803                 if (cur_fill_num > ooa_hdr->alloc_num) {
2804                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2805                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2806                 } else {
2807                         ooa_hdr->dropped_num = 0;
2808                         ooa_hdr->status = CTL_OOA_OK;
2809                 }
2810
2811                 free(entries, M_CTL);
2812                 break;
2813         }
2814         case CTL_DELAY_IO: {
2815                 struct ctl_io_delay_info *delay_info;
2816
2817                 delay_info = (struct ctl_io_delay_info *)addr;
2818
2819 #ifdef CTL_IO_DELAY
2820                 mtx_lock(&softc->ctl_lock);
2821                 if (delay_info->lun_id >= ctl_max_luns ||
2822                     (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) {
2823                         mtx_unlock(&softc->ctl_lock);
2824                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2825                         break;
2826                 }
2827                 mtx_lock(&lun->lun_lock);
2828                 mtx_unlock(&softc->ctl_lock);
2829                 delay_info->status = CTL_DELAY_STATUS_OK;
2830                 switch (delay_info->delay_type) {
2831                 case CTL_DELAY_TYPE_CONT:
2832                 case CTL_DELAY_TYPE_ONESHOT:
2833                         break;
2834                 default:
2835                         delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE;
2836                         break;
2837                 }
2838                 switch (delay_info->delay_loc) {
2839                 case CTL_DELAY_LOC_DATAMOVE:
2840                         lun->delay_info.datamove_type = delay_info->delay_type;
2841                         lun->delay_info.datamove_delay = delay_info->delay_secs;
2842                         break;
2843                 case CTL_DELAY_LOC_DONE:
2844                         lun->delay_info.done_type = delay_info->delay_type;
2845                         lun->delay_info.done_delay = delay_info->delay_secs;
2846                         break;
2847                 default:
2848                         delay_info->status = CTL_DELAY_STATUS_INVALID_LOC;
2849                         break;
2850                 }
2851                 mtx_unlock(&lun->lun_lock);
2852 #else
2853                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2854 #endif /* CTL_IO_DELAY */
2855                 break;
2856         }
2857 #ifdef CTL_LEGACY_STATS
2858         case CTL_GETSTATS: {
2859                 struct ctl_stats *stats = (struct ctl_stats *)addr;
2860                 int i;
2861
2862                 /*
2863                  * XXX KDM no locking here.  If the LUN list changes,
2864                  * things can blow up.
2865                  */
2866                 i = 0;
2867                 stats->status = CTL_SS_OK;
2868                 stats->fill_len = 0;
2869                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2870                         if (stats->fill_len + sizeof(lun->legacy_stats) >
2871                             stats->alloc_len) {
2872                                 stats->status = CTL_SS_NEED_MORE_SPACE;
2873                                 break;
2874                         }
2875                         retval = copyout(&lun->legacy_stats, &stats->lun_stats[i++],
2876                                          sizeof(lun->legacy_stats));
2877                         if (retval != 0)
2878                                 break;
2879                         stats->fill_len += sizeof(lun->legacy_stats);
2880                 }
2881                 stats->num_luns = softc->num_luns;
2882                 stats->flags = CTL_STATS_FLAG_NONE;
2883 #ifdef CTL_TIME_IO
2884                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
2885 #endif
2886                 getnanouptime(&stats->timestamp);
2887                 break;
2888         }
2889 #endif /* CTL_LEGACY_STATS */
2890         case CTL_ERROR_INJECT: {
2891                 struct ctl_error_desc *err_desc, *new_err_desc;
2892
2893                 err_desc = (struct ctl_error_desc *)addr;
2894
2895                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2896                                       M_WAITOK | M_ZERO);
2897                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2898
2899                 mtx_lock(&softc->ctl_lock);
2900                 if (err_desc->lun_id >= ctl_max_luns ||
2901                     (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) {
2902                         mtx_unlock(&softc->ctl_lock);
2903                         free(new_err_desc, M_CTL);
2904                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2905                                __func__, (uintmax_t)err_desc->lun_id);
2906                         retval = EINVAL;
2907                         break;
2908                 }
2909                 mtx_lock(&lun->lun_lock);
2910                 mtx_unlock(&softc->ctl_lock);
2911
2912                 /*
2913                  * We could do some checking here to verify the validity
2914                  * of the request, but given the complexity of error
2915                  * injection requests, the checking logic would be fairly
2916                  * complex.
2917                  *
2918                  * For now, if the request is invalid, it just won't get
2919                  * executed and might get deleted.
2920                  */
2921                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2922
2923                 /*
2924                  * XXX KDM check to make sure the serial number is unique,
2925                  * in case we somehow manage to wrap.  That shouldn't
2926                  * happen for a very long time, but it's the right thing to
2927                  * do.
2928                  */
2929                 new_err_desc->serial = lun->error_serial;
2930                 err_desc->serial = lun->error_serial;
2931                 lun->error_serial++;
2932
2933                 mtx_unlock(&lun->lun_lock);
2934                 break;
2935         }
2936         case CTL_ERROR_INJECT_DELETE: {
2937                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2938                 int delete_done;
2939
2940                 delete_desc = (struct ctl_error_desc *)addr;
2941                 delete_done = 0;
2942
2943                 mtx_lock(&softc->ctl_lock);
2944                 if (delete_desc->lun_id >= ctl_max_luns ||
2945                     (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) {
2946                         mtx_unlock(&softc->ctl_lock);
2947                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2948                                __func__, (uintmax_t)delete_desc->lun_id);
2949                         retval = EINVAL;
2950                         break;
2951                 }
2952                 mtx_lock(&lun->lun_lock);
2953                 mtx_unlock(&softc->ctl_lock);
2954                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2955                         if (desc->serial != delete_desc->serial)
2956                                 continue;
2957
2958                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2959                                       links);
2960                         free(desc, M_CTL);
2961                         delete_done = 1;
2962                 }
2963                 mtx_unlock(&lun->lun_lock);
2964                 if (delete_done == 0) {
2965                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2966                                "error serial %ju on LUN %u\n", __func__,
2967                                delete_desc->serial, delete_desc->lun_id);
2968                         retval = EINVAL;
2969                         break;
2970                 }
2971                 break;
2972         }
2973         case CTL_DUMP_STRUCTS: {
2974                 int j, k;
2975                 struct ctl_port *port;
2976                 struct ctl_frontend *fe;
2977
2978                 mtx_lock(&softc->ctl_lock);
2979                 printf("CTL Persistent Reservation information start:\n");
2980                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2981                         mtx_lock(&lun->lun_lock);
2982                         if ((lun->flags & CTL_LUN_DISABLED) != 0) {
2983                                 mtx_unlock(&lun->lun_lock);
2984                                 continue;
2985                         }
2986
2987                         for (j = 0; j < ctl_max_ports; j++) {
2988                                 if (lun->pr_keys[j] == NULL)
2989                                         continue;
2990                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2991                                         if (lun->pr_keys[j][k] == 0)
2992                                                 continue;
2993                                         printf("  LUN %ju port %d iid %d key "
2994                                                "%#jx\n", lun->lun, j, k,
2995                                                (uintmax_t)lun->pr_keys[j][k]);
2996                                 }
2997                         }
2998                         mtx_unlock(&lun->lun_lock);
2999                 }
3000                 printf("CTL Persistent Reservation information end\n");
3001                 printf("CTL Ports:\n");
3002                 STAILQ_FOREACH(port, &softc->port_list, links) {
3003                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3004                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3005                                port->frontend->name, port->port_type,
3006                                port->physical_port, port->virtual_port,
3007                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3008                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3009                                 if (port->wwpn_iid[j].in_use == 0 &&
3010                                     port->wwpn_iid[j].wwpn == 0 &&
3011                                     port->wwpn_iid[j].name == NULL)
3012                                         continue;
3013
3014                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
3015                                     j, port->wwpn_iid[j].in_use,
3016                                     (uintmax_t)port->wwpn_iid[j].wwpn,
3017                                     port->wwpn_iid[j].name);
3018                         }
3019                 }
3020                 printf("CTL Port information end\n");
3021                 mtx_unlock(&softc->ctl_lock);
3022                 /*
3023                  * XXX KDM calling this without a lock.  We'd likely want
3024                  * to drop the lock before calling the frontend's dump
3025                  * routine anyway.
3026                  */
3027                 printf("CTL Frontends:\n");
3028                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
3029                         printf("  Frontend '%s'\n", fe->name);
3030                         if (fe->fe_dump != NULL)
3031                                 fe->fe_dump();
3032                 }
3033                 printf("CTL Frontend information end\n");
3034                 break;
3035         }
3036         case CTL_LUN_REQ: {
3037                 struct ctl_lun_req *lun_req;
3038                 struct ctl_backend_driver *backend;
3039
3040                 lun_req = (struct ctl_lun_req *)addr;
3041
3042                 backend = ctl_backend_find(lun_req->backend);
3043                 if (backend == NULL) {
3044                         lun_req->status = CTL_LUN_ERROR;
3045                         snprintf(lun_req->error_str,
3046                                  sizeof(lun_req->error_str),
3047                                  "Backend \"%s\" not found.",
3048                                  lun_req->backend);
3049                         break;
3050                 }
3051                 if (lun_req->num_be_args > 0) {
3052                         lun_req->kern_be_args = ctl_copyin_args(
3053                                 lun_req->num_be_args,
3054                                 lun_req->be_args,
3055                                 lun_req->error_str,
3056                                 sizeof(lun_req->error_str));
3057                         if (lun_req->kern_be_args == NULL) {
3058                                 lun_req->status = CTL_LUN_ERROR;
3059                                 break;
3060                         }
3061                 }
3062
3063                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3064
3065                 if (lun_req->num_be_args > 0) {
3066                         ctl_copyout_args(lun_req->num_be_args,
3067                                       lun_req->kern_be_args);
3068                         ctl_free_args(lun_req->num_be_args,
3069                                       lun_req->kern_be_args);
3070                 }
3071                 break;
3072         }
3073         case CTL_LUN_LIST: {
3074                 struct sbuf *sb;
3075                 struct ctl_lun_list *list;
3076                 struct ctl_option *opt;
3077
3078                 list = (struct ctl_lun_list *)addr;
3079
3080                 /*
3081                  * Allocate a fixed length sbuf here, based on the length
3082                  * of the user's buffer.  We could allocate an auto-extending
3083                  * buffer, and then tell the user how much larger our
3084                  * amount of data is than his buffer, but that presents
3085                  * some problems:
3086                  *
3087                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3088                  *     we can't hold a lock while calling them with an
3089                  *     auto-extending buffer.
3090                  *
3091                  * 2.  There is not currently a LUN reference counting
3092                  *     mechanism, outside of outstanding transactions on
3093                  *     the LUN's OOA queue.  So a LUN could go away on us
3094                  *     while we're getting the LUN number, backend-specific
3095                  *     information, etc.  Thus, given the way things
3096                  *     currently work, we need to hold the CTL lock while
3097                  *     grabbing LUN information.
3098                  *
3099                  * So, from the user's standpoint, the best thing to do is
3100                  * allocate what he thinks is a reasonable buffer length,
3101                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3102                  * double the buffer length and try again.  (And repeat
3103                  * that until he succeeds.)
3104                  */
3105                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3106                 if (sb == NULL) {
3107                         list->status = CTL_LUN_LIST_ERROR;
3108                         snprintf(list->error_str, sizeof(list->error_str),
3109                                  "Unable to allocate %d bytes for LUN list",
3110                                  list->alloc_len);
3111                         break;
3112                 }
3113
3114                 sbuf_printf(sb, "<ctllunlist>\n");
3115
3116                 mtx_lock(&softc->ctl_lock);
3117                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3118                         mtx_lock(&lun->lun_lock);
3119                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3120                                              (uintmax_t)lun->lun);
3121
3122                         /*
3123                          * Bail out as soon as we see that we've overfilled
3124                          * the buffer.
3125                          */
3126                         if (retval != 0)
3127                                 break;
3128
3129                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3130                                              "</backend_type>\n",
3131                                              (lun->backend == NULL) ?  "none" :
3132                                              lun->backend->name);
3133
3134                         if (retval != 0)
3135                                 break;
3136
3137                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3138                                              lun->be_lun->lun_type);
3139
3140                         if (retval != 0)
3141                                 break;
3142
3143                         if (lun->backend == NULL) {
3144                                 retval = sbuf_printf(sb, "</lun>\n");
3145                                 if (retval != 0)
3146                                         break;
3147                                 continue;
3148                         }
3149
3150                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3151                                              (lun->be_lun->maxlba > 0) ?
3152                                              lun->be_lun->maxlba + 1 : 0);
3153
3154                         if (retval != 0)
3155                                 break;
3156
3157                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3158                                              lun->be_lun->blocksize);
3159
3160                         if (retval != 0)
3161                                 break;
3162
3163                         retval = sbuf_printf(sb, "\t<serial_number>");
3164
3165                         if (retval != 0)
3166                                 break;
3167
3168                         retval = ctl_sbuf_printf_esc(sb,
3169                             lun->be_lun->serial_num,
3170                             sizeof(lun->be_lun->serial_num));
3171
3172                         if (retval != 0)
3173                                 break;
3174
3175                         retval = sbuf_printf(sb, "</serial_number>\n");
3176
3177                         if (retval != 0)
3178                                 break;
3179
3180                         retval = sbuf_printf(sb, "\t<device_id>");
3181
3182                         if (retval != 0)
3183                                 break;
3184
3185                         retval = ctl_sbuf_printf_esc(sb,
3186                             lun->be_lun->device_id,
3187                             sizeof(lun->be_lun->device_id));
3188
3189                         if (retval != 0)
3190                                 break;
3191
3192                         retval = sbuf_printf(sb, "</device_id>\n");
3193
3194                         if (retval != 0)
3195                                 break;
3196
3197                         if (lun->backend->lun_info != NULL) {
3198                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3199                                 if (retval != 0)
3200                                         break;
3201                         }
3202                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3203                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3204                                     opt->name, opt->value, opt->name);
3205                                 if (retval != 0)
3206                                         break;
3207                         }
3208
3209                         retval = sbuf_printf(sb, "</lun>\n");
3210
3211                         if (retval != 0)
3212                                 break;
3213                         mtx_unlock(&lun->lun_lock);
3214                 }
3215                 if (lun != NULL)
3216                         mtx_unlock(&lun->lun_lock);
3217                 mtx_unlock(&softc->ctl_lock);
3218
3219                 if ((retval != 0)
3220                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3221                         retval = 0;
3222                         sbuf_delete(sb);
3223                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3224                         snprintf(list->error_str, sizeof(list->error_str),
3225                                  "Out of space, %d bytes is too small",
3226                                  list->alloc_len);
3227                         break;
3228                 }
3229
3230                 sbuf_finish(sb);
3231
3232                 retval = copyout(sbuf_data(sb), list->lun_xml,
3233                                  sbuf_len(sb) + 1);
3234
3235                 list->fill_len = sbuf_len(sb) + 1;
3236                 list->status = CTL_LUN_LIST_OK;
3237                 sbuf_delete(sb);
3238                 break;
3239         }
3240         case CTL_ISCSI: {
3241                 struct ctl_iscsi *ci;
3242                 struct ctl_frontend *fe;
3243
3244                 ci = (struct ctl_iscsi *)addr;
3245
3246                 fe = ctl_frontend_find("iscsi");
3247                 if (fe == NULL) {
3248                         ci->status = CTL_ISCSI_ERROR;
3249                         snprintf(ci->error_str, sizeof(ci->error_str),
3250                             "Frontend \"iscsi\" not found.");
3251                         break;
3252                 }
3253
3254                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3255                 break;
3256         }
3257         case CTL_PORT_REQ: {
3258                 struct ctl_req *req;
3259                 struct ctl_frontend *fe;
3260
3261                 req = (struct ctl_req *)addr;
3262
3263                 fe = ctl_frontend_find(req->driver);
3264                 if (fe == NULL) {
3265                         req->status = CTL_LUN_ERROR;
3266                         snprintf(req->error_str, sizeof(req->error_str),
3267                             "Frontend \"%s\" not found.", req->driver);
3268                         break;
3269                 }
3270                 if (req->num_args > 0) {
3271                         req->kern_args = ctl_copyin_args(req->num_args,
3272                             req->args, req->error_str, sizeof(req->error_str));
3273                         if (req->kern_args == NULL) {
3274                                 req->status = CTL_LUN_ERROR;
3275                                 break;
3276                         }
3277                 }
3278
3279                 if (fe->ioctl)
3280                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3281                 else
3282                         retval = ENODEV;
3283
3284                 if (req->num_args > 0) {
3285                         ctl_copyout_args(req->num_args, req->kern_args);
3286                         ctl_free_args(req->num_args, req->kern_args);
3287                 }
3288                 break;
3289         }
3290         case CTL_PORT_LIST: {
3291                 struct sbuf *sb;
3292                 struct ctl_port *port;
3293                 struct ctl_lun_list *list;
3294                 struct ctl_option *opt;
3295                 int j;
3296                 uint32_t plun;
3297
3298                 list = (struct ctl_lun_list *)addr;
3299
3300                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3301                 if (sb == NULL) {
3302                         list->status = CTL_LUN_LIST_ERROR;
3303                         snprintf(list->error_str, sizeof(list->error_str),
3304                                  "Unable to allocate %d bytes for LUN list",
3305                                  list->alloc_len);
3306                         break;
3307                 }
3308
3309                 sbuf_printf(sb, "<ctlportlist>\n");
3310
3311                 mtx_lock(&softc->ctl_lock);
3312                 STAILQ_FOREACH(port, &softc->port_list, links) {
3313                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3314                                              (uintmax_t)port->targ_port);
3315
3316                         /*
3317                          * Bail out as soon as we see that we've overfilled
3318                          * the buffer.
3319                          */
3320                         if (retval != 0)
3321                                 break;
3322
3323                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3324                             "</frontend_type>\n", port->frontend->name);
3325                         if (retval != 0)
3326                                 break;
3327
3328                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3329                                              port->port_type);
3330                         if (retval != 0)
3331                                 break;
3332
3333                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3334                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3335                         if (retval != 0)
3336                                 break;
3337
3338                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3339                             port->port_name);
3340                         if (retval != 0)
3341                                 break;
3342
3343                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3344                             port->physical_port);
3345                         if (retval != 0)
3346                                 break;
3347
3348                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3349                             port->virtual_port);
3350                         if (retval != 0)
3351                                 break;
3352
3353                         if (port->target_devid != NULL) {
3354                                 sbuf_printf(sb, "\t<target>");
3355                                 ctl_id_sbuf(port->target_devid, sb);
3356                                 sbuf_printf(sb, "</target>\n");
3357                         }
3358
3359                         if (port->port_devid != NULL) {
3360                                 sbuf_printf(sb, "\t<port>");
3361                                 ctl_id_sbuf(port->port_devid, sb);
3362                                 sbuf_printf(sb, "</port>\n");
3363                         }
3364
3365                         if (port->port_info != NULL) {
3366                                 retval = port->port_info(port->onoff_arg, sb);
3367                                 if (retval != 0)
3368                                         break;
3369                         }
3370                         STAILQ_FOREACH(opt, &port->options, links) {
3371                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3372                                     opt->name, opt->value, opt->name);
3373                                 if (retval != 0)
3374                                         break;
3375                         }
3376
3377                         if (port->lun_map != NULL) {
3378                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3379                                 for (j = 0; j < port->lun_map_size; j++) {
3380                                         plun = ctl_lun_map_from_port(port, j);
3381                                         if (plun == UINT32_MAX)
3382                                                 continue;
3383                                         sbuf_printf(sb,
3384                                             "\t<lun id=\"%u\">%u</lun>\n",
3385                                             j, plun);
3386                                 }
3387                         }
3388
3389                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3390                                 if (port->wwpn_iid[j].in_use == 0 ||
3391                                     (port->wwpn_iid[j].wwpn == 0 &&
3392                                      port->wwpn_iid[j].name == NULL))
3393                                         continue;
3394
3395                                 if (port->wwpn_iid[j].name != NULL)
3396                                         retval = sbuf_printf(sb,
3397                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3398                                             j, port->wwpn_iid[j].name);
3399                                 else
3400                                         retval = sbuf_printf(sb,
3401                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3402                                             j, port->wwpn_iid[j].wwpn);
3403                                 if (retval != 0)
3404                                         break;
3405                         }
3406                         if (retval != 0)
3407                                 break;
3408
3409                         retval = sbuf_printf(sb, "</targ_port>\n");
3410                         if (retval != 0)
3411                                 break;
3412                 }
3413                 mtx_unlock(&softc->ctl_lock);
3414
3415                 if ((retval != 0)
3416                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3417                         retval = 0;
3418                         sbuf_delete(sb);
3419                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3420                         snprintf(list->error_str, sizeof(list->error_str),
3421                                  "Out of space, %d bytes is too small",
3422                                  list->alloc_len);
3423                         break;
3424                 }
3425
3426                 sbuf_finish(sb);
3427
3428                 retval = copyout(sbuf_data(sb), list->lun_xml,
3429                                  sbuf_len(sb) + 1);
3430
3431                 list->fill_len = sbuf_len(sb) + 1;
3432                 list->status = CTL_LUN_LIST_OK;
3433                 sbuf_delete(sb);
3434                 break;
3435         }
3436         case CTL_LUN_MAP: {
3437                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3438                 struct ctl_port *port;
3439
3440                 mtx_lock(&softc->ctl_lock);
3441                 if (lm->port < softc->port_min ||
3442                     lm->port >= softc->port_max ||
3443                     (port = softc->ctl_ports[lm->port]) == NULL) {
3444                         mtx_unlock(&softc->ctl_lock);
3445                         return (ENXIO);
3446                 }
3447                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3448                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3449                                 if (ctl_lun_map_to_port(port, lun->lun) ==
3450                                     UINT32_MAX)
3451                                         continue;
3452                                 mtx_lock(&lun->lun_lock);
3453                                 ctl_est_ua_port(lun, lm->port, -1,
3454                                     CTL_UA_LUN_CHANGE);
3455                                 mtx_unlock(&lun->lun_lock);
3456                         }
3457                 }
3458                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3459                 if (lm->plun != UINT32_MAX) {
3460                         if (lm->lun == UINT32_MAX)
3461                                 retval = ctl_lun_map_unset(port, lm->plun);
3462                         else if (lm->lun < ctl_max_luns &&
3463                             softc->ctl_luns[lm->lun] != NULL)
3464                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3465                         else
3466                                 return (ENXIO);
3467                 } else {
3468                         if (lm->lun == UINT32_MAX)
3469                                 retval = ctl_lun_map_deinit(port);
3470                         else
3471                                 retval = ctl_lun_map_init(port);
3472                 }
3473                 if (port->status & CTL_PORT_STATUS_ONLINE)
3474                         ctl_isc_announce_port(port);
3475                 break;
3476         }
3477         case CTL_GET_LUN_STATS: {
3478                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3479                 int i;
3480
3481                 /*
3482                  * XXX KDM no locking here.  If the LUN list changes,
3483                  * things can blow up.
3484                  */
3485                 i = 0;
3486                 stats->status = CTL_SS_OK;
3487                 stats->fill_len = 0;
3488                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3489                         if (lun->lun < stats->first_item)
3490                                 continue;
3491                         if (stats->fill_len + sizeof(lun->stats) >
3492                             stats->alloc_len) {
3493                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3494                                 break;
3495                         }
3496                         retval = copyout(&lun->stats, &stats->stats[i++],
3497                                          sizeof(lun->stats));
3498                         if (retval != 0)
3499                                 break;
3500                         stats->fill_len += sizeof(lun->stats);
3501                 }
3502                 stats->num_items = softc->num_luns;
3503                 stats->flags = CTL_STATS_FLAG_NONE;
3504 #ifdef CTL_TIME_IO
3505                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3506 #endif
3507                 getnanouptime(&stats->timestamp);
3508                 break;
3509         }
3510         case CTL_GET_PORT_STATS: {
3511                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3512                 int i;
3513
3514                 /*
3515                  * XXX KDM no locking here.  If the LUN list changes,
3516                  * things can blow up.
3517                  */
3518                 i = 0;
3519                 stats->status = CTL_SS_OK;
3520                 stats->fill_len = 0;
3521                 STAILQ_FOREACH(port, &softc->port_list, links) {
3522                         if (port->targ_port < stats->first_item)
3523                                 continue;
3524                         if (stats->fill_len + sizeof(port->stats) >
3525                             stats->alloc_len) {
3526                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3527                                 break;
3528                         }
3529                         retval = copyout(&port->stats, &stats->stats[i++],
3530                                          sizeof(port->stats));
3531                         if (retval != 0)
3532                                 break;
3533                         stats->fill_len += sizeof(port->stats);
3534                 }
3535                 stats->num_items = softc->num_ports;
3536                 stats->flags = CTL_STATS_FLAG_NONE;
3537 #ifdef CTL_TIME_IO
3538                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3539 #endif
3540                 getnanouptime(&stats->timestamp);
3541                 break;
3542         }
3543         default: {
3544                 /* XXX KDM should we fix this? */
3545 #if 0
3546                 struct ctl_backend_driver *backend;
3547                 unsigned int type;
3548                 int found;
3549
3550                 found = 0;
3551
3552                 /*
3553                  * We encode the backend type as the ioctl type for backend
3554                  * ioctls.  So parse it out here, and then search for a
3555                  * backend of this type.
3556                  */
3557                 type = _IOC_TYPE(cmd);
3558
3559                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3560                         if (backend->type == type) {
3561                                 found = 1;
3562                                 break;
3563                         }
3564                 }
3565                 if (found == 0) {
3566                         printf("ctl: unknown ioctl command %#lx or backend "
3567                                "%d\n", cmd, type);
3568                         retval = EINVAL;
3569                         break;
3570                 }
3571                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3572 #endif
3573                 retval = ENOTTY;
3574                 break;
3575         }
3576         }
3577         return (retval);
3578 }
3579
3580 uint32_t
3581 ctl_get_initindex(struct ctl_nexus *nexus)
3582 {
3583         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3584 }
3585
3586 int
3587 ctl_lun_map_init(struct ctl_port *port)
3588 {
3589         struct ctl_softc *softc = port->ctl_softc;
3590         struct ctl_lun *lun;
3591         int size = ctl_lun_map_size;
3592         uint32_t i;
3593
3594         if (port->lun_map == NULL || port->lun_map_size < size) {
3595                 port->lun_map_size = 0;
3596                 free(port->lun_map, M_CTL);
3597                 port->lun_map = malloc(size * sizeof(uint32_t),
3598                     M_CTL, M_NOWAIT);
3599         }
3600         if (port->lun_map == NULL)
3601                 return (ENOMEM);
3602         for (i = 0; i < size; i++)
3603                 port->lun_map[i] = UINT32_MAX;
3604         port->lun_map_size = size;
3605         if (port->status & CTL_PORT_STATUS_ONLINE) {
3606                 if (port->lun_disable != NULL) {
3607                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3608                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3609                 }
3610                 ctl_isc_announce_port(port);
3611         }
3612         return (0);
3613 }
3614
3615 int
3616 ctl_lun_map_deinit(struct ctl_port *port)
3617 {
3618         struct ctl_softc *softc = port->ctl_softc;
3619         struct ctl_lun *lun;
3620
3621         if (port->lun_map == NULL)
3622                 return (0);
3623         port->lun_map_size = 0;
3624         free(port->lun_map, M_CTL);
3625         port->lun_map = NULL;
3626         if (port->status & CTL_PORT_STATUS_ONLINE) {
3627                 if (port->lun_enable != NULL) {
3628                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3629                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3630                 }
3631                 ctl_isc_announce_port(port);
3632         }
3633         return (0);
3634 }
3635
3636 int
3637 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3638 {
3639         int status;
3640         uint32_t old;
3641
3642         if (port->lun_map == NULL) {
3643                 status = ctl_lun_map_init(port);
3644                 if (status != 0)
3645                         return (status);
3646         }
3647         if (plun >= port->lun_map_size)
3648                 return (EINVAL);
3649         old = port->lun_map[plun];
3650         port->lun_map[plun] = glun;
3651         if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) {
3652                 if (port->lun_enable != NULL)
3653                         port->lun_enable(port->targ_lun_arg, plun);
3654                 ctl_isc_announce_port(port);
3655         }
3656         return (0);
3657 }
3658
3659 int
3660 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3661 {
3662         uint32_t old;
3663
3664         if (port->lun_map == NULL || plun >= port->lun_map_size)
3665                 return (0);
3666         old = port->lun_map[plun];
3667         port->lun_map[plun] = UINT32_MAX;
3668         if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) {
3669                 if (port->lun_disable != NULL)
3670                         port->lun_disable(port->targ_lun_arg, plun);
3671                 ctl_isc_announce_port(port);
3672         }
3673         return (0);
3674 }
3675
3676 uint32_t
3677 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3678 {
3679
3680         if (port == NULL)
3681                 return (UINT32_MAX);
3682         if (port->lun_map == NULL)
3683                 return (lun_id);
3684         if (lun_id > port->lun_map_size)
3685                 return (UINT32_MAX);
3686         return (port->lun_map[lun_id]);
3687 }
3688
3689 uint32_t
3690 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3691 {
3692         uint32_t i;
3693
3694         if (port == NULL)
3695                 return (UINT32_MAX);
3696         if (port->lun_map == NULL)
3697                 return (lun_id);
3698         for (i = 0; i < port->lun_map_size; i++) {
3699                 if (port->lun_map[i] == lun_id)
3700                         return (i);
3701         }
3702         return (UINT32_MAX);
3703 }
3704
3705 uint32_t
3706 ctl_decode_lun(uint64_t encoded)
3707 {
3708         uint8_t lun[8];
3709         uint32_t result = 0xffffffff;
3710
3711         be64enc(lun, encoded);
3712         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3713         case RPL_LUNDATA_ATYP_PERIPH:
3714                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3715                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3716                         result = lun[1];
3717                 break;
3718         case RPL_LUNDATA_ATYP_FLAT:
3719                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3720                     lun[6] == 0 && lun[7] == 0)
3721                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3722                 break;
3723         case RPL_LUNDATA_ATYP_EXTLUN:
3724                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3725                 case 0x02:
3726                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3727                         case 0x00:
3728                                 result = lun[1];
3729                                 break;
3730                         case 0x10:
3731                                 result = (lun[1] << 16) + (lun[2] << 8) +
3732                                     lun[3];
3733                                 break;
3734                         case 0x20:
3735                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3736                                         result = (lun[2] << 24) +
3737                                             (lun[3] << 16) + (lun[4] << 8) +
3738                                             lun[5];
3739                                 break;
3740                         }
3741                         break;
3742                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3743                         result = 0xffffffff;
3744                         break;
3745                 }
3746                 break;
3747         }
3748         return (result);
3749 }
3750
3751 uint64_t
3752 ctl_encode_lun(uint32_t decoded)
3753 {
3754         uint64_t l = decoded;
3755
3756         if (l <= 0xff)
3757                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3758         if (l <= 0x3fff)
3759                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3760         if (l <= 0xffffff)
3761                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3762                     (l << 32));
3763         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3764 }
3765
3766 int
3767 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3768 {
3769         int i;
3770
3771         for (i = first; i < last; i++) {
3772                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3773                         return (i);
3774         }
3775         return (-1);
3776 }
3777
3778 int
3779 ctl_set_mask(uint32_t *mask, uint32_t bit)
3780 {
3781         uint32_t chunk, piece;
3782
3783         chunk = bit >> 5;
3784         piece = bit % (sizeof(uint32_t) * 8);
3785
3786         if ((mask[chunk] & (1 << piece)) != 0)
3787                 return (-1);
3788         else
3789                 mask[chunk] |= (1 << piece);
3790
3791         return (0);
3792 }
3793
3794 int
3795 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3796 {
3797         uint32_t chunk, piece;
3798
3799         chunk = bit >> 5;
3800         piece = bit % (sizeof(uint32_t) * 8);
3801
3802         if ((mask[chunk] & (1 << piece)) == 0)
3803                 return (-1);
3804         else
3805                 mask[chunk] &= ~(1 << piece);
3806
3807         return (0);
3808 }
3809
3810 int
3811 ctl_is_set(uint32_t *mask, uint32_t bit)
3812 {
3813         uint32_t chunk, piece;
3814
3815         chunk = bit >> 5;
3816         piece = bit % (sizeof(uint32_t) * 8);
3817
3818         if ((mask[chunk] & (1 << piece)) == 0)
3819                 return (0);
3820         else
3821                 return (1);
3822 }
3823
3824 static uint64_t
3825 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3826 {
3827         uint64_t *t;
3828
3829         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3830         if (t == NULL)
3831                 return (0);
3832         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3833 }
3834
3835 static void
3836 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3837 {
3838         uint64_t *t;
3839
3840         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3841         if (t == NULL)
3842                 return;
3843         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3844 }
3845
3846 static void
3847 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3848 {
3849         uint64_t *p;
3850         u_int i;
3851
3852         i = residx/CTL_MAX_INIT_PER_PORT;
3853         if (lun->pr_keys[i] != NULL)
3854                 return;
3855         mtx_unlock(&lun->lun_lock);
3856         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3857             M_WAITOK | M_ZERO);
3858         mtx_lock(&lun->lun_lock);
3859         if (lun->pr_keys[i] == NULL)
3860                 lun->pr_keys[i] = p;
3861         else
3862                 free(p, M_CTL);
3863 }
3864
3865 static void
3866 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3867 {
3868         uint64_t *t;
3869
3870         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3871         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3872         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3873 }
3874
3875 /*
3876  * ctl_softc, pool_name, total_ctl_io are passed in.
3877  * npool is passed out.
3878  */
3879 int
3880 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3881                 uint32_t total_ctl_io, void **npool)
3882 {
3883         struct ctl_io_pool *pool;
3884
3885         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3886                                             M_NOWAIT | M_ZERO);
3887         if (pool == NULL)
3888                 return (ENOMEM);
3889
3890         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3891         pool->ctl_softc = ctl_softc;
3892 #ifdef IO_POOLS
3893         pool->zone = uma_zsecond_create(pool->name, NULL,
3894             NULL, NULL, NULL, ctl_softc->io_zone);
3895         /* uma_prealloc(pool->zone, total_ctl_io); */
3896 #else
3897         pool->zone = ctl_softc->io_zone;
3898 #endif
3899
3900         *npool = pool;
3901         return (0);
3902 }
3903
3904 void
3905 ctl_pool_free(struct ctl_io_pool *pool)
3906 {
3907
3908         if (pool == NULL)
3909                 return;
3910
3911 #ifdef IO_POOLS
3912         uma_zdestroy(pool->zone);
3913 #endif
3914         free(pool, M_CTL);
3915 }
3916
3917 union ctl_io *
3918 ctl_alloc_io(void *pool_ref)
3919 {
3920         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3921         union ctl_io *io;
3922
3923         io = uma_zalloc(pool->zone, M_WAITOK);
3924         if (io != NULL) {
3925                 io->io_hdr.pool = pool_ref;
3926                 CTL_SOFTC(io) = pool->ctl_softc;
3927         }
3928         return (io);
3929 }
3930
3931 union ctl_io *
3932 ctl_alloc_io_nowait(void *pool_ref)
3933 {
3934         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3935         union ctl_io *io;
3936
3937         io = uma_zalloc(pool->zone, M_NOWAIT);
3938         if (io != NULL) {
3939                 io->io_hdr.pool = pool_ref;
3940                 CTL_SOFTC(io) = pool->ctl_softc;
3941         }
3942         return (io);
3943 }
3944
3945 void
3946 ctl_free_io(union ctl_io *io)
3947 {
3948         struct ctl_io_pool *pool;
3949
3950         if (io == NULL)
3951                 return;
3952
3953         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3954         uma_zfree(pool->zone, io);
3955 }
3956
3957 void
3958 ctl_zero_io(union ctl_io *io)
3959 {
3960         struct ctl_io_pool *pool;
3961
3962         if (io == NULL)
3963                 return;
3964
3965         /*
3966          * May need to preserve linked list pointers at some point too.
3967          */
3968         pool = io->io_hdr.pool;
3969         memset(io, 0, sizeof(*io));
3970         io->io_hdr.pool = pool;
3971         CTL_SOFTC(io) = pool->ctl_softc;
3972 }
3973
3974 int
3975 ctl_expand_number(const char *buf, uint64_t *num)
3976 {
3977         char *endptr;
3978         uint64_t number;
3979         unsigned shift;
3980
3981         number = strtoq(buf, &endptr, 0);
3982
3983         switch (tolower((unsigned char)*endptr)) {
3984         case 'e':
3985                 shift = 60;
3986                 break;
3987         case 'p':
3988                 shift = 50;
3989                 break;
3990         case 't':
3991                 shift = 40;
3992                 break;
3993         case 'g':
3994                 shift = 30;
3995                 break;
3996         case 'm':
3997                 shift = 20;
3998                 break;
3999         case 'k':
4000                 shift = 10;
4001                 break;
4002         case 'b':
4003         case '\0': /* No unit. */
4004                 *num = number;
4005                 return (0);
4006         default:
4007                 /* Unrecognized unit. */
4008                 return (-1);
4009         }
4010
4011         if ((number << shift) >> shift != number) {
4012                 /* Overflow */
4013                 return (-1);
4014         }
4015         *num = number << shift;
4016         return (0);
4017 }
4018
4019
4020 /*
4021  * This routine could be used in the future to load default and/or saved
4022  * mode page parameters for a particuar lun.
4023  */
4024 static int
4025 ctl_init_page_index(struct ctl_lun *lun)
4026 {
4027         int i, page_code;
4028         struct ctl_page_index *page_index;
4029         const char *value;
4030         uint64_t ival;
4031
4032         memcpy(&lun->mode_pages.index, page_index_template,
4033                sizeof(page_index_template));
4034
4035         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4036
4037                 page_index = &lun->mode_pages.index[i];
4038                 if (lun->be_lun->lun_type == T_DIRECT &&
4039                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4040                         continue;
4041                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4042                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4043                         continue;
4044                 if (lun->be_lun->lun_type == T_CDROM &&
4045                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4046                         continue;
4047
4048                 page_code = page_index->page_code & SMPH_PC_MASK;
4049                 switch (page_code) {
4050                 case SMS_RW_ERROR_RECOVERY_PAGE: {
4051                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4052                             ("subpage %#x for page %#x is incorrect!",
4053                             page_index->subpage, page_code));
4054                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4055                                &rw_er_page_default,
4056                                sizeof(rw_er_page_default));
4057                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4058                                &rw_er_page_changeable,
4059                                sizeof(rw_er_page_changeable));
4060                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4061                                &rw_er_page_default,
4062                                sizeof(rw_er_page_default));
4063                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4064                                &rw_er_page_default,
4065                                sizeof(rw_er_page_default));
4066                         page_index->page_data =
4067                                 (uint8_t *)lun->mode_pages.rw_er_page;
4068                         break;
4069                 }
4070                 case SMS_FORMAT_DEVICE_PAGE: {
4071                         struct scsi_format_page *format_page;
4072
4073                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4074                             ("subpage %#x for page %#x is incorrect!",
4075                             page_index->subpage, page_code));
4076
4077                         /*
4078                          * Sectors per track are set above.  Bytes per
4079                          * sector need to be set here on a per-LUN basis.
4080                          */
4081                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4082                                &format_page_default,
4083                                sizeof(format_page_default));
4084                         memcpy(&lun->mode_pages.format_page[
4085                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4086                                sizeof(format_page_changeable));
4087                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4088                                &format_page_default,
4089                                sizeof(format_page_default));
4090                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4091                                &format_page_default,
4092                                sizeof(format_page_default));
4093
4094                         format_page = &lun->mode_pages.format_page[
4095                                 CTL_PAGE_CURRENT];
4096                         scsi_ulto2b(lun->be_lun->blocksize,
4097                                     format_page->bytes_per_sector);
4098
4099                         format_page = &lun->mode_pages.format_page[
4100                                 CTL_PAGE_DEFAULT];
4101                         scsi_ulto2b(lun->be_lun->blocksize,
4102                                     format_page->bytes_per_sector);
4103
4104                         format_page = &lun->mode_pages.format_page[
4105                                 CTL_PAGE_SAVED];
4106                         scsi_ulto2b(lun->be_lun->blocksize,
4107                                     format_page->bytes_per_sector);
4108
4109                         page_index->page_data =
4110                                 (uint8_t *)lun->mode_pages.format_page;
4111                         break;
4112                 }
4113                 case SMS_RIGID_DISK_PAGE: {
4114                         struct scsi_rigid_disk_page *rigid_disk_page;
4115                         uint32_t sectors_per_cylinder;
4116                         uint64_t cylinders;
4117 #ifndef __XSCALE__
4118                         int shift;
4119 #endif /* !__XSCALE__ */
4120
4121                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4122                             ("subpage %#x for page %#x is incorrect!",
4123                             page_index->subpage, page_code));
4124
4125                         /*
4126                          * Rotation rate and sectors per track are set
4127                          * above.  We calculate the cylinders here based on
4128                          * capacity.  Due to the number of heads and
4129                          * sectors per track we're using, smaller arrays
4130                          * may turn out to have 0 cylinders.  Linux and
4131                          * FreeBSD don't pay attention to these mode pages
4132                          * to figure out capacity, but Solaris does.  It
4133                          * seems to deal with 0 cylinders just fine, and
4134                          * works out a fake geometry based on the capacity.
4135                          */
4136                         memcpy(&lun->mode_pages.rigid_disk_page[
4137                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4138                                sizeof(rigid_disk_page_default));
4139                         memcpy(&lun->mode_pages.rigid_disk_page[
4140                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4141                                sizeof(rigid_disk_page_changeable));
4142
4143                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4144                                 CTL_DEFAULT_HEADS;
4145
4146                         /*
4147                          * The divide method here will be more accurate,
4148                          * probably, but results in floating point being
4149                          * used in the kernel on i386 (__udivdi3()).  On the
4150                          * XScale, though, __udivdi3() is implemented in
4151                          * software.
4152                          *
4153                          * The shift method for cylinder calculation is
4154                          * accurate if sectors_per_cylinder is a power of
4155                          * 2.  Otherwise it might be slightly off -- you
4156                          * might have a bit of a truncation problem.
4157                          */
4158 #ifdef  __XSCALE__
4159                         cylinders = (lun->be_lun->maxlba + 1) /
4160                                 sectors_per_cylinder;
4161 #else
4162                         for (shift = 31; shift > 0; shift--) {
4163                                 if (sectors_per_cylinder & (1 << shift))
4164                                         break;
4165                         }
4166                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4167 #endif
4168
4169                         /*
4170                          * We've basically got 3 bytes, or 24 bits for the
4171                          * cylinder size in the mode page.  If we're over,
4172                          * just round down to 2^24.
4173                          */
4174                         if (cylinders > 0xffffff)
4175                                 cylinders = 0xffffff;
4176
4177                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4178                                 CTL_PAGE_DEFAULT];
4179                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4180
4181                         if ((value = ctl_get_opt(&lun->be_lun->options,
4182                             "rpm")) != NULL) {
4183                                 scsi_ulto2b(strtol(value, NULL, 0),
4184                                      rigid_disk_page->rotation_rate);
4185                         }
4186
4187                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4188                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4189                                sizeof(rigid_disk_page_default));
4190                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4191                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4192                                sizeof(rigid_disk_page_default));
4193
4194                         page_index->page_data =
4195                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4196                         break;
4197                 }
4198                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4199                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4200                             ("subpage %#x for page %#x is incorrect!",
4201                             page_index->subpage, page_code));
4202                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4203                                &verify_er_page_default,
4204                                sizeof(verify_er_page_default));
4205                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4206                                &verify_er_page_changeable,
4207                                sizeof(verify_er_page_changeable));
4208                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4209                                &verify_er_page_default,
4210                                sizeof(verify_er_page_default));
4211                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4212                                &verify_er_page_default,
4213                                sizeof(verify_er_page_default));
4214                         page_index->page_data =
4215                                 (uint8_t *)lun->mode_pages.verify_er_page;
4216                         break;
4217                 }
4218                 case SMS_CACHING_PAGE: {
4219                         struct scsi_caching_page *caching_page;
4220
4221                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4222                             ("subpage %#x for page %#x is incorrect!",
4223                             page_index->subpage, page_code));
4224                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4225                                &caching_page_default,
4226                                sizeof(caching_page_default));
4227                         memcpy(&lun->mode_pages.caching_page[
4228                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4229                                sizeof(caching_page_changeable));
4230                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4231                                &caching_page_default,
4232                                sizeof(caching_page_default));
4233                         caching_page = &lun->mode_pages.caching_page[
4234                             CTL_PAGE_SAVED];
4235                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
4236                         if (value != NULL && strcmp(value, "off") == 0)
4237                                 caching_page->flags1 &= ~SCP_WCE;
4238                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4239                         if (value != NULL && strcmp(value, "off") == 0)
4240                                 caching_page->flags1 |= SCP_RCD;
4241                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4242                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4243                                sizeof(caching_page_default));
4244                         page_index->page_data =
4245                                 (uint8_t *)lun->mode_pages.caching_page;
4246                         break;
4247                 }
4248                 case SMS_CONTROL_MODE_PAGE: {
4249                         switch (page_index->subpage) {
4250                         case SMS_SUBPAGE_PAGE_0: {
4251                                 struct scsi_control_page *control_page;
4252
4253                                 memcpy(&lun->mode_pages.control_page[
4254                                     CTL_PAGE_DEFAULT],
4255                                        &control_page_default,
4256                                        sizeof(control_page_default));
4257                                 memcpy(&lun->mode_pages.control_page[
4258                                     CTL_PAGE_CHANGEABLE],
4259                                        &control_page_changeable,
4260                                        sizeof(control_page_changeable));
4261                                 memcpy(&lun->mode_pages.control_page[
4262                                     CTL_PAGE_SAVED],
4263                                        &control_page_default,
4264                                        sizeof(control_page_default));
4265                                 control_page = &lun->mode_pages.control_page[
4266                                     CTL_PAGE_SAVED];
4267                                 value = ctl_get_opt(&lun->be_lun->options,
4268                                     "reordering");
4269                                 if (value != NULL &&
4270                                     strcmp(value, "unrestricted") == 0) {
4271                                         control_page->queue_flags &=
4272                                             ~SCP_QUEUE_ALG_MASK;
4273                                         control_page->queue_flags |=
4274                                             SCP_QUEUE_ALG_UNRESTRICTED;
4275                                 }
4276                                 memcpy(&lun->mode_pages.control_page[
4277                                     CTL_PAGE_CURRENT],
4278                                        &lun->mode_pages.control_page[
4279                                     CTL_PAGE_SAVED],
4280                                        sizeof(control_page_default));
4281                                 page_index->page_data =
4282                                     (uint8_t *)lun->mode_pages.control_page;
4283                                 break;
4284                         }
4285                         case 0x01:
4286                                 memcpy(&lun->mode_pages.control_ext_page[
4287                                     CTL_PAGE_DEFAULT],
4288                                        &control_ext_page_default,
4289                                        sizeof(control_ext_page_default));
4290                                 memcpy(&lun->mode_pages.control_ext_page[
4291                                     CTL_PAGE_CHANGEABLE],
4292                                        &control_ext_page_changeable,
4293                                        sizeof(control_ext_page_changeable));
4294                                 memcpy(&lun->mode_pages.control_ext_page[
4295                                     CTL_PAGE_SAVED],
4296                                        &control_ext_page_default,
4297                                        sizeof(control_ext_page_default));
4298                                 memcpy(&lun->mode_pages.control_ext_page[
4299                                     CTL_PAGE_CURRENT],
4300                                        &lun->mode_pages.control_ext_page[
4301                                     CTL_PAGE_SAVED],
4302                                        sizeof(control_ext_page_default));
4303                                 page_index->page_data =
4304                                     (uint8_t *)lun->mode_pages.control_ext_page;
4305                                 break;
4306                         default:
4307                                 panic("subpage %#x for page %#x is incorrect!",
4308                                       page_index->subpage, page_code);
4309                         }
4310                         break;
4311                 }
4312                 case SMS_INFO_EXCEPTIONS_PAGE: {
4313                         switch (page_index->subpage) {
4314                         case SMS_SUBPAGE_PAGE_0:
4315                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4316                                        &ie_page_default,
4317                                        sizeof(ie_page_default));
4318                                 memcpy(&lun->mode_pages.ie_page[
4319                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4320                                        sizeof(ie_page_changeable));
4321                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4322                                        &ie_page_default,
4323                                        sizeof(ie_page_default));
4324                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4325                                        &ie_page_default,
4326                                        sizeof(ie_page_default));
4327                                 page_index->page_data =
4328                                         (uint8_t *)lun->mode_pages.ie_page;
4329                                 break;
4330                         case 0x02: {
4331                                 struct ctl_logical_block_provisioning_page *page;
4332
4333                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4334                                        &lbp_page_default,
4335                                        sizeof(lbp_page_default));
4336                                 memcpy(&lun->mode_pages.lbp_page[
4337                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4338                                        sizeof(lbp_page_changeable));
4339                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4340                                        &lbp_page_default,
4341                                        sizeof(lbp_page_default));
4342                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4343                                 value = ctl_get_opt(&lun->be_lun->options,
4344                                     "avail-threshold");
4345                                 if (value != NULL &&
4346                                     ctl_expand_number(value, &ival) == 0) {
4347                                         page->descr[0].flags |= SLBPPD_ENABLED |
4348                                             SLBPPD_ARMING_DEC;
4349                                         if (lun->be_lun->blocksize)
4350                                                 ival /= lun->be_lun->blocksize;
4351                                         else
4352                                                 ival /= 512;
4353                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4354                                             page->descr[0].count);
4355                                 }
4356                                 value = ctl_get_opt(&lun->be_lun->options,
4357                                     "used-threshold");
4358                                 if (value != NULL &&
4359                                     ctl_expand_number(value, &ival) == 0) {
4360                                         page->descr[1].flags |= SLBPPD_ENABLED |
4361                                             SLBPPD_ARMING_INC;
4362                                         if (lun->be_lun->blocksize)
4363                                                 ival /= lun->be_lun->blocksize;
4364                                         else
4365                                                 ival /= 512;
4366                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4367                                             page->descr[1].count);
4368                                 }
4369                                 value = ctl_get_opt(&lun->be_lun->options,
4370                                     "pool-avail-threshold");
4371                                 if (value != NULL &&
4372                                     ctl_expand_number(value, &ival) == 0) {
4373                                         page->descr[2].flags |= SLBPPD_ENABLED |
4374                                             SLBPPD_ARMING_DEC;
4375                                         if (lun->be_lun->blocksize)
4376                                                 ival /= lun->be_lun->blocksize;
4377                                         else
4378                                                 ival /= 512;
4379                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4380                                             page->descr[2].count);
4381                                 }
4382                                 value = ctl_get_opt(&lun->be_lun->options,
4383                                     "pool-used-threshold");
4384                                 if (value != NULL &&
4385                                     ctl_expand_number(value, &ival) == 0) {
4386                                         page->descr[3].flags |= SLBPPD_ENABLED |
4387                                             SLBPPD_ARMING_INC;
4388                                         if (lun->be_lun->blocksize)
4389                                                 ival /= lun->be_lun->blocksize;
4390                                         else
4391                                                 ival /= 512;
4392                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4393                                             page->descr[3].count);
4394                                 }
4395                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4396                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4397                                        sizeof(lbp_page_default));
4398                                 page_index->page_data =
4399                                         (uint8_t *)lun->mode_pages.lbp_page;
4400                                 break;
4401                         }
4402                         default:
4403                                 panic("subpage %#x for page %#x is incorrect!",
4404                                       page_index->subpage, page_code);
4405                         }
4406                         break;
4407                 }
4408                 case SMS_CDDVD_CAPS_PAGE:{
4409                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4410                             ("subpage %#x for page %#x is incorrect!",
4411                             page_index->subpage, page_code));
4412                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4413                                &cddvd_page_default,
4414                                sizeof(cddvd_page_default));
4415                         memcpy(&lun->mode_pages.cddvd_page[
4416                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4417                                sizeof(cddvd_page_changeable));
4418                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4419                                &cddvd_page_default,
4420                                sizeof(cddvd_page_default));
4421                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4422                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4423                                sizeof(cddvd_page_default));
4424                         page_index->page_data =
4425                                 (uint8_t *)lun->mode_pages.cddvd_page;
4426                         break;
4427                 }
4428                 default:
4429                         panic("invalid page code value %#x", page_code);
4430                 }
4431         }
4432
4433         return (CTL_RETVAL_COMPLETE);
4434 }
4435
4436 static int
4437 ctl_init_log_page_index(struct ctl_lun *lun)
4438 {
4439         struct ctl_page_index *page_index;
4440         int i, j, k, prev;
4441
4442         memcpy(&lun->log_pages.index, log_page_index_template,
4443                sizeof(log_page_index_template));
4444
4445         prev = -1;
4446         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4447
4448                 page_index = &lun->log_pages.index[i];
4449                 if (lun->be_lun->lun_type == T_DIRECT &&
4450                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4451                         continue;
4452                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4453                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4454                         continue;
4455                 if (lun->be_lun->lun_type == T_CDROM &&
4456                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4457                         continue;
4458
4459                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4460                     lun->backend->lun_attr == NULL)
4461                         continue;
4462
4463                 if (page_index->page_code != prev) {
4464                         lun->log_pages.pages_page[j] = page_index->page_code;
4465                         prev = page_index->page_code;
4466                         j++;
4467                 }
4468                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4469                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4470                 k++;
4471         }
4472         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4473         lun->log_pages.index[0].page_len = j;
4474         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4475         lun->log_pages.index[1].page_len = k * 2;
4476         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4477         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4478         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4479         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4480         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.ie_page;
4481         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.ie_page);
4482
4483         return (CTL_RETVAL_COMPLETE);
4484 }
4485
4486 static int
4487 hex2bin(const char *str, uint8_t *buf, int buf_size)
4488 {
4489         int i;
4490         u_char c;
4491
4492         memset(buf, 0, buf_size);
4493         while (isspace(str[0]))
4494                 str++;
4495         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4496                 str += 2;
4497         buf_size *= 2;
4498         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4499                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4500                         str++;
4501                 c = str[i];
4502                 if (isdigit(c))
4503                         c -= '0';
4504                 else if (isalpha(c))
4505                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4506                 else
4507                         break;
4508                 if (c >= 16)
4509                         break;
4510                 if ((i & 1) == 0)
4511                         buf[i / 2] |= (c << 4);
4512                 else
4513                         buf[i / 2] |= c;
4514         }
4515         return ((i + 1) / 2);
4516 }
4517
4518 /*
4519  * LUN allocation.
4520  *
4521  * Requirements:
4522  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4523  *   wants us to allocate the LUN and he can block.
4524  * - ctl_softc is always set
4525  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4526  *
4527  * Returns 0 for success, non-zero (errno) for failure.
4528  */
4529 static int
4530 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4531               struct ctl_be_lun *const be_lun)
4532 {
4533         struct ctl_lun *nlun, *lun;
4534         struct scsi_vpd_id_descriptor *desc;
4535         struct scsi_vpd_id_t10 *t10id;
4536         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4537         int lun_number, lun_malloced;
4538         int devidlen, idlen1, idlen2 = 0, len;
4539
4540         if (be_lun == NULL)
4541                 return (EINVAL);
4542
4543         /*
4544          * We currently only support Direct Access or Processor LUN types.
4545          */
4546         switch (be_lun->lun_type) {
4547         case T_DIRECT:
4548         case T_PROCESSOR:
4549         case T_CDROM:
4550                 break;
4551         case T_SEQUENTIAL:
4552         case T_CHANGER:
4553         default:
4554                 be_lun->lun_config_status(be_lun->be_lun,
4555                                           CTL_LUN_CONFIG_FAILURE);
4556                 break;
4557         }
4558         if (ctl_lun == NULL) {
4559                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4560                 lun_malloced = 1;
4561         } else {
4562                 lun_malloced = 0;
4563                 lun = ctl_lun;
4564         }
4565
4566         memset(lun, 0, sizeof(*lun));
4567         if (lun_malloced)
4568                 lun->flags = CTL_LUN_MALLOCED;
4569
4570         lun->pending_sense = malloc(sizeof(struct scsi_sense_data *) *
4571             ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO);
4572         lun->pending_ua = malloc(sizeof(ctl_ua_type *) * ctl_max_ports,
4573             M_DEVBUF, M_WAITOK | M_ZERO);
4574         lun->pr_keys = malloc(sizeof(uint64_t *) * ctl_max_ports,
4575             M_DEVBUF, M_WAITOK | M_ZERO);
4576
4577         /* Generate LUN ID. */
4578         devidlen = max(CTL_DEVID_MIN_LEN,
4579             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4580         idlen1 = sizeof(*t10id) + devidlen;
4581         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4582         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4583         if (scsiname != NULL) {
4584                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4585                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4586         }
4587         eui = ctl_get_opt(&be_lun->options, "eui");
4588         if (eui != NULL) {
4589                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4590         }
4591         naa = ctl_get_opt(&be_lun->options, "naa");
4592         if (naa != NULL) {
4593                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4594         }
4595         uuid = ctl_get_opt(&be_lun->options, "uuid");
4596         if (uuid != NULL) {
4597                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4598         }
4599         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4600             M_CTL, M_WAITOK | M_ZERO);
4601         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4602         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4603         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4604         desc->length = idlen1;
4605         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4606         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4607         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4608                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4609         } else {
4610                 strncpy(t10id->vendor, vendor,
4611                     min(sizeof(t10id->vendor), strlen(vendor)));
4612         }
4613         strncpy((char *)t10id->vendor_spec_id,
4614             (char *)be_lun->device_id, devidlen);
4615         if (scsiname != NULL) {
4616                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4617                     desc->length);
4618                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4619                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4620                     SVPD_ID_TYPE_SCSI_NAME;
4621                 desc->length = idlen2;
4622                 strlcpy(desc->identifier, scsiname, idlen2);
4623         }
4624         if (eui != NULL) {
4625                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4626                     desc->length);
4627                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4628                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4629                     SVPD_ID_TYPE_EUI64;
4630                 desc->length = hex2bin(eui, desc->identifier, 16);
4631                 desc->length = desc->length > 12 ? 16 :
4632                     (desc->length > 8 ? 12 : 8);
4633                 len -= 16 - desc->length;
4634         }
4635         if (naa != NULL) {
4636                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4637                     desc->length);
4638                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4639                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4640                     SVPD_ID_TYPE_NAA;
4641                 desc->length = hex2bin(naa, desc->identifier, 16);
4642                 desc->length = desc->length > 8 ? 16 : 8;
4643                 len -= 16 - desc->length;
4644         }
4645         if (uuid != NULL) {
4646                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4647                     desc->length);
4648                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4649                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4650                     SVPD_ID_TYPE_UUID;
4651                 desc->identifier[0] = 0x10;
4652                 hex2bin(uuid, &desc->identifier[2], 16);
4653                 desc->length = 18;
4654         }
4655         lun->lun_devid->len = len;
4656
4657         mtx_lock(&ctl_softc->ctl_lock);
4658         /*
4659          * See if the caller requested a particular LUN number.  If so, see
4660          * if it is available.  Otherwise, allocate the first available LUN.
4661          */
4662         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4663                 if ((be_lun->req_lun_id > (ctl_max_luns - 1))
4664                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4665                         mtx_unlock(&ctl_softc->ctl_lock);
4666                         if (be_lun->req_lun_id > (ctl_max_luns - 1)) {
4667                                 printf("ctl: requested LUN ID %d is higher "
4668                                        "than ctl_max_luns - 1 (%d)\n",
4669                                        be_lun->req_lun_id, ctl_max_luns - 1);
4670                         } else {
4671                                 /*
4672                                  * XXX KDM return an error, or just assign
4673                                  * another LUN ID in this case??
4674                                  */
4675                                 printf("ctl: requested LUN ID %d is already "
4676                                        "in use\n", be_lun->req_lun_id);
4677                         }
4678 fail:
4679                         free(lun->lun_devid, M_CTL);
4680                         if (lun->flags & CTL_LUN_MALLOCED)
4681                                 free(lun, M_CTL);
4682                         be_lun->lun_config_status(be_lun->be_lun,
4683                                                   CTL_LUN_CONFIG_FAILURE);
4684                         return (ENOSPC);
4685                 }
4686                 lun_number = be_lun->req_lun_id;
4687         } else {
4688                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, ctl_max_luns);
4689                 if (lun_number == -1) {
4690                         mtx_unlock(&ctl_softc->ctl_lock);
4691                         printf("ctl: can't allocate LUN, out of LUNs\n");
4692                         goto fail;
4693                 }
4694         }
4695         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4696         mtx_unlock(&ctl_softc->ctl_lock);
4697
4698         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4699         lun->lun = lun_number;
4700         lun->be_lun = be_lun;
4701         /*
4702          * The processor LUN is always enabled.  Disk LUNs come on line
4703          * disabled, and must be enabled by the backend.
4704          */
4705         lun->flags |= CTL_LUN_DISABLED;
4706         lun->backend = be_lun->be;
4707         be_lun->ctl_lun = lun;
4708         be_lun->lun_id = lun_number;
4709         atomic_add_int(&be_lun->be->num_luns, 1);
4710         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4711                 lun->flags |= CTL_LUN_EJECTED;
4712         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4713                 lun->flags |= CTL_LUN_NO_MEDIA;
4714         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4715                 lun->flags |= CTL_LUN_STOPPED;
4716
4717         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4718                 lun->flags |= CTL_LUN_PRIMARY_SC;
4719
4720         value = ctl_get_opt(&be_lun->options, "removable");
4721         if (value != NULL) {
4722                 if (strcmp(value, "on") == 0)
4723                         lun->flags |= CTL_LUN_REMOVABLE;
4724         } else if (be_lun->lun_type == T_CDROM)
4725                 lun->flags |= CTL_LUN_REMOVABLE;
4726
4727         lun->ctl_softc = ctl_softc;
4728 #ifdef CTL_TIME_IO
4729         lun->last_busy = getsbinuptime();
4730 #endif
4731         TAILQ_INIT(&lun->ooa_queue);
4732         TAILQ_INIT(&lun->blocked_queue);
4733         STAILQ_INIT(&lun->error_list);
4734         lun->ie_reported = 1;
4735         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4736         ctl_tpc_lun_init(lun);
4737         if (lun->flags & CTL_LUN_REMOVABLE) {
4738                 lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4,
4739                     M_CTL, M_WAITOK);
4740         }
4741
4742         /*
4743          * Initialize the mode and log page index.
4744          */
4745         ctl_init_page_index(lun);
4746         ctl_init_log_page_index(lun);
4747
4748         /* Setup statistics gathering */
4749 #ifdef CTL_LEGACY_STATS
4750         lun->legacy_stats.device_type = be_lun->lun_type;
4751         lun->legacy_stats.lun_number = lun_number;
4752         lun->legacy_stats.blocksize = be_lun->blocksize;
4753         if (be_lun->blocksize == 0)
4754                 lun->legacy_stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4755         lun->legacy_stats.ports = malloc(sizeof(struct ctl_lun_io_port_stats) *
4756             ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO);
4757         for (len = 0; len < ctl_max_ports; len++)
4758                 lun->legacy_stats.ports[len].targ_port = len;
4759 #endif /* CTL_LEGACY_STATS */
4760         lun->stats.item = lun_number;
4761
4762         /*
4763          * Now, before we insert this lun on the lun list, set the lun
4764          * inventory changed UA for all other luns.
4765          */
4766         mtx_lock(&ctl_softc->ctl_lock);
4767         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4768                 mtx_lock(&nlun->lun_lock);
4769                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4770                 mtx_unlock(&nlun->lun_lock);
4771         }
4772         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4773         ctl_softc->ctl_luns[lun_number] = lun;
4774         ctl_softc->num_luns++;
4775         mtx_unlock(&ctl_softc->ctl_lock);
4776
4777         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4778         return (0);
4779 }
4780
4781 /*
4782  * Delete a LUN.
4783  * Assumptions:
4784  * - LUN has already been marked invalid and any pending I/O has been taken
4785  *   care of.
4786  */
4787 static int
4788 ctl_free_lun(struct ctl_lun *lun)
4789 {
4790         struct ctl_softc *softc = lun->ctl_softc;
4791         struct ctl_lun *nlun;
4792         int i;
4793
4794         KASSERT(TAILQ_EMPTY(&lun->ooa_queue),
4795             ("Freeing a LUN %p with outstanding I/O!\n", lun));
4796
4797         mtx_lock(&softc->ctl_lock);
4798         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4799         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4800         softc->ctl_luns[lun->lun] = NULL;
4801         softc->num_luns--;
4802         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4803                 mtx_lock(&nlun->lun_lock);
4804                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4805                 mtx_unlock(&nlun->lun_lock);
4806         }
4807         mtx_unlock(&softc->ctl_lock);
4808
4809         /*
4810          * Tell the backend to free resources, if this LUN has a backend.
4811          */
4812         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4813         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4814
4815         lun->ie_reportcnt = UINT32_MAX;
4816         callout_drain(&lun->ie_callout);
4817         ctl_tpc_lun_shutdown(lun);
4818         mtx_destroy(&lun->lun_lock);
4819         free(lun->lun_devid, M_CTL);
4820         for (i = 0; i < ctl_max_ports; i++)
4821                 free(lun->pending_ua[i], M_CTL);
4822         free(lun->pending_ua, M_DEVBUF);
4823         for (i = 0; i < ctl_max_ports; i++)
4824                 free(lun->pr_keys[i], M_CTL);
4825         free(lun->pr_keys, M_DEVBUF);
4826         free(lun->write_buffer, M_CTL);
4827         free(lun->prevent, M_CTL);
4828         if (lun->flags & CTL_LUN_MALLOCED)
4829                 free(lun, M_CTL);
4830
4831         return (0);
4832 }
4833
4834 static void
4835 ctl_create_lun(struct ctl_be_lun *be_lun)
4836 {
4837
4838         /*
4839          * ctl_alloc_lun() should handle all potential failure cases.
4840          */
4841         ctl_alloc_lun(control_softc, NULL, be_lun);
4842 }
4843
4844 int
4845 ctl_add_lun(struct ctl_be_lun *be_lun)
4846 {
4847         struct ctl_softc *softc = control_softc;
4848
4849         mtx_lock(&softc->ctl_lock);
4850         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4851         mtx_unlock(&softc->ctl_lock);
4852         wakeup(&softc->pending_lun_queue);
4853
4854         return (0);
4855 }
4856
4857 int
4858 ctl_enable_lun(struct ctl_be_lun *be_lun)
4859 {
4860         struct ctl_softc *softc;
4861         struct ctl_port *port, *nport;
4862         struct ctl_lun *lun;
4863         int retval;
4864
4865         lun = (struct ctl_lun *)be_lun->ctl_lun;
4866         softc = lun->ctl_softc;
4867
4868         mtx_lock(&softc->ctl_lock);
4869         mtx_lock(&lun->lun_lock);
4870         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4871                 /*
4872                  * eh?  Why did we get called if the LUN is already
4873                  * enabled?
4874                  */
4875                 mtx_unlock(&lun->lun_lock);
4876                 mtx_unlock(&softc->ctl_lock);
4877                 return (0);
4878         }
4879         lun->flags &= ~CTL_LUN_DISABLED;
4880         mtx_unlock(&lun->lun_lock);
4881
4882         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4883                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4884                     port->lun_map != NULL || port->lun_enable == NULL)
4885                         continue;
4886
4887                 /*
4888                  * Drop the lock while we call the FETD's enable routine.
4889                  * This can lead to a callback into CTL (at least in the
4890                  * case of the internal initiator frontend.
4891                  */
4892                 mtx_unlock(&softc->ctl_lock);
4893                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4894                 mtx_lock(&softc->ctl_lock);
4895                 if (retval != 0) {
4896                         printf("%s: FETD %s port %d returned error "
4897                                "%d for lun_enable on lun %jd\n",
4898                                __func__, port->port_name, port->targ_port,
4899                                retval, (intmax_t)lun->lun);
4900                 }
4901         }
4902
4903         mtx_unlock(&softc->ctl_lock);
4904         ctl_isc_announce_lun(lun);
4905
4906         return (0);
4907 }
4908
4909 int
4910 ctl_disable_lun(struct ctl_be_lun *be_lun)
4911 {
4912         struct ctl_softc *softc;
4913         struct ctl_port *port;
4914         struct ctl_lun *lun;
4915         int retval;
4916
4917         lun = (struct ctl_lun *)be_lun->ctl_lun;
4918         softc = lun->ctl_softc;
4919
4920         mtx_lock(&softc->ctl_lock);
4921         mtx_lock(&lun->lun_lock);
4922         if (lun->flags & CTL_LUN_DISABLED) {
4923                 mtx_unlock(&lun->lun_lock);
4924                 mtx_unlock(&softc->ctl_lock);
4925                 return (0);
4926         }
4927         lun->flags |= CTL_LUN_DISABLED;
4928         mtx_unlock(&lun->lun_lock);
4929
4930         STAILQ_FOREACH(port, &softc->port_list, links) {
4931                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4932                     port->lun_map != NULL || port->lun_disable == NULL)
4933                         continue;
4934
4935                 /*
4936                  * Drop the lock before we call the frontend's disable
4937                  * routine, to avoid lock order reversals.
4938                  *
4939                  * XXX KDM what happens if the frontend list changes while
4940                  * we're traversing it?  It's unlikely, but should be handled.
4941                  */
4942                 mtx_unlock(&softc->ctl_lock);
4943                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4944                 mtx_lock(&softc->ctl_lock);
4945                 if (retval != 0) {
4946                         printf("%s: FETD %s port %d returned error "
4947                                "%d for lun_disable on lun %jd\n",
4948                                __func__, port->port_name, port->targ_port,
4949                                retval, (intmax_t)lun->lun);
4950                 }
4951         }
4952
4953         mtx_unlock(&softc->ctl_lock);
4954         ctl_isc_announce_lun(lun);
4955
4956         return (0);
4957 }
4958
4959 int
4960 ctl_start_lun(struct ctl_be_lun *be_lun)
4961 {
4962         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4963
4964         mtx_lock(&lun->lun_lock);
4965         lun->flags &= ~CTL_LUN_STOPPED;
4966         mtx_unlock(&lun->lun_lock);
4967         return (0);
4968 }
4969
4970 int
4971 ctl_stop_lun(struct ctl_be_lun *be_lun)
4972 {
4973         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4974
4975         mtx_lock(&lun->lun_lock);
4976         lun->flags |= CTL_LUN_STOPPED;
4977         mtx_unlock(&lun->lun_lock);
4978         return (0);
4979 }
4980
4981 int
4982 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4983 {
4984         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4985
4986         mtx_lock(&lun->lun_lock);
4987         lun->flags |= CTL_LUN_NO_MEDIA;
4988         mtx_unlock(&lun->lun_lock);
4989         return (0);
4990 }
4991
4992 int
4993 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4994 {
4995         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4996         union ctl_ha_msg msg;
4997
4998         mtx_lock(&lun->lun_lock);
4999         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
5000         if (lun->flags & CTL_LUN_REMOVABLE)
5001                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
5002         mtx_unlock(&lun->lun_lock);
5003         if ((lun->flags & CTL_LUN_REMOVABLE) &&
5004             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5005                 bzero(&msg.ua, sizeof(msg.ua));
5006                 msg.hdr.msg_type = CTL_MSG_UA;
5007                 msg.hdr.nexus.initid = -1;
5008                 msg.hdr.nexus.targ_port = -1;
5009                 msg.hdr.nexus.targ_lun = lun->lun;
5010                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5011                 msg.ua.ua_all = 1;
5012                 msg.ua.ua_set = 1;
5013                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
5014                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5015                     M_WAITOK);
5016         }
5017         return (0);
5018 }
5019
5020 int
5021 ctl_lun_ejected(struct ctl_be_lun *be_lun)
5022 {
5023         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5024
5025         mtx_lock(&lun->lun_lock);
5026         lun->flags |= CTL_LUN_EJECTED;
5027         mtx_unlock(&lun->lun_lock);
5028         return (0);
5029 }
5030
5031 int
5032 ctl_lun_primary(struct ctl_be_lun *be_lun)
5033 {
5034         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5035
5036         mtx_lock(&lun->lun_lock);
5037         lun->flags |= CTL_LUN_PRIMARY_SC;
5038         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5039         mtx_unlock(&lun->lun_lock);
5040         ctl_isc_announce_lun(lun);
5041         return (0);
5042 }
5043
5044 int
5045 ctl_lun_secondary(struct ctl_be_lun *be_lun)
5046 {
5047         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5048
5049         mtx_lock(&lun->lun_lock);
5050         lun->flags &= ~CTL_LUN_PRIMARY_SC;
5051         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5052         mtx_unlock(&lun->lun_lock);
5053         ctl_isc_announce_lun(lun);
5054         return (0);
5055 }
5056
5057 int
5058 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
5059 {
5060         struct ctl_softc *softc;
5061         struct ctl_lun *lun;
5062
5063         lun = (struct ctl_lun *)be_lun->ctl_lun;
5064         softc = lun->ctl_softc;
5065
5066         mtx_lock(&lun->lun_lock);
5067
5068         /*
5069          * The LUN needs to be disabled before it can be marked invalid.
5070          */
5071         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
5072                 mtx_unlock(&lun->lun_lock);
5073                 return (-1);
5074         }
5075         /*
5076          * Mark the LUN invalid.
5077          */
5078         lun->flags |= CTL_LUN_INVALID;
5079
5080         /*
5081          * If there is nothing in the OOA queue, go ahead and free the LUN.
5082          * If we have something in the OOA queue, we'll free it when the
5083          * last I/O completes.
5084          */
5085         if (TAILQ_EMPTY(&lun->ooa_queue)) {
5086                 mtx_unlock(&lun->lun_lock);
5087                 ctl_free_lun(lun);
5088         } else
5089                 mtx_unlock(&lun->lun_lock);
5090
5091         return (0);
5092 }
5093
5094 void
5095 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5096 {
5097         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5098         union ctl_ha_msg msg;
5099
5100         mtx_lock(&lun->lun_lock);
5101         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
5102         mtx_unlock(&lun->lun_lock);
5103         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5104                 /* Send msg to other side. */
5105                 bzero(&msg.ua, sizeof(msg.ua));
5106                 msg.hdr.msg_type = CTL_MSG_UA;
5107                 msg.hdr.nexus.initid = -1;
5108                 msg.hdr.nexus.targ_port = -1;
5109                 msg.hdr.nexus.targ_lun = lun->lun;
5110                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5111                 msg.ua.ua_all = 1;
5112                 msg.ua.ua_set = 1;
5113                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
5114                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5115                     M_WAITOK);
5116         }
5117 }
5118
5119 /*
5120  * Backend "memory move is complete" callback for requests that never
5121  * make it down to say RAIDCore's configuration code.
5122  */
5123 int
5124 ctl_config_move_done(union ctl_io *io)
5125 {
5126         int retval;
5127
5128         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5129         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5130             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5131
5132         if ((io->io_hdr.port_status != 0) &&
5133             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5134              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5135                 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
5136                     /*retry_count*/ io->io_hdr.port_status);
5137         } else if (io->scsiio.kern_data_resid != 0 &&
5138             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
5139             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5140              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5141                 ctl_set_invalid_field_ciu(&io->scsiio);
5142         }
5143
5144         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5145                 ctl_data_print(io);
5146         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5147             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5148              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5149             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5150                 /*
5151                  * XXX KDM just assuming a single pointer here, and not a
5152                  * S/G list.  If we start using S/G lists for config data,
5153                  * we'll need to know how to clean them up here as well.
5154                  */
5155                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5156                         free(io->scsiio.kern_data_ptr, M_CTL);
5157                 ctl_done(io);
5158                 retval = CTL_RETVAL_COMPLETE;
5159         } else {
5160                 /*
5161                  * XXX KDM now we need to continue data movement.  Some
5162                  * options:
5163                  * - call ctl_scsiio() again?  We don't do this for data
5164                  *   writes, because for those at least we know ahead of
5165                  *   time where the write will go and how long it is.  For
5166                  *   config writes, though, that information is largely
5167                  *   contained within the write itself, thus we need to
5168                  *   parse out the data again.
5169                  *
5170                  * - Call some other function once the data is in?
5171                  */
5172
5173                 /*
5174                  * XXX KDM call ctl_scsiio() again for now, and check flag
5175                  * bits to see whether we're allocated or not.
5176                  */
5177                 retval = ctl_scsiio(&io->scsiio);
5178         }
5179         return (retval);
5180 }
5181
5182 /*
5183  * This gets called by a backend driver when it is done with a
5184  * data_submit method.
5185  */
5186 void
5187 ctl_data_submit_done(union ctl_io *io)
5188 {
5189         /*
5190          * If the IO_CONT flag is set, we need to call the supplied
5191          * function to continue processing the I/O, instead of completing
5192          * the I/O just yet.
5193          *
5194          * If there is an error, though, we don't want to keep processing.
5195          * Instead, just send status back to the initiator.
5196          */
5197         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5198             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5199             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5200              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5201                 io->scsiio.io_cont(io);
5202                 return;
5203         }
5204         ctl_done(io);
5205 }
5206
5207 /*
5208  * This gets called by a backend driver when it is done with a
5209  * configuration write.
5210  */
5211 void
5212 ctl_config_write_done(union ctl_io *io)
5213 {
5214         uint8_t *buf;
5215
5216         /*
5217          * If the IO_CONT flag is set, we need to call the supplied
5218          * function to continue processing the I/O, instead of completing
5219          * the I/O just yet.
5220          *
5221          * If there is an error, though, we don't want to keep processing.
5222          * Instead, just send status back to the initiator.
5223          */
5224         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5225             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5226             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5227              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5228                 io->scsiio.io_cont(io);
5229                 return;
5230         }
5231         /*
5232          * Since a configuration write can be done for commands that actually
5233          * have data allocated, like write buffer, and commands that have
5234          * no data, like start/stop unit, we need to check here.
5235          */
5236         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5237                 buf = io->scsiio.kern_data_ptr;
5238         else
5239                 buf = NULL;
5240         ctl_done(io);
5241         if (buf)
5242                 free(buf, M_CTL);
5243 }
5244
5245 void
5246 ctl_config_read_done(union ctl_io *io)
5247 {
5248         uint8_t *buf;
5249
5250         /*
5251          * If there is some error -- we are done, skip data transfer.
5252          */
5253         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5254             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5255              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5256                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5257                         buf = io->scsiio.kern_data_ptr;
5258                 else
5259                         buf = NULL;
5260                 ctl_done(io);
5261                 if (buf)
5262                         free(buf, M_CTL);
5263                 return;
5264         }
5265
5266         /*
5267          * If the IO_CONT flag is set, we need to call the supplied
5268          * function to continue processing the I/O, instead of completing
5269          * the I/O just yet.
5270          */
5271         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5272                 io->scsiio.io_cont(io);
5273                 return;
5274         }
5275
5276         ctl_datamove(io);
5277 }
5278
5279 /*
5280  * SCSI release command.
5281  */
5282 int
5283 ctl_scsi_release(struct ctl_scsiio *ctsio)
5284 {
5285         struct ctl_lun *lun = CTL_LUN(ctsio);
5286         uint32_t residx;
5287
5288         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5289
5290         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5291
5292         /*
5293          * XXX KDM right now, we only support LUN reservation.  We don't
5294          * support 3rd party reservations, or extent reservations, which
5295          * might actually need the parameter list.  If we've gotten this
5296          * far, we've got a LUN reservation.  Anything else got kicked out
5297          * above.  So, according to SPC, ignore the length.
5298          */
5299
5300         mtx_lock(&lun->lun_lock);
5301
5302         /*
5303          * According to SPC, it is not an error for an intiator to attempt
5304          * to release a reservation on a LUN that isn't reserved, or that
5305          * is reserved by another initiator.  The reservation can only be
5306          * released, though, by the initiator who made it or by one of
5307          * several reset type events.
5308          */
5309         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5310                         lun->flags &= ~CTL_LUN_RESERVED;
5311
5312         mtx_unlock(&lun->lun_lock);
5313
5314         ctl_set_success(ctsio);
5315         ctl_done((union ctl_io *)ctsio);
5316         return (CTL_RETVAL_COMPLETE);
5317 }
5318
5319 int
5320 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5321 {
5322         struct ctl_lun *lun = CTL_LUN(ctsio);
5323         uint32_t residx;
5324
5325         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5326
5327         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5328
5329         /*
5330          * XXX KDM right now, we only support LUN reservation.  We don't
5331          * support 3rd party reservations, or extent reservations, which
5332          * might actually need the parameter list.  If we've gotten this
5333          * far, we've got a LUN reservation.  Anything else got kicked out
5334          * above.  So, according to SPC, ignore the length.
5335          */
5336
5337         mtx_lock(&lun->lun_lock);
5338         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5339                 ctl_set_reservation_conflict(ctsio);
5340                 goto bailout;
5341         }
5342
5343         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5344         if (lun->flags & CTL_LUN_PR_RESERVED) {
5345                 ctl_set_success(ctsio);
5346                 goto bailout;
5347         }
5348
5349         lun->flags |= CTL_LUN_RESERVED;
5350         lun->res_idx = residx;
5351         ctl_set_success(ctsio);
5352
5353 bailout:
5354         mtx_unlock(&lun->lun_lock);
5355         ctl_done((union ctl_io *)ctsio);
5356         return (CTL_RETVAL_COMPLETE);
5357 }
5358
5359 int
5360 ctl_start_stop(struct ctl_scsiio *ctsio)
5361 {
5362         struct ctl_lun *lun = CTL_LUN(ctsio);
5363         struct scsi_start_stop_unit *cdb;
5364         int retval;
5365
5366         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5367
5368         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5369
5370         if ((cdb->how & SSS_PC_MASK) == 0) {
5371                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5372                     (cdb->how & SSS_START) == 0) {
5373                         uint32_t residx;
5374
5375                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5376                         if (ctl_get_prkey(lun, residx) == 0 ||
5377                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5378
5379                                 ctl_set_reservation_conflict(ctsio);
5380                                 ctl_done((union ctl_io *)ctsio);
5381                                 return (CTL_RETVAL_COMPLETE);
5382                         }
5383                 }
5384
5385                 if ((cdb->how & SSS_LOEJ) &&
5386                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5387                         ctl_set_invalid_field(ctsio,
5388                                               /*sks_valid*/ 1,
5389                                               /*command*/ 1,
5390                                               /*field*/ 4,
5391                                               /*bit_valid*/ 1,
5392                                               /*bit*/ 1);
5393                         ctl_done((union ctl_io *)ctsio);
5394                         return (CTL_RETVAL_COMPLETE);
5395                 }
5396
5397                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5398                     lun->prevent_count > 0) {
5399                         /* "Medium removal prevented" */
5400                         ctl_set_sense(ctsio, /*current_error*/ 1,
5401                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5402                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5403                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5404                         ctl_done((union ctl_io *)ctsio);
5405                         return (CTL_RETVAL_COMPLETE);
5406                 }
5407         }
5408
5409         retval = lun->backend->config_write((union ctl_io *)ctsio);
5410         return (retval);
5411 }
5412
5413 int
5414 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5415 {
5416         struct ctl_lun *lun = CTL_LUN(ctsio);
5417         struct scsi_prevent *cdb;
5418         int retval;
5419         uint32_t initidx;
5420
5421         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5422
5423         cdb = (struct scsi_prevent *)ctsio->cdb;
5424
5425         if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) {
5426                 ctl_set_invalid_opcode(ctsio);
5427                 ctl_done((union ctl_io *)ctsio);
5428                 return (CTL_RETVAL_COMPLETE);
5429         }
5430
5431         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5432         mtx_lock(&lun->lun_lock);
5433         if ((cdb->how & PR_PREVENT) &&
5434             ctl_is_set(lun->prevent, initidx) == 0) {
5435                 ctl_set_mask(lun->prevent, initidx);
5436                 lun->prevent_count++;
5437         } else if ((cdb->how & PR_PREVENT) == 0 &&
5438             ctl_is_set(lun->prevent, initidx)) {
5439                 ctl_clear_mask(lun->prevent, initidx);
5440                 lun->prevent_count--;
5441         }
5442         mtx_unlock(&lun->lun_lock);
5443         retval = lun->backend->config_write((union ctl_io *)ctsio);
5444         return (retval);
5445 }
5446
5447 /*
5448  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5449  * we don't really do anything with the LBA and length fields if the user
5450  * passes them in.  Instead we'll just flush out the cache for the entire
5451  * LUN.
5452  */
5453 int
5454 ctl_sync_cache(struct ctl_scsiio *ctsio)
5455 {
5456         struct ctl_lun *lun = CTL_LUN(ctsio);
5457         struct ctl_lba_len_flags *lbalen;
5458         uint64_t starting_lba;
5459         uint32_t block_count;
5460         int retval;
5461         uint8_t byte2;
5462
5463         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5464
5465         retval = 0;
5466
5467         switch (ctsio->cdb[0]) {
5468         case SYNCHRONIZE_CACHE: {
5469                 struct scsi_sync_cache *cdb;
5470                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5471
5472                 starting_lba = scsi_4btoul(cdb->begin_lba);
5473                 block_count = scsi_2btoul(cdb->lb_count);
5474                 byte2 = cdb->byte2;
5475                 break;
5476         }
5477         case SYNCHRONIZE_CACHE_16: {
5478                 struct scsi_sync_cache_16 *cdb;
5479                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5480
5481                 starting_lba = scsi_8btou64(cdb->begin_lba);
5482                 block_count = scsi_4btoul(cdb->lb_count);
5483                 byte2 = cdb->byte2;
5484                 break;
5485         }
5486         default:
5487                 ctl_set_invalid_opcode(ctsio);
5488                 ctl_done((union ctl_io *)ctsio);
5489                 goto bailout;
5490                 break; /* NOTREACHED */
5491         }
5492
5493         /*
5494          * We check the LBA and length, but don't do anything with them.
5495          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5496          * get flushed.  This check will just help satisfy anyone who wants
5497          * to see an error for an out of range LBA.
5498          */
5499         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5500                 ctl_set_lba_out_of_range(ctsio,
5501                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5502                 ctl_done((union ctl_io *)ctsio);
5503                 goto bailout;
5504         }
5505
5506         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5507         lbalen->lba = starting_lba;
5508         lbalen->len = block_count;
5509         lbalen->flags = byte2;
5510         retval = lun->backend->config_write((union ctl_io *)ctsio);
5511
5512 bailout:
5513         return (retval);
5514 }
5515
5516 int
5517 ctl_format(struct ctl_scsiio *ctsio)
5518 {
5519         struct scsi_format *cdb;
5520         int length, defect_list_len;
5521
5522         CTL_DEBUG_PRINT(("ctl_format\n"));
5523
5524         cdb = (struct scsi_format *)ctsio->cdb;
5525
5526         length = 0;
5527         if (cdb->byte2 & SF_FMTDATA) {
5528                 if (cdb->byte2 & SF_LONGLIST)
5529                         length = sizeof(struct scsi_format_header_long);
5530                 else
5531                         length = sizeof(struct scsi_format_header_short);
5532         }
5533
5534         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5535          && (length > 0)) {
5536                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5537                 ctsio->kern_data_len = length;
5538                 ctsio->kern_total_len = length;
5539                 ctsio->kern_rel_offset = 0;
5540                 ctsio->kern_sg_entries = 0;
5541                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5542                 ctsio->be_move_done = ctl_config_move_done;
5543                 ctl_datamove((union ctl_io *)ctsio);
5544
5545                 return (CTL_RETVAL_COMPLETE);
5546         }
5547
5548         defect_list_len = 0;
5549
5550         if (cdb->byte2 & SF_FMTDATA) {
5551                 if (cdb->byte2 & SF_LONGLIST) {
5552                         struct scsi_format_header_long *header;
5553
5554                         header = (struct scsi_format_header_long *)
5555                                 ctsio->kern_data_ptr;
5556
5557                         defect_list_len = scsi_4btoul(header->defect_list_len);
5558                         if (defect_list_len != 0) {
5559                                 ctl_set_invalid_field(ctsio,
5560                                                       /*sks_valid*/ 1,
5561                                                       /*command*/ 0,
5562                                                       /*field*/ 2,
5563                                                       /*bit_valid*/ 0,
5564                                                       /*bit*/ 0);
5565                                 goto bailout;
5566                         }
5567                 } else {
5568                         struct scsi_format_header_short *header;
5569
5570                         header = (struct scsi_format_header_short *)
5571                                 ctsio->kern_data_ptr;
5572
5573                         defect_list_len = scsi_2btoul(header->defect_list_len);
5574                         if (defect_list_len != 0) {
5575                                 ctl_set_invalid_field(ctsio,
5576                                                       /*sks_valid*/ 1,
5577                                                       /*command*/ 0,
5578                                                       /*field*/ 2,
5579                                                       /*bit_valid*/ 0,
5580                                                       /*bit*/ 0);
5581                                 goto bailout;
5582                         }
5583                 }
5584         }
5585
5586         ctl_set_success(ctsio);
5587 bailout:
5588
5589         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5590                 free(ctsio->kern_data_ptr, M_CTL);
5591                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5592         }
5593
5594         ctl_done((union ctl_io *)ctsio);
5595         return (CTL_RETVAL_COMPLETE);
5596 }
5597
5598 int
5599 ctl_read_buffer(struct ctl_scsiio *ctsio)
5600 {
5601         struct ctl_lun *lun = CTL_LUN(ctsio);
5602         uint64_t buffer_offset;
5603         uint32_t len;
5604         uint8_t byte2;
5605         static uint8_t descr[4];
5606         static uint8_t echo_descr[4] = { 0 };
5607
5608         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5609
5610         switch (ctsio->cdb[0]) {
5611         case READ_BUFFER: {
5612                 struct scsi_read_buffer *cdb;
5613
5614                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5615                 buffer_offset = scsi_3btoul(cdb->offset);
5616                 len = scsi_3btoul(cdb->length);
5617                 byte2 = cdb->byte2;
5618                 break;
5619         }
5620         case READ_BUFFER_16: {
5621                 struct scsi_read_buffer_16 *cdb;
5622
5623                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5624                 buffer_offset = scsi_8btou64(cdb->offset);
5625                 len = scsi_4btoul(cdb->length);
5626                 byte2 = cdb->byte2;
5627                 break;
5628         }
5629         default: /* This shouldn't happen. */
5630                 ctl_set_invalid_opcode(ctsio);
5631                 ctl_done((union ctl_io *)ctsio);
5632                 return (CTL_RETVAL_COMPLETE);
5633         }
5634
5635         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5636             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5637                 ctl_set_invalid_field(ctsio,
5638                                       /*sks_valid*/ 1,
5639                                       /*command*/ 1,
5640                                       /*field*/ 6,
5641                                       /*bit_valid*/ 0,
5642                                       /*bit*/ 0);
5643                 ctl_done((union ctl_io *)ctsio);
5644                 return (CTL_RETVAL_COMPLETE);
5645         }
5646
5647         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5648                 descr[0] = 0;
5649                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5650                 ctsio->kern_data_ptr = descr;
5651                 len = min(len, sizeof(descr));
5652         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5653                 ctsio->kern_data_ptr = echo_descr;
5654                 len = min(len, sizeof(echo_descr));
5655         } else {
5656                 if (lun->write_buffer == NULL) {
5657                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5658                             M_CTL, M_WAITOK);
5659                 }
5660                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5661         }
5662         ctsio->kern_data_len = len;
5663         ctsio->kern_total_len = len;
5664         ctsio->kern_rel_offset = 0;
5665         ctsio->kern_sg_entries = 0;
5666         ctl_set_success(ctsio);
5667         ctsio->be_move_done = ctl_config_move_done;
5668         ctl_datamove((union ctl_io *)ctsio);
5669         return (CTL_RETVAL_COMPLETE);
5670 }
5671
5672 int
5673 ctl_write_buffer(struct ctl_scsiio *ctsio)
5674 {
5675         struct ctl_lun *lun = CTL_LUN(ctsio);
5676         struct scsi_write_buffer *cdb;
5677         int buffer_offset, len;
5678
5679         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5680
5681         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5682
5683         len = scsi_3btoul(cdb->length);
5684         buffer_offset = scsi_3btoul(cdb->offset);
5685
5686         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5687                 ctl_set_invalid_field(ctsio,
5688                                       /*sks_valid*/ 1,
5689                                       /*command*/ 1,
5690                                       /*field*/ 6,
5691                                       /*bit_valid*/ 0,
5692                                       /*bit*/ 0);
5693                 ctl_done((union ctl_io *)ctsio);
5694                 return (CTL_RETVAL_COMPLETE);
5695         }
5696
5697         /*
5698          * If we've got a kernel request that hasn't been malloced yet,
5699          * malloc it and tell the caller the data buffer is here.
5700          */
5701         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5702                 if (lun->write_buffer == NULL) {
5703                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5704                             M_CTL, M_WAITOK);
5705                 }
5706                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5707                 ctsio->kern_data_len = len;
5708                 ctsio->kern_total_len = len;
5709                 ctsio->kern_rel_offset = 0;
5710                 ctsio->kern_sg_entries = 0;
5711                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5712                 ctsio->be_move_done = ctl_config_move_done;
5713                 ctl_datamove((union ctl_io *)ctsio);
5714
5715                 return (CTL_RETVAL_COMPLETE);
5716         }
5717
5718         ctl_set_success(ctsio);
5719         ctl_done((union ctl_io *)ctsio);
5720         return (CTL_RETVAL_COMPLETE);
5721 }
5722
5723 int
5724 ctl_write_same(struct ctl_scsiio *ctsio)
5725 {
5726         struct ctl_lun *lun = CTL_LUN(ctsio);
5727         struct ctl_lba_len_flags *lbalen;
5728         uint64_t lba;
5729         uint32_t num_blocks;
5730         int len, retval;
5731         uint8_t byte2;
5732
5733         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5734
5735         switch (ctsio->cdb[0]) {
5736         case WRITE_SAME_10: {
5737                 struct scsi_write_same_10 *cdb;
5738
5739                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5740
5741                 lba = scsi_4btoul(cdb->addr);
5742                 num_blocks = scsi_2btoul(cdb->length);
5743                 byte2 = cdb->byte2;
5744                 break;
5745         }
5746         case WRITE_SAME_16: {
5747                 struct scsi_write_same_16 *cdb;
5748
5749                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5750
5751                 lba = scsi_8btou64(cdb->addr);
5752                 num_blocks = scsi_4btoul(cdb->length);
5753                 byte2 = cdb->byte2;
5754                 break;
5755         }
5756         default:
5757                 /*
5758                  * We got a command we don't support.  This shouldn't
5759                  * happen, commands should be filtered out above us.
5760                  */
5761                 ctl_set_invalid_opcode(ctsio);
5762                 ctl_done((union ctl_io *)ctsio);
5763
5764                 return (CTL_RETVAL_COMPLETE);
5765                 break; /* NOTREACHED */
5766         }
5767
5768         /* ANCHOR flag can be used only together with UNMAP */
5769         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5770                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5771                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5772                 ctl_done((union ctl_io *)ctsio);
5773                 return (CTL_RETVAL_COMPLETE);
5774         }
5775
5776         /*
5777          * The first check is to make sure we're in bounds, the second
5778          * check is to catch wrap-around problems.  If the lba + num blocks
5779          * is less than the lba, then we've wrapped around and the block
5780          * range is invalid anyway.
5781          */
5782         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5783          || ((lba + num_blocks) < lba)) {
5784                 ctl_set_lba_out_of_range(ctsio,
5785                     MAX(lba, lun->be_lun->maxlba + 1));
5786                 ctl_done((union ctl_io *)ctsio);
5787                 return (CTL_RETVAL_COMPLETE);
5788         }
5789
5790         /* Zero number of blocks means "to the last logical block" */
5791         if (num_blocks == 0) {
5792                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5793                         ctl_set_invalid_field(ctsio,
5794                                               /*sks_valid*/ 0,
5795                                               /*command*/ 1,
5796                                               /*field*/ 0,
5797                                               /*bit_valid*/ 0,
5798                                               /*bit*/ 0);
5799                         ctl_done((union ctl_io *)ctsio);
5800                         return (CTL_RETVAL_COMPLETE);
5801                 }
5802                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5803         }
5804
5805         len = lun->be_lun->blocksize;
5806
5807         /*
5808          * If we've got a kernel request that hasn't been malloced yet,
5809          * malloc it and tell the caller the data buffer is here.
5810          */
5811         if ((byte2 & SWS_NDOB) == 0 &&
5812             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5813                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5814                 ctsio->kern_data_len = len;
5815                 ctsio->kern_total_len = len;
5816                 ctsio->kern_rel_offset = 0;
5817                 ctsio->kern_sg_entries = 0;
5818                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5819                 ctsio->be_move_done = ctl_config_move_done;
5820                 ctl_datamove((union ctl_io *)ctsio);
5821
5822                 return (CTL_RETVAL_COMPLETE);
5823         }
5824
5825         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5826         lbalen->lba = lba;
5827         lbalen->len = num_blocks;
5828         lbalen->flags = byte2;
5829         retval = lun->backend->config_write((union ctl_io *)ctsio);
5830
5831         return (retval);
5832 }
5833
5834 int
5835 ctl_unmap(struct ctl_scsiio *ctsio)
5836 {
5837         struct ctl_lun *lun = CTL_LUN(ctsio);
5838         struct scsi_unmap *cdb;
5839         struct ctl_ptr_len_flags *ptrlen;
5840         struct scsi_unmap_header *hdr;
5841         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5842         uint64_t lba;
5843         uint32_t num_blocks;
5844         int len, retval;
5845         uint8_t byte2;
5846
5847         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5848
5849         cdb = (struct scsi_unmap *)ctsio->cdb;
5850         len = scsi_2btoul(cdb->length);
5851         byte2 = cdb->byte2;
5852
5853         /*
5854          * If we've got a kernel request that hasn't been malloced yet,
5855          * malloc it and tell the caller the data buffer is here.
5856          */
5857         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5858                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5859                 ctsio->kern_data_len = len;
5860                 ctsio->kern_total_len = len;
5861                 ctsio->kern_rel_offset = 0;
5862                 ctsio->kern_sg_entries = 0;
5863                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5864                 ctsio->be_move_done = ctl_config_move_done;
5865                 ctl_datamove((union ctl_io *)ctsio);
5866
5867                 return (CTL_RETVAL_COMPLETE);
5868         }
5869
5870         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5871         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5872         if (len < sizeof (*hdr) ||
5873             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5874             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5875             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5876                 ctl_set_invalid_field(ctsio,
5877                                       /*sks_valid*/ 0,
5878                                       /*command*/ 0,
5879                                       /*field*/ 0,
5880                                       /*bit_valid*/ 0,
5881                                       /*bit*/ 0);
5882                 goto done;
5883         }
5884         len = scsi_2btoul(hdr->desc_length);
5885         buf = (struct scsi_unmap_desc *)(hdr + 1);
5886         end = buf + len / sizeof(*buf);
5887
5888         endnz = buf;
5889         for (range = buf; range < end; range++) {
5890                 lba = scsi_8btou64(range->lba);
5891                 num_blocks = scsi_4btoul(range->length);
5892                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5893                  || ((lba + num_blocks) < lba)) {
5894                         ctl_set_lba_out_of_range(ctsio,
5895                             MAX(lba, lun->be_lun->maxlba + 1));
5896                         ctl_done((union ctl_io *)ctsio);
5897                         return (CTL_RETVAL_COMPLETE);
5898                 }
5899                 if (num_blocks != 0)
5900                         endnz = range + 1;
5901         }
5902
5903         /*
5904          * Block backend can not handle zero last range.
5905          * Filter it out and return if there is nothing left.
5906          */
5907         len = (uint8_t *)endnz - (uint8_t *)buf;
5908         if (len == 0) {
5909                 ctl_set_success(ctsio);
5910                 goto done;
5911         }
5912
5913         mtx_lock(&lun->lun_lock);
5914         ptrlen = (struct ctl_ptr_len_flags *)
5915             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5916         ptrlen->ptr = (void *)buf;
5917         ptrlen->len = len;
5918         ptrlen->flags = byte2;
5919         ctl_check_blocked(lun);
5920         mtx_unlock(&lun->lun_lock);
5921
5922         retval = lun->backend->config_write((union ctl_io *)ctsio);
5923         return (retval);
5924
5925 done:
5926         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5927                 free(ctsio->kern_data_ptr, M_CTL);
5928                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5929         }
5930         ctl_done((union ctl_io *)ctsio);
5931         return (CTL_RETVAL_COMPLETE);
5932 }
5933
5934 int
5935 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5936                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5937 {
5938         struct ctl_lun *lun = CTL_LUN(ctsio);
5939         uint8_t *current_cp;
5940         int set_ua;
5941         uint32_t initidx;
5942
5943         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5944         set_ua = 0;
5945
5946         current_cp = (page_index->page_data + (page_index->page_len *
5947             CTL_PAGE_CURRENT));
5948
5949         mtx_lock(&lun->lun_lock);
5950         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5951                 memcpy(current_cp, page_ptr, page_index->page_len);
5952                 set_ua = 1;
5953         }
5954         if (set_ua != 0)
5955                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5956         mtx_unlock(&lun->lun_lock);
5957         if (set_ua) {
5958                 ctl_isc_announce_mode(lun,
5959                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5960                     page_index->page_code, page_index->subpage);
5961         }
5962         return (CTL_RETVAL_COMPLETE);
5963 }
5964
5965 static void
5966 ctl_ie_timer(void *arg)
5967 {
5968         struct ctl_lun *lun = arg;
5969         uint64_t t;
5970
5971         if (lun->ie_asc == 0)
5972                 return;
5973
5974         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5975                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5976         else
5977                 lun->ie_reported = 0;
5978
5979         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5980                 lun->ie_reportcnt++;
5981                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5982                 if (t == 0 || t == UINT32_MAX)
5983                         t = 3000;  /* 5 min */
5984                 callout_schedule(&lun->ie_callout, t * hz / 10);
5985         }
5986 }
5987
5988 int
5989 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5990                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5991 {
5992         struct ctl_lun *lun = CTL_LUN(ctsio);
5993         struct scsi_info_exceptions_page *pg;
5994         uint64_t t;
5995
5996         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5997
5998         pg = (struct scsi_info_exceptions_page *)page_ptr;
5999         mtx_lock(&lun->lun_lock);
6000         if (pg->info_flags & SIEP_FLAGS_TEST) {
6001                 lun->ie_asc = 0x5d;
6002                 lun->ie_ascq = 0xff;
6003                 if (pg->mrie == SIEP_MRIE_UA) {
6004                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
6005                         lun->ie_reported = 1;
6006                 } else {
6007                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
6008                         lun->ie_reported = -1;
6009                 }
6010                 lun->ie_reportcnt = 1;
6011                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
6012                         lun->ie_reportcnt++;
6013                         t = scsi_4btoul(pg->interval_timer);
6014                         if (t == 0 || t == UINT32_MAX)
6015                                 t = 3000;  /* 5 min */
6016                         callout_reset(&lun->ie_callout, t * hz / 10,
6017                             ctl_ie_timer, lun);
6018                 }
6019         } else {
6020                 lun->ie_asc = 0;
6021                 lun->ie_ascq = 0;
6022                 lun->ie_reported = 1;
6023                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
6024                 lun->ie_reportcnt = UINT32_MAX;
6025                 callout_stop(&lun->ie_callout);
6026         }
6027         mtx_unlock(&lun->lun_lock);
6028         return (CTL_RETVAL_COMPLETE);
6029 }
6030
6031 static int
6032 ctl_do_mode_select(union ctl_io *io)
6033 {
6034         struct ctl_lun *lun = CTL_LUN(io);
6035         struct scsi_mode_page_header *page_header;
6036         struct ctl_page_index *page_index;
6037         struct ctl_scsiio *ctsio;
6038         int page_len, page_len_offset, page_len_size;
6039         union ctl_modepage_info *modepage_info;
6040         uint16_t *len_left, *len_used;
6041         int retval, i;
6042
6043         ctsio = &io->scsiio;
6044         page_index = NULL;
6045         page_len = 0;
6046
6047         modepage_info = (union ctl_modepage_info *)
6048                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6049         len_left = &modepage_info->header.len_left;
6050         len_used = &modepage_info->header.len_used;
6051
6052 do_next_page:
6053
6054         page_header = (struct scsi_mode_page_header *)
6055                 (ctsio->kern_data_ptr + *len_used);
6056
6057         if (*len_left == 0) {
6058                 free(ctsio->kern_data_ptr, M_CTL);
6059                 ctl_set_success(ctsio);
6060                 ctl_done((union ctl_io *)ctsio);
6061                 return (CTL_RETVAL_COMPLETE);
6062         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6063
6064                 free(ctsio->kern_data_ptr, M_CTL);
6065                 ctl_set_param_len_error(ctsio);
6066                 ctl_done((union ctl_io *)ctsio);
6067                 return (CTL_RETVAL_COMPLETE);
6068
6069         } else if ((page_header->page_code & SMPH_SPF)
6070                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6071
6072                 free(ctsio->kern_data_ptr, M_CTL);
6073                 ctl_set_param_len_error(ctsio);
6074                 ctl_done((union ctl_io *)ctsio);
6075                 return (CTL_RETVAL_COMPLETE);
6076         }
6077
6078
6079         /*
6080          * XXX KDM should we do something with the block descriptor?
6081          */
6082         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6083                 page_index = &lun->mode_pages.index[i];
6084                 if (lun->be_lun->lun_type == T_DIRECT &&
6085                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6086                         continue;
6087                 if (lun->be_lun->lun_type == T_PROCESSOR &&
6088                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6089                         continue;
6090                 if (lun->be_lun->lun_type == T_CDROM &&
6091                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6092                         continue;
6093
6094                 if ((page_index->page_code & SMPH_PC_MASK) !=
6095                     (page_header->page_code & SMPH_PC_MASK))
6096                         continue;
6097
6098                 /*
6099                  * If neither page has a subpage code, then we've got a
6100                  * match.
6101                  */
6102                 if (((page_index->page_code & SMPH_SPF) == 0)
6103                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6104                         page_len = page_header->page_length;
6105                         break;
6106                 }
6107
6108                 /*
6109                  * If both pages have subpages, then the subpage numbers
6110                  * have to match.
6111                  */
6112                 if ((page_index->page_code & SMPH_SPF)
6113                   && (page_header->page_code & SMPH_SPF)) {
6114                         struct scsi_mode_page_header_sp *sph;
6115
6116                         sph = (struct scsi_mode_page_header_sp *)page_header;
6117                         if (page_index->subpage == sph->subpage) {
6118                                 page_len = scsi_2btoul(sph->page_length);
6119                                 break;
6120                         }
6121                 }
6122         }
6123
6124         /*
6125          * If we couldn't find the page, or if we don't have a mode select
6126          * handler for it, send back an error to the user.
6127          */
6128         if ((i >= CTL_NUM_MODE_PAGES)
6129          || (page_index->select_handler == NULL)) {
6130                 ctl_set_invalid_field(ctsio,
6131                                       /*sks_valid*/ 1,
6132                                       /*command*/ 0,
6133                                       /*field*/ *len_used,
6134                                       /*bit_valid*/ 0,
6135                                       /*bit*/ 0);
6136                 free(ctsio->kern_data_ptr, M_CTL);
6137                 ctl_done((union ctl_io *)ctsio);
6138                 return (CTL_RETVAL_COMPLETE);
6139         }
6140
6141         if (page_index->page_code & SMPH_SPF) {
6142                 page_len_offset = 2;
6143                 page_len_size = 2;
6144         } else {
6145                 page_len_size = 1;
6146                 page_len_offset = 1;
6147         }
6148
6149         /*
6150          * If the length the initiator gives us isn't the one we specify in
6151          * the mode page header, or if they didn't specify enough data in
6152          * the CDB to avoid truncating this page, kick out the request.
6153          */
6154         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6155                 ctl_set_invalid_field(ctsio,
6156                                       /*sks_valid*/ 1,
6157                                       /*command*/ 0,
6158                                       /*field*/ *len_used + page_len_offset,
6159                                       /*bit_valid*/ 0,
6160                                       /*bit*/ 0);
6161                 free(ctsio->kern_data_ptr, M_CTL);
6162                 ctl_done((union ctl_io *)ctsio);
6163                 return (CTL_RETVAL_COMPLETE);
6164         }
6165         if (*len_left < page_index->page_len) {
6166                 free(ctsio->kern_data_ptr, M_CTL);
6167                 ctl_set_param_len_error(ctsio);
6168                 ctl_done((union ctl_io *)ctsio);
6169                 return (CTL_RETVAL_COMPLETE);
6170         }
6171
6172         /*
6173          * Run through the mode page, checking to make sure that the bits
6174          * the user changed are actually legal for him to change.
6175          */
6176         for (i = 0; i < page_index->page_len; i++) {
6177                 uint8_t *user_byte, *change_mask, *current_byte;
6178                 int bad_bit;
6179                 int j;
6180
6181                 user_byte = (uint8_t *)page_header + i;
6182                 change_mask = page_index->page_data +
6183                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6184                 current_byte = page_index->page_data +
6185                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6186
6187                 /*
6188                  * Check to see whether the user set any bits in this byte
6189                  * that he is not allowed to set.
6190                  */
6191                 if ((*user_byte & ~(*change_mask)) ==
6192                     (*current_byte & ~(*change_mask)))
6193                         continue;
6194
6195                 /*
6196                  * Go through bit by bit to determine which one is illegal.
6197                  */
6198                 bad_bit = 0;
6199                 for (j = 7; j >= 0; j--) {
6200                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6201                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6202                                 bad_bit = i;
6203                                 break;
6204                         }
6205                 }
6206                 ctl_set_invalid_field(ctsio,
6207                                       /*sks_valid*/ 1,
6208                                       /*command*/ 0,
6209                                       /*field*/ *len_used + i,
6210                                       /*bit_valid*/ 1,
6211                                       /*bit*/ bad_bit);
6212                 free(ctsio->kern_data_ptr, M_CTL);
6213                 ctl_done((union ctl_io *)ctsio);
6214                 return (CTL_RETVAL_COMPLETE);
6215         }
6216
6217         /*
6218          * Decrement these before we call the page handler, since we may
6219          * end up getting called back one way or another before the handler
6220          * returns to this context.
6221          */
6222         *len_left -= page_index->page_len;
6223         *len_used += page_index->page_len;
6224
6225         retval = page_index->select_handler(ctsio, page_index,
6226                                             (uint8_t *)page_header);
6227
6228         /*
6229          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6230          * wait until this queued command completes to finish processing
6231          * the mode page.  If it returns anything other than
6232          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6233          * already set the sense information, freed the data pointer, and
6234          * completed the io for us.
6235          */
6236         if (retval != CTL_RETVAL_COMPLETE)
6237                 goto bailout_no_done;
6238
6239         /*
6240          * If the initiator sent us more than one page, parse the next one.
6241          */
6242         if (*len_left > 0)
6243                 goto do_next_page;
6244
6245         ctl_set_success(ctsio);
6246         free(ctsio->kern_data_ptr, M_CTL);
6247         ctl_done((union ctl_io *)ctsio);
6248
6249 bailout_no_done:
6250
6251         return (CTL_RETVAL_COMPLETE);
6252
6253 }
6254
6255 int
6256 ctl_mode_select(struct ctl_scsiio *ctsio)
6257 {
6258         struct ctl_lun *lun = CTL_LUN(ctsio);
6259         union ctl_modepage_info *modepage_info;
6260         int bd_len, i, header_size, param_len, pf, rtd, sp;
6261         uint32_t initidx;
6262
6263         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6264         switch (ctsio->cdb[0]) {
6265         case MODE_SELECT_6: {
6266                 struct scsi_mode_select_6 *cdb;
6267
6268                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6269
6270                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6271                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6272                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6273                 param_len = cdb->length;
6274                 header_size = sizeof(struct scsi_mode_header_6);
6275                 break;
6276         }
6277         case MODE_SELECT_10: {
6278                 struct scsi_mode_select_10 *cdb;
6279
6280                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6281
6282                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6283                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6284                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6285                 param_len = scsi_2btoul(cdb->length);
6286                 header_size = sizeof(struct scsi_mode_header_10);
6287                 break;
6288         }
6289         default:
6290                 ctl_set_invalid_opcode(ctsio);
6291                 ctl_done((union ctl_io *)ctsio);
6292                 return (CTL_RETVAL_COMPLETE);
6293         }
6294
6295         if (rtd) {
6296                 if (param_len != 0) {
6297                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 0,
6298                             /*command*/ 1, /*field*/ 0,
6299                             /*bit_valid*/ 0, /*bit*/ 0);
6300                         ctl_done((union ctl_io *)ctsio);
6301                         return (CTL_RETVAL_COMPLETE);
6302                 }
6303
6304                 /* Revert to defaults. */
6305                 ctl_init_page_index(lun);
6306                 mtx_lock(&lun->lun_lock);
6307                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6308                 mtx_unlock(&lun->lun_lock);
6309                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6310                         ctl_isc_announce_mode(lun, -1,
6311                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
6312                             lun->mode_pages.index[i].subpage);
6313                 }
6314                 ctl_set_success(ctsio);
6315                 ctl_done((union ctl_io *)ctsio);
6316                 return (CTL_RETVAL_COMPLETE);
6317         }
6318
6319         /*
6320          * From SPC-3:
6321          * "A parameter list length of zero indicates that the Data-Out Buffer
6322          * shall be empty. This condition shall not be considered as an error."
6323          */
6324         if (param_len == 0) {
6325                 ctl_set_success(ctsio);
6326                 ctl_done((union ctl_io *)ctsio);
6327                 return (CTL_RETVAL_COMPLETE);
6328         }
6329
6330         /*
6331          * Since we'll hit this the first time through, prior to
6332          * allocation, we don't need to free a data buffer here.
6333          */
6334         if (param_len < header_size) {
6335                 ctl_set_param_len_error(ctsio);
6336                 ctl_done((union ctl_io *)ctsio);
6337                 return (CTL_RETVAL_COMPLETE);
6338         }
6339
6340         /*
6341          * Allocate the data buffer and grab the user's data.  In theory,
6342          * we shouldn't have to sanity check the parameter list length here
6343          * because the maximum size is 64K.  We should be able to malloc
6344          * that much without too many problems.
6345          */
6346         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6347                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6348                 ctsio->kern_data_len = param_len;
6349                 ctsio->kern_total_len = param_len;
6350                 ctsio->kern_rel_offset = 0;
6351                 ctsio->kern_sg_entries = 0;
6352                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6353                 ctsio->be_move_done = ctl_config_move_done;
6354                 ctl_datamove((union ctl_io *)ctsio);
6355
6356                 return (CTL_RETVAL_COMPLETE);
6357         }
6358
6359         switch (ctsio->cdb[0]) {
6360         case MODE_SELECT_6: {
6361                 struct scsi_mode_header_6 *mh6;
6362
6363                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6364                 bd_len = mh6->blk_desc_len;
6365                 break;
6366         }
6367         case MODE_SELECT_10: {
6368                 struct scsi_mode_header_10 *mh10;
6369
6370                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6371                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6372                 break;
6373         }
6374         default:
6375                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6376         }
6377
6378         if (param_len < (header_size + bd_len)) {
6379                 free(ctsio->kern_data_ptr, M_CTL);
6380                 ctl_set_param_len_error(ctsio);
6381                 ctl_done((union ctl_io *)ctsio);
6382                 return (CTL_RETVAL_COMPLETE);
6383         }
6384
6385         /*
6386          * Set the IO_CONT flag, so that if this I/O gets passed to
6387          * ctl_config_write_done(), it'll get passed back to
6388          * ctl_do_mode_select() for further processing, or completion if
6389          * we're all done.
6390          */
6391         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6392         ctsio->io_cont = ctl_do_mode_select;
6393
6394         modepage_info = (union ctl_modepage_info *)
6395                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6396         memset(modepage_info, 0, sizeof(*modepage_info));
6397         modepage_info->header.len_left = param_len - header_size - bd_len;
6398         modepage_info->header.len_used = header_size + bd_len;
6399
6400         return (ctl_do_mode_select((union ctl_io *)ctsio));
6401 }
6402
6403 int
6404 ctl_mode_sense(struct ctl_scsiio *ctsio)
6405 {
6406         struct ctl_lun *lun = CTL_LUN(ctsio);
6407         int pc, page_code, dbd, llba, subpage;
6408         int alloc_len, page_len, header_len, total_len;
6409         struct scsi_mode_block_descr *block_desc;
6410         struct ctl_page_index *page_index;
6411
6412         dbd = 0;
6413         llba = 0;
6414         block_desc = NULL;
6415
6416         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6417
6418         switch (ctsio->cdb[0]) {
6419         case MODE_SENSE_6: {
6420                 struct scsi_mode_sense_6 *cdb;
6421
6422                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6423
6424                 header_len = sizeof(struct scsi_mode_hdr_6);
6425                 if (cdb->byte2 & SMS_DBD)
6426                         dbd = 1;
6427                 else
6428                         header_len += sizeof(struct scsi_mode_block_descr);
6429
6430                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6431                 page_code = cdb->page & SMS_PAGE_CODE;
6432                 subpage = cdb->subpage;
6433                 alloc_len = cdb->length;
6434                 break;
6435         }
6436         case MODE_SENSE_10: {
6437                 struct scsi_mode_sense_10 *cdb;
6438
6439                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6440
6441                 header_len = sizeof(struct scsi_mode_hdr_10);
6442
6443                 if (cdb->byte2 & SMS_DBD)
6444                         dbd = 1;
6445                 else
6446                         header_len += sizeof(struct scsi_mode_block_descr);
6447                 if (cdb->byte2 & SMS10_LLBAA)
6448                         llba = 1;
6449                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6450                 page_code = cdb->page & SMS_PAGE_CODE;
6451                 subpage = cdb->subpage;
6452                 alloc_len = scsi_2btoul(cdb->length);
6453                 break;
6454         }
6455         default:
6456                 ctl_set_invalid_opcode(ctsio);
6457                 ctl_done((union ctl_io *)ctsio);
6458                 return (CTL_RETVAL_COMPLETE);
6459                 break; /* NOTREACHED */
6460         }
6461
6462         /*
6463          * We have to make a first pass through to calculate the size of
6464          * the pages that match the user's query.  Then we allocate enough
6465          * memory to hold it, and actually copy the data into the buffer.
6466          */
6467         switch (page_code) {
6468         case SMS_ALL_PAGES_PAGE: {
6469                 u_int i;
6470
6471                 page_len = 0;
6472
6473                 /*
6474                  * At the moment, values other than 0 and 0xff here are
6475                  * reserved according to SPC-3.
6476                  */
6477                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6478                  && (subpage != SMS_SUBPAGE_ALL)) {
6479                         ctl_set_invalid_field(ctsio,
6480                                               /*sks_valid*/ 1,
6481                                               /*command*/ 1,
6482                                               /*field*/ 3,
6483                                               /*bit_valid*/ 0,
6484                                               /*bit*/ 0);
6485                         ctl_done((union ctl_io *)ctsio);
6486                         return (CTL_RETVAL_COMPLETE);
6487                 }
6488
6489                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6490                         page_index = &lun->mode_pages.index[i];
6491
6492                         /* Make sure the page is supported for this dev type */
6493                         if (lun->be_lun->lun_type == T_DIRECT &&
6494                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6495                                 continue;
6496                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6497                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6498                                 continue;
6499                         if (lun->be_lun->lun_type == T_CDROM &&
6500                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6501                                 continue;
6502
6503                         /*
6504                          * We don't use this subpage if the user didn't
6505                          * request all subpages.
6506                          */
6507                         if ((page_index->subpage != 0)
6508                          && (subpage == SMS_SUBPAGE_PAGE_0))
6509                                 continue;
6510
6511 #if 0
6512                         printf("found page %#x len %d\n",
6513                                page_index->page_code & SMPH_PC_MASK,
6514                                page_index->page_len);
6515 #endif
6516                         page_len += page_index->page_len;
6517                 }
6518                 break;
6519         }
6520         default: {
6521                 u_int i;
6522
6523                 page_len = 0;
6524
6525                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6526                         page_index = &lun->mode_pages.index[i];
6527
6528                         /* Make sure the page is supported for this dev type */
6529                         if (lun->be_lun->lun_type == T_DIRECT &&
6530                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6531                                 continue;
6532                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6533                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6534                                 continue;
6535                         if (lun->be_lun->lun_type == T_CDROM &&
6536                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6537                                 continue;
6538
6539                         /* Look for the right page code */
6540                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6541                                 continue;
6542
6543                         /* Look for the right subpage or the subpage wildcard*/
6544                         if ((page_index->subpage != subpage)
6545                          && (subpage != SMS_SUBPAGE_ALL))
6546                                 continue;
6547
6548 #if 0
6549                         printf("found page %#x len %d\n",
6550                                page_index->page_code & SMPH_PC_MASK,
6551                                page_index->page_len);
6552 #endif
6553
6554                         page_len += page_index->page_len;
6555                 }
6556
6557                 if (page_len == 0) {
6558                         ctl_set_invalid_field(ctsio,
6559                                               /*sks_valid*/ 1,
6560                                               /*command*/ 1,
6561                                               /*field*/ 2,
6562                                               /*bit_valid*/ 1,
6563                                               /*bit*/ 5);
6564                         ctl_done((union ctl_io *)ctsio);
6565                         return (CTL_RETVAL_COMPLETE);
6566                 }
6567                 break;
6568         }
6569         }
6570
6571         total_len = header_len + page_len;
6572 #if 0
6573         printf("header_len = %d, page_len = %d, total_len = %d\n",
6574                header_len, page_len, total_len);
6575 #endif
6576
6577         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6578         ctsio->kern_sg_entries = 0;
6579         ctsio->kern_rel_offset = 0;
6580         ctsio->kern_data_len = min(total_len, alloc_len);
6581         ctsio->kern_total_len = ctsio->kern_data_len;
6582
6583         switch (ctsio->cdb[0]) {
6584         case MODE_SENSE_6: {
6585                 struct scsi_mode_hdr_6 *header;
6586
6587                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6588
6589                 header->datalen = MIN(total_len - 1, 254);
6590                 if (lun->be_lun->lun_type == T_DIRECT) {
6591                         header->dev_specific = 0x10; /* DPOFUA */
6592                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6593                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6594                                 header->dev_specific |= 0x80; /* WP */
6595                 }
6596                 if (dbd)
6597                         header->block_descr_len = 0;
6598                 else
6599                         header->block_descr_len =
6600                                 sizeof(struct scsi_mode_block_descr);
6601                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6602                 break;
6603         }
6604         case MODE_SENSE_10: {
6605                 struct scsi_mode_hdr_10 *header;
6606                 int datalen;
6607
6608                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6609
6610                 datalen = MIN(total_len - 2, 65533);
6611                 scsi_ulto2b(datalen, header->datalen);
6612                 if (lun->be_lun->lun_type == T_DIRECT) {
6613                         header->dev_specific = 0x10; /* DPOFUA */
6614                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6615                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6616                                 header->dev_specific |= 0x80; /* WP */
6617                 }
6618                 if (dbd)
6619                         scsi_ulto2b(0, header->block_descr_len);
6620                 else
6621                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6622                                     header->block_descr_len);
6623                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6624                 break;
6625         }
6626         default:
6627                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6628         }
6629
6630         /*
6631          * If we've got a disk, use its blocksize in the block
6632          * descriptor.  Otherwise, just set it to 0.
6633          */
6634         if (dbd == 0) {
6635                 if (lun->be_lun->lun_type == T_DIRECT)
6636                         scsi_ulto3b(lun->be_lun->blocksize,
6637                                     block_desc->block_len);
6638                 else
6639                         scsi_ulto3b(0, block_desc->block_len);
6640         }
6641
6642         switch (page_code) {
6643         case SMS_ALL_PAGES_PAGE: {
6644                 int i, data_used;
6645
6646                 data_used = header_len;
6647                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6648                         struct ctl_page_index *page_index;
6649
6650                         page_index = &lun->mode_pages.index[i];
6651                         if (lun->be_lun->lun_type == T_DIRECT &&
6652                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6653                                 continue;
6654                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6655                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6656                                 continue;
6657                         if (lun->be_lun->lun_type == T_CDROM &&
6658                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6659                                 continue;
6660
6661                         /*
6662                          * We don't use this subpage if the user didn't
6663                          * request all subpages.  We already checked (above)
6664                          * to make sure the user only specified a subpage
6665                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6666                          */
6667                         if ((page_index->subpage != 0)
6668                          && (subpage == SMS_SUBPAGE_PAGE_0))
6669                                 continue;
6670
6671                         /*
6672                          * Call the handler, if it exists, to update the
6673                          * page to the latest values.
6674                          */
6675                         if (page_index->sense_handler != NULL)
6676                                 page_index->sense_handler(ctsio, page_index,pc);
6677
6678                         memcpy(ctsio->kern_data_ptr + data_used,
6679                                page_index->page_data +
6680                                (page_index->page_len * pc),
6681                                page_index->page_len);
6682                         data_used += page_index->page_len;
6683                 }
6684                 break;
6685         }
6686         default: {
6687                 int i, data_used;
6688
6689                 data_used = header_len;
6690
6691                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6692                         struct ctl_page_index *page_index;
6693
6694                         page_index = &lun->mode_pages.index[i];
6695
6696                         /* Look for the right page code */
6697                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6698                                 continue;
6699
6700                         /* Look for the right subpage or the subpage wildcard*/
6701                         if ((page_index->subpage != subpage)
6702                          && (subpage != SMS_SUBPAGE_ALL))
6703                                 continue;
6704
6705                         /* Make sure the page is supported for this dev type */
6706                         if (lun->be_lun->lun_type == T_DIRECT &&
6707                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6708                                 continue;
6709                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6710                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6711                                 continue;
6712                         if (lun->be_lun->lun_type == T_CDROM &&
6713                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6714                                 continue;
6715
6716                         /*
6717                          * Call the handler, if it exists, to update the
6718                          * page to the latest values.
6719                          */
6720                         if (page_index->sense_handler != NULL)
6721                                 page_index->sense_handler(ctsio, page_index,pc);
6722
6723                         memcpy(ctsio->kern_data_ptr + data_used,
6724                                page_index->page_data +
6725                                (page_index->page_len * pc),
6726                                page_index->page_len);
6727                         data_used += page_index->page_len;
6728                 }
6729                 break;
6730         }
6731         }
6732
6733         ctl_set_success(ctsio);
6734         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6735         ctsio->be_move_done = ctl_config_move_done;
6736         ctl_datamove((union ctl_io *)ctsio);
6737         return (CTL_RETVAL_COMPLETE);
6738 }
6739
6740 int
6741 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6742                                struct ctl_page_index *page_index,
6743                                int pc)
6744 {
6745         struct ctl_lun *lun = CTL_LUN(ctsio);
6746         struct scsi_log_param_header *phdr;
6747         uint8_t *data;
6748         uint64_t val;
6749
6750         data = page_index->page_data;
6751
6752         if (lun->backend->lun_attr != NULL &&
6753             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6754              != UINT64_MAX) {
6755                 phdr = (struct scsi_log_param_header *)data;
6756                 scsi_ulto2b(0x0001, phdr->param_code);
6757                 phdr->param_control = SLP_LBIN | SLP_LP;
6758                 phdr->param_len = 8;
6759                 data = (uint8_t *)(phdr + 1);
6760                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6761                 data[4] = 0x02; /* per-pool */
6762                 data += phdr->param_len;
6763         }
6764
6765         if (lun->backend->lun_attr != NULL &&
6766             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6767              != UINT64_MAX) {
6768                 phdr = (struct scsi_log_param_header *)data;
6769                 scsi_ulto2b(0x0002, phdr->param_code);
6770                 phdr->param_control = SLP_LBIN | SLP_LP;
6771                 phdr->param_len = 8;
6772                 data = (uint8_t *)(phdr + 1);
6773                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6774                 data[4] = 0x01; /* per-LUN */
6775                 data += phdr->param_len;
6776         }
6777
6778         if (lun->backend->lun_attr != NULL &&
6779             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6780              != UINT64_MAX) {
6781                 phdr = (struct scsi_log_param_header *)data;
6782                 scsi_ulto2b(0x00f1, phdr->param_code);
6783                 phdr->param_control = SLP_LBIN | SLP_LP;
6784                 phdr->param_len = 8;
6785                 data = (uint8_t *)(phdr + 1);
6786                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6787                 data[4] = 0x02; /* per-pool */
6788                 data += phdr->param_len;
6789         }
6790
6791         if (lun->backend->lun_attr != NULL &&
6792             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6793              != UINT64_MAX) {
6794                 phdr = (struct scsi_log_param_header *)data;
6795                 scsi_ulto2b(0x00f2, phdr->param_code);
6796                 phdr->param_control = SLP_LBIN | SLP_LP;
6797                 phdr->param_len = 8;
6798                 data = (uint8_t *)(phdr + 1);
6799                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6800                 data[4] = 0x02; /* per-pool */
6801                 data += phdr->param_len;
6802         }
6803
6804         page_index->page_len = data - page_index->page_data;
6805         return (0);
6806 }
6807
6808 int
6809 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6810                                struct ctl_page_index *page_index,
6811                                int pc)
6812 {
6813         struct ctl_lun *lun = CTL_LUN(ctsio);
6814         struct stat_page *data;
6815         struct bintime *t;
6816
6817         data = (struct stat_page *)page_index->page_data;
6818
6819         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6820         data->sap.hdr.param_control = SLP_LBIN;
6821         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6822             sizeof(struct scsi_log_param_header);
6823         scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
6824             data->sap.read_num);
6825         scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
6826             data->sap.write_num);
6827         if (lun->be_lun->blocksize > 0) {
6828                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
6829                     lun->be_lun->blocksize, data->sap.recvieved_lba);
6830                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
6831                     lun->be_lun->blocksize, data->sap.transmitted_lba);
6832         }
6833         t = &lun->stats.time[CTL_STATS_READ];
6834         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6835             data->sap.read_int);
6836         t = &lun->stats.time[CTL_STATS_WRITE];
6837         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6838             data->sap.write_int);
6839         scsi_u64to8b(0, data->sap.weighted_num);
6840         scsi_u64to8b(0, data->sap.weighted_int);
6841         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6842         data->it.hdr.param_control = SLP_LBIN;
6843         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6844             sizeof(struct scsi_log_param_header);
6845 #ifdef CTL_TIME_IO
6846         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6847 #endif
6848         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6849         data->it.hdr.param_control = SLP_LBIN;
6850         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6851             sizeof(struct scsi_log_param_header);
6852         scsi_ulto4b(3, data->ti.exponent);
6853         scsi_ulto4b(1, data->ti.integer);
6854         return (0);
6855 }
6856
6857 int
6858 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6859                                struct ctl_page_index *page_index,
6860                                int pc)
6861 {
6862         struct ctl_lun *lun = CTL_LUN(ctsio);
6863         struct scsi_log_informational_exceptions *data;
6864
6865         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6866
6867         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6868         data->hdr.param_control = SLP_LBIN;
6869         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6870             sizeof(struct scsi_log_param_header);
6871         data->ie_asc = lun->ie_asc;
6872         data->ie_ascq = lun->ie_ascq;
6873         data->temperature = 0xff;
6874         return (0);
6875 }
6876
6877 int
6878 ctl_log_sense(struct ctl_scsiio *ctsio)
6879 {
6880         struct ctl_lun *lun = CTL_LUN(ctsio);
6881         int i, pc, page_code, subpage;
6882         int alloc_len, total_len;
6883         struct ctl_page_index *page_index;
6884         struct scsi_log_sense *cdb;
6885         struct scsi_log_header *header;
6886
6887         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6888
6889         cdb = (struct scsi_log_sense *)ctsio->cdb;
6890         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6891         page_code = cdb->page & SLS_PAGE_CODE;
6892         subpage = cdb->subpage;
6893         alloc_len = scsi_2btoul(cdb->length);
6894
6895         page_index = NULL;
6896         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6897                 page_index = &lun->log_pages.index[i];
6898
6899                 /* Look for the right page code */
6900                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6901                         continue;
6902
6903                 /* Look for the right subpage or the subpage wildcard*/
6904                 if (page_index->subpage != subpage)
6905                         continue;
6906
6907                 break;
6908         }
6909         if (i >= CTL_NUM_LOG_PAGES) {
6910                 ctl_set_invalid_field(ctsio,
6911                                       /*sks_valid*/ 1,
6912                                       /*command*/ 1,
6913                                       /*field*/ 2,
6914                                       /*bit_valid*/ 0,
6915                                       /*bit*/ 0);
6916                 ctl_done((union ctl_io *)ctsio);
6917                 return (CTL_RETVAL_COMPLETE);
6918         }
6919
6920         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6921
6922         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6923         ctsio->kern_sg_entries = 0;
6924         ctsio->kern_rel_offset = 0;
6925         ctsio->kern_data_len = min(total_len, alloc_len);
6926         ctsio->kern_total_len = ctsio->kern_data_len;
6927
6928         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6929         header->page = page_index->page_code;
6930         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6931                 header->page |= SL_DS;
6932         if (page_index->subpage) {
6933                 header->page |= SL_SPF;
6934                 header->subpage = page_index->subpage;
6935         }
6936         scsi_ulto2b(page_index->page_len, header->datalen);
6937
6938         /*
6939          * Call the handler, if it exists, to update the
6940          * page to the latest values.
6941          */
6942         if (page_index->sense_handler != NULL)
6943                 page_index->sense_handler(ctsio, page_index, pc);
6944
6945         memcpy(header + 1, page_index->page_data, page_index->page_len);
6946
6947         ctl_set_success(ctsio);
6948         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6949         ctsio->be_move_done = ctl_config_move_done;
6950         ctl_datamove((union ctl_io *)ctsio);
6951         return (CTL_RETVAL_COMPLETE);
6952 }
6953
6954 int
6955 ctl_read_capacity(struct ctl_scsiio *ctsio)
6956 {
6957         struct ctl_lun *lun = CTL_LUN(ctsio);
6958         struct scsi_read_capacity *cdb;
6959         struct scsi_read_capacity_data *data;
6960         uint32_t lba;
6961
6962         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6963
6964         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6965
6966         lba = scsi_4btoul(cdb->addr);
6967         if (((cdb->pmi & SRC_PMI) == 0)
6968          && (lba != 0)) {
6969                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6970                                       /*sks_valid*/ 1,
6971                                       /*command*/ 1,
6972                                       /*field*/ 2,
6973                                       /*bit_valid*/ 0,
6974                                       /*bit*/ 0);
6975                 ctl_done((union ctl_io *)ctsio);
6976                 return (CTL_RETVAL_COMPLETE);
6977         }
6978
6979         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6980         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6981         ctsio->kern_data_len = sizeof(*data);
6982         ctsio->kern_total_len = sizeof(*data);
6983         ctsio->kern_rel_offset = 0;
6984         ctsio->kern_sg_entries = 0;
6985
6986         /*
6987          * If the maximum LBA is greater than 0xfffffffe, the user must
6988          * issue a SERVICE ACTION IN (16) command, with the read capacity
6989          * serivce action set.
6990          */
6991         if (lun->be_lun->maxlba > 0xfffffffe)
6992                 scsi_ulto4b(0xffffffff, data->addr);
6993         else
6994                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6995
6996         /*
6997          * XXX KDM this may not be 512 bytes...
6998          */
6999         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7000
7001         ctl_set_success(ctsio);
7002         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7003         ctsio->be_move_done = ctl_config_move_done;
7004         ctl_datamove((union ctl_io *)ctsio);
7005         return (CTL_RETVAL_COMPLETE);
7006 }
7007
7008 int
7009 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7010 {
7011         struct ctl_lun *lun = CTL_LUN(ctsio);
7012         struct scsi_read_capacity_16 *cdb;
7013         struct scsi_read_capacity_data_long *data;
7014         uint64_t lba;
7015         uint32_t alloc_len;
7016
7017         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7018
7019         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7020
7021         alloc_len = scsi_4btoul(cdb->alloc_len);
7022         lba = scsi_8btou64(cdb->addr);
7023
7024         if ((cdb->reladr & SRC16_PMI)
7025          && (lba != 0)) {
7026                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7027                                       /*sks_valid*/ 1,
7028                                       /*command*/ 1,
7029                                       /*field*/ 2,
7030                                       /*bit_valid*/ 0,
7031                                       /*bit*/ 0);
7032                 ctl_done((union ctl_io *)ctsio);
7033                 return (CTL_RETVAL_COMPLETE);
7034         }
7035
7036         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7037         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7038         ctsio->kern_rel_offset = 0;
7039         ctsio->kern_sg_entries = 0;
7040         ctsio->kern_data_len = min(sizeof(*data), alloc_len);
7041         ctsio->kern_total_len = ctsio->kern_data_len;
7042
7043         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7044         /* XXX KDM this may not be 512 bytes... */
7045         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7046         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7047         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7048         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7049                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7050
7051         ctl_set_success(ctsio);
7052         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7053         ctsio->be_move_done = ctl_config_move_done;
7054         ctl_datamove((union ctl_io *)ctsio);
7055         return (CTL_RETVAL_COMPLETE);
7056 }
7057
7058 int
7059 ctl_get_lba_status(struct ctl_scsiio *ctsio)
7060 {
7061         struct ctl_lun *lun = CTL_LUN(ctsio);
7062         struct scsi_get_lba_status *cdb;
7063         struct scsi_get_lba_status_data *data;
7064         struct ctl_lba_len_flags *lbalen;
7065         uint64_t lba;
7066         uint32_t alloc_len, total_len;
7067         int retval;
7068
7069         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7070
7071         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7072         lba = scsi_8btou64(cdb->addr);
7073         alloc_len = scsi_4btoul(cdb->alloc_len);
7074
7075         if (lba > lun->be_lun->maxlba) {
7076                 ctl_set_lba_out_of_range(ctsio, lba);
7077                 ctl_done((union ctl_io *)ctsio);
7078                 return (CTL_RETVAL_COMPLETE);
7079         }
7080
7081         total_len = sizeof(*data) + sizeof(data->descr[0]);
7082         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7083         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7084         ctsio->kern_rel_offset = 0;
7085         ctsio->kern_sg_entries = 0;
7086         ctsio->kern_data_len = min(total_len, alloc_len);
7087         ctsio->kern_total_len = ctsio->kern_data_len;
7088
7089         /* Fill dummy data in case backend can't tell anything. */
7090         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7091         scsi_u64to8b(lba, data->descr[0].addr);
7092         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7093             data->descr[0].length);
7094         data->descr[0].status = 0; /* Mapped or unknown. */
7095
7096         ctl_set_success(ctsio);
7097         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7098         ctsio->be_move_done = ctl_config_move_done;
7099
7100         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7101         lbalen->lba = lba;
7102         lbalen->len = total_len;
7103         lbalen->flags = 0;
7104         retval = lun->backend->config_read((union ctl_io *)ctsio);
7105         return (retval);
7106 }
7107
7108 int
7109 ctl_read_defect(struct ctl_scsiio *ctsio)
7110 {
7111         struct scsi_read_defect_data_10 *ccb10;
7112         struct scsi_read_defect_data_12 *ccb12;
7113         struct scsi_read_defect_data_hdr_10 *data10;
7114         struct scsi_read_defect_data_hdr_12 *data12;
7115         uint32_t alloc_len, data_len;
7116         uint8_t format;
7117
7118         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7119
7120         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7121                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7122                 format = ccb10->format;
7123                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7124                 data_len = sizeof(*data10);
7125         } else {
7126                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7127                 format = ccb12->format;
7128                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7129                 data_len = sizeof(*data12);
7130         }
7131         if (alloc_len == 0) {
7132                 ctl_set_success(ctsio);
7133                 ctl_done((union ctl_io *)ctsio);
7134                 return (CTL_RETVAL_COMPLETE);
7135         }
7136
7137         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7138         ctsio->kern_rel_offset = 0;
7139         ctsio->kern_sg_entries = 0;
7140         ctsio->kern_data_len = min(data_len, alloc_len);
7141         ctsio->kern_total_len = ctsio->kern_data_len;
7142
7143         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7144                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7145                     ctsio->kern_data_ptr;
7146                 data10->format = format;
7147                 scsi_ulto2b(0, data10->length);
7148         } else {
7149                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7150                     ctsio->kern_data_ptr;
7151                 data12->format = format;
7152                 scsi_ulto2b(0, data12->generation);
7153                 scsi_ulto4b(0, data12->length);
7154         }
7155
7156         ctl_set_success(ctsio);
7157         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7158         ctsio->be_move_done = ctl_config_move_done;
7159         ctl_datamove((union ctl_io *)ctsio);
7160         return (CTL_RETVAL_COMPLETE);
7161 }
7162
7163 int
7164 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7165 {
7166         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7167         struct ctl_lun *lun = CTL_LUN(ctsio);
7168         struct scsi_maintenance_in *cdb;
7169         int retval;
7170         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7171         int num_ha_groups, num_target_ports, shared_group;
7172         struct ctl_port *port;
7173         struct scsi_target_group_data *rtg_ptr;
7174         struct scsi_target_group_data_extended *rtg_ext_ptr;
7175         struct scsi_target_port_group_descriptor *tpg_desc;
7176
7177         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7178
7179         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7180         retval = CTL_RETVAL_COMPLETE;
7181
7182         switch (cdb->byte2 & STG_PDF_MASK) {
7183         case STG_PDF_LENGTH:
7184                 ext = 0;
7185                 break;
7186         case STG_PDF_EXTENDED:
7187                 ext = 1;
7188                 break;
7189         default:
7190                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7191                                       /*sks_valid*/ 1,
7192                                       /*command*/ 1,
7193                                       /*field*/ 2,
7194                                       /*bit_valid*/ 1,
7195                                       /*bit*/ 5);
7196                 ctl_done((union ctl_io *)ctsio);
7197                 return(retval);
7198         }
7199
7200         num_target_ports = 0;
7201         shared_group = (softc->is_single != 0);
7202         mtx_lock(&softc->ctl_lock);
7203         STAILQ_FOREACH(port, &softc->port_list, links) {
7204                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7205                         continue;
7206                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7207                         continue;
7208                 num_target_ports++;
7209                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7210                         shared_group = 1;
7211         }
7212         mtx_unlock(&softc->ctl_lock);
7213         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7214
7215         if (ext)
7216                 total_len = sizeof(struct scsi_target_group_data_extended);
7217         else
7218                 total_len = sizeof(struct scsi_target_group_data);
7219         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7220                 (shared_group + num_ha_groups) +
7221             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7222
7223         alloc_len = scsi_4btoul(cdb->length);
7224
7225         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7226         ctsio->kern_sg_entries = 0;
7227         ctsio->kern_rel_offset = 0;
7228         ctsio->kern_data_len = min(total_len, alloc_len);
7229         ctsio->kern_total_len = ctsio->kern_data_len;
7230
7231         if (ext) {
7232                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7233                     ctsio->kern_data_ptr;
7234                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7235                 rtg_ext_ptr->format_type = 0x10;
7236                 rtg_ext_ptr->implicit_transition_time = 0;
7237                 tpg_desc = &rtg_ext_ptr->groups[0];
7238         } else {
7239                 rtg_ptr = (struct scsi_target_group_data *)
7240                     ctsio->kern_data_ptr;
7241                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7242                 tpg_desc = &rtg_ptr->groups[0];
7243         }
7244
7245         mtx_lock(&softc->ctl_lock);
7246         pg = softc->port_min / softc->port_cnt;
7247         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7248                 /* Some shelf is known to be primary. */
7249                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7250                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7251                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7252                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7253                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7254                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7255                 else
7256                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7257                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7258                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7259                 } else {
7260                         ts = os;
7261                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7262                 }
7263         } else {
7264                 /* No known primary shelf. */
7265                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7266                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7267                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7268                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7269                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7270                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7271                 } else {
7272                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7273                 }
7274         }
7275         if (shared_group) {
7276                 tpg_desc->pref_state = ts;
7277                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7278                     TPG_U_SUP | TPG_T_SUP;
7279                 scsi_ulto2b(1, tpg_desc->target_port_group);
7280                 tpg_desc->status = TPG_IMPLICIT;
7281                 pc = 0;
7282                 STAILQ_FOREACH(port, &softc->port_list, links) {
7283                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7284                                 continue;
7285                         if (!softc->is_single &&
7286                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7287                                 continue;
7288                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7289                                 continue;
7290                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7291                             relative_target_port_identifier);
7292                         pc++;
7293                 }
7294                 tpg_desc->target_port_count = pc;
7295                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7296                     &tpg_desc->descriptors[pc];
7297         }
7298         for (g = 0; g < num_ha_groups; g++) {
7299                 tpg_desc->pref_state = (g == pg) ? ts : os;
7300                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7301                     TPG_U_SUP | TPG_T_SUP;
7302                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7303                 tpg_desc->status = TPG_IMPLICIT;
7304                 pc = 0;
7305                 STAILQ_FOREACH(port, &softc->port_list, links) {
7306                         if (port->targ_port < g * softc->port_cnt ||
7307                             port->targ_port >= (g + 1) * softc->port_cnt)
7308                                 continue;
7309                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7310                                 continue;
7311                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7312                                 continue;
7313                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7314                                 continue;
7315                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7316                             relative_target_port_identifier);
7317                         pc++;
7318                 }
7319                 tpg_desc->target_port_count = pc;
7320                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7321                     &tpg_desc->descriptors[pc];
7322         }
7323         mtx_unlock(&softc->ctl_lock);
7324
7325         ctl_set_success(ctsio);
7326         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7327         ctsio->be_move_done = ctl_config_move_done;
7328         ctl_datamove((union ctl_io *)ctsio);
7329         return(retval);
7330 }
7331
7332 int
7333 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7334 {
7335         struct ctl_lun *lun = CTL_LUN(ctsio);
7336         struct scsi_report_supported_opcodes *cdb;
7337         const struct ctl_cmd_entry *entry, *sentry;
7338         struct scsi_report_supported_opcodes_all *all;
7339         struct scsi_report_supported_opcodes_descr *descr;
7340         struct scsi_report_supported_opcodes_one *one;
7341         int retval;
7342         int alloc_len, total_len;
7343         int opcode, service_action, i, j, num;
7344
7345         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7346
7347         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7348         retval = CTL_RETVAL_COMPLETE;
7349
7350         opcode = cdb->requested_opcode;
7351         service_action = scsi_2btoul(cdb->requested_service_action);
7352         switch (cdb->options & RSO_OPTIONS_MASK) {
7353         case RSO_OPTIONS_ALL:
7354                 num = 0;
7355                 for (i = 0; i < 256; i++) {
7356                         entry = &ctl_cmd_table[i];
7357                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7358                                 for (j = 0; j < 32; j++) {
7359                                         sentry = &((const struct ctl_cmd_entry *)
7360                                             entry->execute)[j];
7361                                         if (ctl_cmd_applicable(
7362                                             lun->be_lun->lun_type, sentry))
7363                                                 num++;
7364                                 }
7365                         } else {
7366                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7367                                     entry))
7368                                         num++;
7369                         }
7370                 }
7371                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7372                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7373                 break;
7374         case RSO_OPTIONS_OC:
7375                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7376                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7377                                               /*sks_valid*/ 1,
7378                                               /*command*/ 1,
7379                                               /*field*/ 2,
7380                                               /*bit_valid*/ 1,
7381                                               /*bit*/ 2);
7382                         ctl_done((union ctl_io *)ctsio);
7383                         return (CTL_RETVAL_COMPLETE);
7384                 }
7385                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7386                 break;
7387         case RSO_OPTIONS_OC_SA:
7388                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7389                     service_action >= 32) {
7390                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7391                                               /*sks_valid*/ 1,
7392                                               /*command*/ 1,
7393                                               /*field*/ 2,
7394                                               /*bit_valid*/ 1,
7395                                               /*bit*/ 2);
7396                         ctl_done((union ctl_io *)ctsio);
7397                         return (CTL_RETVAL_COMPLETE);
7398                 }
7399                 /* FALLTHROUGH */
7400         case RSO_OPTIONS_OC_ASA:
7401                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7402                 break;
7403         default:
7404                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7405                                       /*sks_valid*/ 1,
7406                                       /*command*/ 1,
7407                                       /*field*/ 2,
7408                                       /*bit_valid*/ 1,
7409                                       /*bit*/ 2);
7410                 ctl_done((union ctl_io *)ctsio);
7411                 return (CTL_RETVAL_COMPLETE);
7412         }
7413
7414         alloc_len = scsi_4btoul(cdb->length);
7415
7416         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7417         ctsio->kern_sg_entries = 0;
7418         ctsio->kern_rel_offset = 0;
7419         ctsio->kern_data_len = min(total_len, alloc_len);
7420         ctsio->kern_total_len = ctsio->kern_data_len;
7421
7422         switch (cdb->options & RSO_OPTIONS_MASK) {
7423         case RSO_OPTIONS_ALL:
7424                 all = (struct scsi_report_supported_opcodes_all *)
7425                     ctsio->kern_data_ptr;
7426                 num = 0;
7427                 for (i = 0; i < 256; i++) {
7428                         entry = &ctl_cmd_table[i];
7429                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7430                                 for (j = 0; j < 32; j++) {
7431                                         sentry = &((const struct ctl_cmd_entry *)
7432                                             entry->execute)[j];
7433                                         if (!ctl_cmd_applicable(
7434                                             lun->be_lun->lun_type, sentry))
7435                                                 continue;
7436                                         descr = &all->descr[num++];
7437                                         descr->opcode = i;
7438                                         scsi_ulto2b(j, descr->service_action);
7439                                         descr->flags = RSO_SERVACTV;
7440                                         scsi_ulto2b(sentry->length,
7441                                             descr->cdb_length);
7442                                 }
7443                         } else {
7444                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7445                                     entry))
7446                                         continue;
7447                                 descr = &all->descr[num++];
7448                                 descr->opcode = i;
7449                                 scsi_ulto2b(0, descr->service_action);
7450                                 descr->flags = 0;
7451                                 scsi_ulto2b(entry->length, descr->cdb_length);
7452                         }
7453                 }
7454                 scsi_ulto4b(
7455                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7456                     all->length);
7457                 break;
7458         case RSO_OPTIONS_OC:
7459                 one = (struct scsi_report_supported_opcodes_one *)
7460                     ctsio->kern_data_ptr;
7461                 entry = &ctl_cmd_table[opcode];
7462                 goto fill_one;
7463         case RSO_OPTIONS_OC_SA:
7464                 one = (struct scsi_report_supported_opcodes_one *)
7465                     ctsio->kern_data_ptr;
7466                 entry = &ctl_cmd_table[opcode];
7467                 entry = &((const struct ctl_cmd_entry *)
7468                     entry->execute)[service_action];
7469 fill_one:
7470                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7471                         one->support = 3;
7472                         scsi_ulto2b(entry->length, one->cdb_length);
7473                         one->cdb_usage[0] = opcode;
7474                         memcpy(&one->cdb_usage[1], entry->usage,
7475                             entry->length - 1);
7476                 } else
7477                         one->support = 1;
7478                 break;
7479         case RSO_OPTIONS_OC_ASA:
7480                 one = (struct scsi_report_supported_opcodes_one *)
7481                     ctsio->kern_data_ptr;
7482                 entry = &ctl_cmd_table[opcode];
7483                 if (entry->flags & CTL_CMD_FLAG_SA5) {
7484                         entry = &((const struct ctl_cmd_entry *)
7485                             entry->execute)[service_action];
7486                 } else if (service_action != 0) {
7487                         one->support = 1;
7488                         break;
7489                 }
7490                 goto fill_one;
7491         }
7492
7493         ctl_set_success(ctsio);
7494         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7495         ctsio->be_move_done = ctl_config_move_done;
7496         ctl_datamove((union ctl_io *)ctsio);
7497         return(retval);
7498 }
7499
7500 int
7501 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7502 {
7503         struct scsi_report_supported_tmf *cdb;
7504         struct scsi_report_supported_tmf_ext_data *data;
7505         int retval;
7506         int alloc_len, total_len;
7507
7508         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7509
7510         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7511
7512         retval = CTL_RETVAL_COMPLETE;
7513
7514         if (cdb->options & RST_REPD)
7515                 total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7516         else
7517                 total_len = sizeof(struct scsi_report_supported_tmf_data);
7518         alloc_len = scsi_4btoul(cdb->length);
7519
7520         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7521         ctsio->kern_sg_entries = 0;
7522         ctsio->kern_rel_offset = 0;
7523         ctsio->kern_data_len = min(total_len, alloc_len);
7524         ctsio->kern_total_len = ctsio->kern_data_len;
7525
7526         data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7527         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7528             RST_TRS;
7529         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7530         data->length = total_len - 4;
7531
7532         ctl_set_success(ctsio);
7533         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7534         ctsio->be_move_done = ctl_config_move_done;
7535         ctl_datamove((union ctl_io *)ctsio);
7536         return (retval);
7537 }
7538
7539 int
7540 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7541 {
7542         struct scsi_report_timestamp *cdb;
7543         struct scsi_report_timestamp_data *data;
7544         struct timeval tv;
7545         int64_t timestamp;
7546         int retval;
7547         int alloc_len, total_len;
7548
7549         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7550
7551         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7552
7553         retval = CTL_RETVAL_COMPLETE;
7554
7555         total_len = sizeof(struct scsi_report_timestamp_data);
7556         alloc_len = scsi_4btoul(cdb->length);
7557
7558         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7559         ctsio->kern_sg_entries = 0;
7560         ctsio->kern_rel_offset = 0;
7561         ctsio->kern_data_len = min(total_len, alloc_len);
7562         ctsio->kern_total_len = ctsio->kern_data_len;
7563
7564         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7565         scsi_ulto2b(sizeof(*data) - 2, data->length);
7566         data->origin = RTS_ORIG_OUTSIDE;
7567         getmicrotime(&tv);
7568         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7569         scsi_ulto4b(timestamp >> 16, data->timestamp);
7570         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7571
7572         ctl_set_success(ctsio);
7573         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7574         ctsio->be_move_done = ctl_config_move_done;
7575         ctl_datamove((union ctl_io *)ctsio);
7576         return (retval);
7577 }
7578
7579 int
7580 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7581 {
7582         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7583         struct ctl_lun *lun = CTL_LUN(ctsio);
7584         struct scsi_per_res_in *cdb;
7585         int alloc_len, total_len = 0;
7586         /* struct scsi_per_res_in_rsrv in_data; */
7587         uint64_t key;
7588
7589         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7590
7591         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7592
7593         alloc_len = scsi_2btoul(cdb->length);
7594
7595 retry:
7596         mtx_lock(&lun->lun_lock);
7597         switch (cdb->action) {
7598         case SPRI_RK: /* read keys */
7599                 total_len = sizeof(struct scsi_per_res_in_keys) +
7600                         lun->pr_key_count *
7601                         sizeof(struct scsi_per_res_key);
7602                 break;
7603         case SPRI_RR: /* read reservation */
7604                 if (lun->flags & CTL_LUN_PR_RESERVED)
7605                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7606                 else
7607                         total_len = sizeof(struct scsi_per_res_in_header);
7608                 break;
7609         case SPRI_RC: /* report capabilities */
7610                 total_len = sizeof(struct scsi_per_res_cap);
7611                 break;
7612         case SPRI_RS: /* read full status */
7613                 total_len = sizeof(struct scsi_per_res_in_header) +
7614                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7615                     lun->pr_key_count;
7616                 break;
7617         default:
7618                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7619         }
7620         mtx_unlock(&lun->lun_lock);
7621
7622         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7623         ctsio->kern_rel_offset = 0;
7624         ctsio->kern_sg_entries = 0;
7625         ctsio->kern_data_len = min(total_len, alloc_len);
7626         ctsio->kern_total_len = ctsio->kern_data_len;
7627
7628         mtx_lock(&lun->lun_lock);
7629         switch (cdb->action) {
7630         case SPRI_RK: { // read keys
7631         struct scsi_per_res_in_keys *res_keys;
7632                 int i, key_count;
7633
7634                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7635
7636                 /*
7637                  * We had to drop the lock to allocate our buffer, which
7638                  * leaves time for someone to come in with another
7639                  * persistent reservation.  (That is unlikely, though,
7640                  * since this should be the only persistent reservation
7641                  * command active right now.)
7642                  */
7643                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7644                     (lun->pr_key_count *
7645                      sizeof(struct scsi_per_res_key)))){
7646                         mtx_unlock(&lun->lun_lock);
7647                         free(ctsio->kern_data_ptr, M_CTL);
7648                         printf("%s: reservation length changed, retrying\n",
7649                                __func__);
7650                         goto retry;
7651                 }
7652
7653                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7654
7655                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7656                              lun->pr_key_count, res_keys->header.length);
7657
7658                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7659                         if ((key = ctl_get_prkey(lun, i)) == 0)
7660                                 continue;
7661
7662                         /*
7663                          * We used lun->pr_key_count to calculate the
7664                          * size to allocate.  If it turns out the number of
7665                          * initiators with the registered flag set is
7666                          * larger than that (i.e. they haven't been kept in
7667                          * sync), we've got a problem.
7668                          */
7669                         if (key_count >= lun->pr_key_count) {
7670                                 key_count++;
7671                                 continue;
7672                         }
7673                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7674                         key_count++;
7675                 }
7676                 break;
7677         }
7678         case SPRI_RR: { // read reservation
7679                 struct scsi_per_res_in_rsrv *res;
7680                 int tmp_len, header_only;
7681
7682                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7683
7684                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7685
7686                 if (lun->flags & CTL_LUN_PR_RESERVED)
7687                 {
7688                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7689                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7690                                     res->header.length);
7691                         header_only = 0;
7692                 } else {
7693                         tmp_len = sizeof(struct scsi_per_res_in_header);
7694                         scsi_ulto4b(0, res->header.length);
7695                         header_only = 1;
7696                 }
7697
7698                 /*
7699                  * We had to drop the lock to allocate our buffer, which
7700                  * leaves time for someone to come in with another
7701                  * persistent reservation.  (That is unlikely, though,
7702                  * since this should be the only persistent reservation
7703                  * command active right now.)
7704                  */
7705                 if (tmp_len != total_len) {
7706                         mtx_unlock(&lun->lun_lock);
7707                         free(ctsio->kern_data_ptr, M_CTL);
7708                         printf("%s: reservation status changed, retrying\n",
7709                                __func__);
7710                         goto retry;
7711                 }
7712
7713                 /*
7714                  * No reservation held, so we're done.
7715                  */
7716                 if (header_only != 0)
7717                         break;
7718
7719                 /*
7720                  * If the registration is an All Registrants type, the key
7721                  * is 0, since it doesn't really matter.
7722                  */
7723                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7724                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7725                             res->data.reservation);
7726                 }
7727                 res->data.scopetype = lun->pr_res_type;
7728                 break;
7729         }
7730         case SPRI_RC:     //report capabilities
7731         {
7732                 struct scsi_per_res_cap *res_cap;
7733                 uint16_t type_mask;
7734
7735                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7736                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7737                 res_cap->flags1 = SPRI_CRH;
7738                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7739                 type_mask = SPRI_TM_WR_EX_AR |
7740                             SPRI_TM_EX_AC_RO |
7741                             SPRI_TM_WR_EX_RO |
7742                             SPRI_TM_EX_AC |
7743                             SPRI_TM_WR_EX |
7744                             SPRI_TM_EX_AC_AR;
7745                 scsi_ulto2b(type_mask, res_cap->type_mask);
7746                 break;
7747         }
7748         case SPRI_RS: { // read full status
7749                 struct scsi_per_res_in_full *res_status;
7750                 struct scsi_per_res_in_full_desc *res_desc;
7751                 struct ctl_port *port;
7752                 int i, len;
7753
7754                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7755
7756                 /*
7757                  * We had to drop the lock to allocate our buffer, which
7758                  * leaves time for someone to come in with another
7759                  * persistent reservation.  (That is unlikely, though,
7760                  * since this should be the only persistent reservation
7761                  * command active right now.)
7762                  */
7763                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7764                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7765                      lun->pr_key_count)){
7766                         mtx_unlock(&lun->lun_lock);
7767                         free(ctsio->kern_data_ptr, M_CTL);
7768                         printf("%s: reservation length changed, retrying\n",
7769                                __func__);
7770                         goto retry;
7771                 }
7772
7773                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7774
7775                 res_desc = &res_status->desc[0];
7776                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7777                         if ((key = ctl_get_prkey(lun, i)) == 0)
7778                                 continue;
7779
7780                         scsi_u64to8b(key, res_desc->res_key.key);
7781                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7782                             (lun->pr_res_idx == i ||
7783                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7784                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7785                                 res_desc->scopetype = lun->pr_res_type;
7786                         }
7787                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7788                             res_desc->rel_trgt_port_id);
7789                         len = 0;
7790                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7791                         if (port != NULL)
7792                                 len = ctl_create_iid(port,
7793                                     i % CTL_MAX_INIT_PER_PORT,
7794                                     res_desc->transport_id);
7795                         scsi_ulto4b(len, res_desc->additional_length);
7796                         res_desc = (struct scsi_per_res_in_full_desc *)
7797                             &res_desc->transport_id[len];
7798                 }
7799                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7800                     res_status->header.length);
7801                 break;
7802         }
7803         default:
7804                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7805         }
7806         mtx_unlock(&lun->lun_lock);
7807
7808         ctl_set_success(ctsio);
7809         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7810         ctsio->be_move_done = ctl_config_move_done;
7811         ctl_datamove((union ctl_io *)ctsio);
7812         return (CTL_RETVAL_COMPLETE);
7813 }
7814
7815 /*
7816  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7817  * it should return.
7818  */
7819 static int
7820 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7821                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7822                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7823                 struct scsi_per_res_out_parms* param)
7824 {
7825         union ctl_ha_msg persis_io;
7826         int i;
7827
7828         mtx_lock(&lun->lun_lock);
7829         if (sa_res_key == 0) {
7830                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7831                         /* validate scope and type */
7832                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7833                              SPR_LU_SCOPE) {
7834                                 mtx_unlock(&lun->lun_lock);
7835                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7836                                                       /*sks_valid*/ 1,
7837                                                       /*command*/ 1,
7838                                                       /*field*/ 2,
7839                                                       /*bit_valid*/ 1,
7840                                                       /*bit*/ 4);
7841                                 ctl_done((union ctl_io *)ctsio);
7842                                 return (1);
7843                         }
7844
7845                         if (type>8 || type==2 || type==4 || type==0) {
7846                                 mtx_unlock(&lun->lun_lock);
7847                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7848                                                       /*sks_valid*/ 1,
7849                                                       /*command*/ 1,
7850                                                       /*field*/ 2,
7851                                                       /*bit_valid*/ 1,
7852                                                       /*bit*/ 0);
7853                                 ctl_done((union ctl_io *)ctsio);
7854                                 return (1);
7855                         }
7856
7857                         /*
7858                          * Unregister everybody else and build UA for
7859                          * them
7860                          */
7861                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7862                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7863                                         continue;
7864
7865                                 ctl_clr_prkey(lun, i);
7866                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7867                         }
7868                         lun->pr_key_count = 1;
7869                         lun->pr_res_type = type;
7870                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7871                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7872                                 lun->pr_res_idx = residx;
7873                         lun->pr_generation++;
7874                         mtx_unlock(&lun->lun_lock);
7875
7876                         /* send msg to other side */
7877                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7878                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7879                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7880                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7881                         persis_io.pr.pr_info.res_type = type;
7882                         memcpy(persis_io.pr.pr_info.sa_res_key,
7883                                param->serv_act_res_key,
7884                                sizeof(param->serv_act_res_key));
7885                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7886                             sizeof(persis_io.pr), M_WAITOK);
7887                 } else {
7888                         /* not all registrants */
7889                         mtx_unlock(&lun->lun_lock);
7890                         free(ctsio->kern_data_ptr, M_CTL);
7891                         ctl_set_invalid_field(ctsio,
7892                                               /*sks_valid*/ 1,
7893                                               /*command*/ 0,
7894                                               /*field*/ 8,
7895                                               /*bit_valid*/ 0,
7896                                               /*bit*/ 0);
7897                         ctl_done((union ctl_io *)ctsio);
7898                         return (1);
7899                 }
7900         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7901                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7902                 int found = 0;
7903
7904                 if (res_key == sa_res_key) {
7905                         /* special case */
7906                         /*
7907                          * The spec implies this is not good but doesn't
7908                          * say what to do. There are two choices either
7909                          * generate a res conflict or check condition
7910                          * with illegal field in parameter data. Since
7911                          * that is what is done when the sa_res_key is
7912                          * zero I'll take that approach since this has
7913                          * to do with the sa_res_key.
7914                          */
7915                         mtx_unlock(&lun->lun_lock);
7916                         free(ctsio->kern_data_ptr, M_CTL);
7917                         ctl_set_invalid_field(ctsio,
7918                                               /*sks_valid*/ 1,
7919                                               /*command*/ 0,
7920                                               /*field*/ 8,
7921                                               /*bit_valid*/ 0,
7922                                               /*bit*/ 0);
7923                         ctl_done((union ctl_io *)ctsio);
7924                         return (1);
7925                 }
7926
7927                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7928                         if (ctl_get_prkey(lun, i) != sa_res_key)
7929                                 continue;
7930
7931                         found = 1;
7932                         ctl_clr_prkey(lun, i);
7933                         lun->pr_key_count--;
7934                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7935                 }
7936                 if (!found) {
7937                         mtx_unlock(&lun->lun_lock);
7938                         free(ctsio->kern_data_ptr, M_CTL);
7939                         ctl_set_reservation_conflict(ctsio);
7940                         ctl_done((union ctl_io *)ctsio);
7941                         return (CTL_RETVAL_COMPLETE);
7942                 }
7943                 lun->pr_generation++;
7944                 mtx_unlock(&lun->lun_lock);
7945
7946                 /* send msg to other side */
7947                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7948                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7949                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7950                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7951                 persis_io.pr.pr_info.res_type = type;
7952                 memcpy(persis_io.pr.pr_info.sa_res_key,
7953                        param->serv_act_res_key,
7954                        sizeof(param->serv_act_res_key));
7955                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7956                     sizeof(persis_io.pr), M_WAITOK);
7957         } else {
7958                 /* Reserved but not all registrants */
7959                 /* sa_res_key is res holder */
7960                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7961                         /* validate scope and type */
7962                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7963                              SPR_LU_SCOPE) {
7964                                 mtx_unlock(&lun->lun_lock);
7965                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7966                                                       /*sks_valid*/ 1,
7967                                                       /*command*/ 1,
7968                                                       /*field*/ 2,
7969                                                       /*bit_valid*/ 1,
7970                                                       /*bit*/ 4);
7971                                 ctl_done((union ctl_io *)ctsio);
7972                                 return (1);
7973                         }
7974
7975                         if (type>8 || type==2 || type==4 || type==0) {
7976                                 mtx_unlock(&lun->lun_lock);
7977                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7978                                                       /*sks_valid*/ 1,
7979                                                       /*command*/ 1,
7980                                                       /*field*/ 2,
7981                                                       /*bit_valid*/ 1,
7982                                                       /*bit*/ 0);
7983                                 ctl_done((union ctl_io *)ctsio);
7984                                 return (1);
7985                         }
7986
7987                         /*
7988                          * Do the following:
7989                          * if sa_res_key != res_key remove all
7990                          * registrants w/sa_res_key and generate UA
7991                          * for these registrants(Registrations
7992                          * Preempted) if it wasn't an exclusive
7993                          * reservation generate UA(Reservations
7994                          * Preempted) for all other registered nexuses
7995                          * if the type has changed. Establish the new
7996                          * reservation and holder. If res_key and
7997                          * sa_res_key are the same do the above
7998                          * except don't unregister the res holder.
7999                          */
8000
8001                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8002                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8003                                         continue;
8004
8005                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
8006                                         ctl_clr_prkey(lun, i);
8007                                         lun->pr_key_count--;
8008                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8009                                 } else if (type != lun->pr_res_type &&
8010                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8011                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8012                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8013                                 }
8014                         }
8015                         lun->pr_res_type = type;
8016                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8017                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8018                                 lun->pr_res_idx = residx;
8019                         else
8020                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8021                         lun->pr_generation++;
8022                         mtx_unlock(&lun->lun_lock);
8023
8024                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8025                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8026                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8027                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8028                         persis_io.pr.pr_info.res_type = type;
8029                         memcpy(persis_io.pr.pr_info.sa_res_key,
8030                                param->serv_act_res_key,
8031                                sizeof(param->serv_act_res_key));
8032                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8033                             sizeof(persis_io.pr), M_WAITOK);
8034                 } else {
8035                         /*
8036                          * sa_res_key is not the res holder just
8037                          * remove registrants
8038                          */
8039                         int found=0;
8040
8041                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8042                                 if (sa_res_key != ctl_get_prkey(lun, i))
8043                                         continue;
8044
8045                                 found = 1;
8046                                 ctl_clr_prkey(lun, i);
8047                                 lun->pr_key_count--;
8048                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8049                         }
8050
8051                         if (!found) {
8052                                 mtx_unlock(&lun->lun_lock);
8053                                 free(ctsio->kern_data_ptr, M_CTL);
8054                                 ctl_set_reservation_conflict(ctsio);
8055                                 ctl_done((union ctl_io *)ctsio);
8056                                 return (1);
8057                         }
8058                         lun->pr_generation++;
8059                         mtx_unlock(&lun->lun_lock);
8060
8061                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8062                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8063                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8064                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8065                         persis_io.pr.pr_info.res_type = type;
8066                         memcpy(persis_io.pr.pr_info.sa_res_key,
8067                                param->serv_act_res_key,
8068                                sizeof(param->serv_act_res_key));
8069                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8070                             sizeof(persis_io.pr), M_WAITOK);
8071                 }
8072         }
8073         return (0);
8074 }
8075
8076 static void
8077 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8078 {
8079         uint64_t sa_res_key;
8080         int i;
8081
8082         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8083
8084         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8085          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8086          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8087                 if (sa_res_key == 0) {
8088                         /*
8089                          * Unregister everybody else and build UA for
8090                          * them
8091                          */
8092                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8093                                 if (i == msg->pr.pr_info.residx ||
8094                                     ctl_get_prkey(lun, i) == 0)
8095                                         continue;
8096
8097                                 ctl_clr_prkey(lun, i);
8098                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8099                         }
8100
8101                         lun->pr_key_count = 1;
8102                         lun->pr_res_type = msg->pr.pr_info.res_type;
8103                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8104                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8105                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8106                 } else {
8107                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8108                                 if (sa_res_key == ctl_get_prkey(lun, i))
8109                                         continue;
8110
8111                                 ctl_clr_prkey(lun, i);
8112                                 lun->pr_key_count--;
8113                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8114                         }
8115                 }
8116         } else {
8117                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8118                         if (i == msg->pr.pr_info.residx ||
8119                             ctl_get_prkey(lun, i) == 0)
8120                                 continue;
8121
8122                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8123                                 ctl_clr_prkey(lun, i);
8124                                 lun->pr_key_count--;
8125                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8126                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8127                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8128                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8129                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8130                         }
8131                 }
8132                 lun->pr_res_type = msg->pr.pr_info.res_type;
8133                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8134                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8135                         lun->pr_res_idx = msg->pr.pr_info.residx;
8136                 else
8137                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8138         }
8139         lun->pr_generation++;
8140
8141 }
8142
8143
8144 int
8145 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8146 {
8147         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8148         struct ctl_lun *lun = CTL_LUN(ctsio);
8149         int retval;
8150         u_int32_t param_len;
8151         struct scsi_per_res_out *cdb;
8152         struct scsi_per_res_out_parms* param;
8153         uint32_t residx;
8154         uint64_t res_key, sa_res_key, key;
8155         uint8_t type;
8156         union ctl_ha_msg persis_io;
8157         int    i;
8158
8159         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8160
8161         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8162         retval = CTL_RETVAL_COMPLETE;
8163
8164         /*
8165          * We only support whole-LUN scope.  The scope & type are ignored for
8166          * register, register and ignore existing key and clear.
8167          * We sometimes ignore scope and type on preempts too!!
8168          * Verify reservation type here as well.
8169          */
8170         type = cdb->scope_type & SPR_TYPE_MASK;
8171         if ((cdb->action == SPRO_RESERVE)
8172          || (cdb->action == SPRO_RELEASE)) {
8173                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8174                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8175                                               /*sks_valid*/ 1,
8176                                               /*command*/ 1,
8177                                               /*field*/ 2,
8178                                               /*bit_valid*/ 1,
8179                                               /*bit*/ 4);
8180                         ctl_done((union ctl_io *)ctsio);
8181                         return (CTL_RETVAL_COMPLETE);
8182                 }
8183
8184                 if (type>8 || type==2 || type==4 || type==0) {
8185                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8186                                               /*sks_valid*/ 1,
8187                                               /*command*/ 1,
8188                                               /*field*/ 2,
8189                                               /*bit_valid*/ 1,
8190                                               /*bit*/ 0);
8191                         ctl_done((union ctl_io *)ctsio);
8192                         return (CTL_RETVAL_COMPLETE);
8193                 }
8194         }
8195
8196         param_len = scsi_4btoul(cdb->length);
8197
8198         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8199                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8200                 ctsio->kern_data_len = param_len;
8201                 ctsio->kern_total_len = param_len;
8202                 ctsio->kern_rel_offset = 0;
8203                 ctsio->kern_sg_entries = 0;
8204                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8205                 ctsio->be_move_done = ctl_config_move_done;
8206                 ctl_datamove((union ctl_io *)ctsio);
8207
8208                 return (CTL_RETVAL_COMPLETE);
8209         }
8210
8211         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8212
8213         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8214         res_key = scsi_8btou64(param->res_key.key);
8215         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8216
8217         /*
8218          * Validate the reservation key here except for SPRO_REG_IGNO
8219          * This must be done for all other service actions
8220          */
8221         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8222                 mtx_lock(&lun->lun_lock);
8223                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8224                         if (res_key != key) {
8225                                 /*
8226                                  * The current key passed in doesn't match
8227                                  * the one the initiator previously
8228                                  * registered.
8229                                  */
8230                                 mtx_unlock(&lun->lun_lock);
8231                                 free(ctsio->kern_data_ptr, M_CTL);
8232                                 ctl_set_reservation_conflict(ctsio);
8233                                 ctl_done((union ctl_io *)ctsio);
8234                                 return (CTL_RETVAL_COMPLETE);
8235                         }
8236                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8237                         /*
8238                          * We are not registered
8239                          */
8240                         mtx_unlock(&lun->lun_lock);
8241                         free(ctsio->kern_data_ptr, M_CTL);
8242                         ctl_set_reservation_conflict(ctsio);
8243                         ctl_done((union ctl_io *)ctsio);
8244                         return (CTL_RETVAL_COMPLETE);
8245                 } else if (res_key != 0) {
8246                         /*
8247                          * We are not registered and trying to register but
8248                          * the register key isn't zero.
8249                          */
8250                         mtx_unlock(&lun->lun_lock);
8251                         free(ctsio->kern_data_ptr, M_CTL);
8252                         ctl_set_reservation_conflict(ctsio);
8253                         ctl_done((union ctl_io *)ctsio);
8254                         return (CTL_RETVAL_COMPLETE);
8255                 }
8256                 mtx_unlock(&lun->lun_lock);
8257         }
8258
8259         switch (cdb->action & SPRO_ACTION_MASK) {
8260         case SPRO_REGISTER:
8261         case SPRO_REG_IGNO: {
8262
8263 #if 0
8264                 printf("Registration received\n");
8265 #endif
8266
8267                 /*
8268                  * We don't support any of these options, as we report in
8269                  * the read capabilities request (see
8270                  * ctl_persistent_reserve_in(), above).
8271                  */
8272                 if ((param->flags & SPR_SPEC_I_PT)
8273                  || (param->flags & SPR_ALL_TG_PT)
8274                  || (param->flags & SPR_APTPL)) {
8275                         int bit_ptr;
8276
8277                         if (param->flags & SPR_APTPL)
8278                                 bit_ptr = 0;
8279                         else if (param->flags & SPR_ALL_TG_PT)
8280                                 bit_ptr = 2;
8281                         else /* SPR_SPEC_I_PT */
8282                                 bit_ptr = 3;
8283
8284                         free(ctsio->kern_data_ptr, M_CTL);
8285                         ctl_set_invalid_field(ctsio,
8286                                               /*sks_valid*/ 1,
8287                                               /*command*/ 0,
8288                                               /*field*/ 20,
8289                                               /*bit_valid*/ 1,
8290                                               /*bit*/ bit_ptr);
8291                         ctl_done((union ctl_io *)ctsio);
8292                         return (CTL_RETVAL_COMPLETE);
8293                 }
8294
8295                 mtx_lock(&lun->lun_lock);
8296
8297                 /*
8298                  * The initiator wants to clear the
8299                  * key/unregister.
8300                  */
8301                 if (sa_res_key == 0) {
8302                         if ((res_key == 0
8303                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8304                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8305                           && ctl_get_prkey(lun, residx) == 0)) {
8306                                 mtx_unlock(&lun->lun_lock);
8307                                 goto done;
8308                         }
8309
8310                         ctl_clr_prkey(lun, residx);
8311                         lun->pr_key_count--;
8312
8313                         if (residx == lun->pr_res_idx) {
8314                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8315                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8316
8317                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8318                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8319                                     lun->pr_key_count) {
8320                                         /*
8321                                          * If the reservation is a registrants
8322                                          * only type we need to generate a UA
8323                                          * for other registered inits.  The
8324                                          * sense code should be RESERVATIONS
8325                                          * RELEASED
8326                                          */
8327
8328                                         for (i = softc->init_min; i < softc->init_max; i++){
8329                                                 if (ctl_get_prkey(lun, i) == 0)
8330                                                         continue;
8331                                                 ctl_est_ua(lun, i,
8332                                                     CTL_UA_RES_RELEASE);
8333                                         }
8334                                 }
8335                                 lun->pr_res_type = 0;
8336                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8337                                 if (lun->pr_key_count==0) {
8338                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8339                                         lun->pr_res_type = 0;
8340                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8341                                 }
8342                         }
8343                         lun->pr_generation++;
8344                         mtx_unlock(&lun->lun_lock);
8345
8346                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8347                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8348                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8349                         persis_io.pr.pr_info.residx = residx;
8350                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8351                             sizeof(persis_io.pr), M_WAITOK);
8352                 } else /* sa_res_key != 0 */ {
8353
8354                         /*
8355                          * If we aren't registered currently then increment
8356                          * the key count and set the registered flag.
8357                          */
8358                         ctl_alloc_prkey(lun, residx);
8359                         if (ctl_get_prkey(lun, residx) == 0)
8360                                 lun->pr_key_count++;
8361                         ctl_set_prkey(lun, residx, sa_res_key);
8362                         lun->pr_generation++;
8363                         mtx_unlock(&lun->lun_lock);
8364
8365                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8366                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8367                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8368                         persis_io.pr.pr_info.residx = residx;
8369                         memcpy(persis_io.pr.pr_info.sa_res_key,
8370                                param->serv_act_res_key,
8371                                sizeof(param->serv_act_res_key));
8372                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8373                             sizeof(persis_io.pr), M_WAITOK);
8374                 }
8375
8376                 break;
8377         }
8378         case SPRO_RESERVE:
8379 #if 0
8380                 printf("Reserve executed type %d\n", type);
8381 #endif
8382                 mtx_lock(&lun->lun_lock);
8383                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8384                         /*
8385                          * if this isn't the reservation holder and it's
8386                          * not a "all registrants" type or if the type is
8387                          * different then we have a conflict
8388                          */
8389                         if ((lun->pr_res_idx != residx
8390                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8391                          || lun->pr_res_type != type) {
8392                                 mtx_unlock(&lun->lun_lock);
8393                                 free(ctsio->kern_data_ptr, M_CTL);
8394                                 ctl_set_reservation_conflict(ctsio);
8395                                 ctl_done((union ctl_io *)ctsio);
8396                                 return (CTL_RETVAL_COMPLETE);
8397                         }
8398                         mtx_unlock(&lun->lun_lock);
8399                 } else /* create a reservation */ {
8400                         /*
8401                          * If it's not an "all registrants" type record
8402                          * reservation holder
8403                          */
8404                         if (type != SPR_TYPE_WR_EX_AR
8405                          && type != SPR_TYPE_EX_AC_AR)
8406                                 lun->pr_res_idx = residx; /* Res holder */
8407                         else
8408                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8409
8410                         lun->flags |= CTL_LUN_PR_RESERVED;
8411                         lun->pr_res_type = type;
8412
8413                         mtx_unlock(&lun->lun_lock);
8414
8415                         /* send msg to other side */
8416                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8417                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8418                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8419                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8420                         persis_io.pr.pr_info.res_type = type;
8421                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8422                             sizeof(persis_io.pr), M_WAITOK);
8423                 }
8424                 break;
8425
8426         case SPRO_RELEASE:
8427                 mtx_lock(&lun->lun_lock);
8428                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8429                         /* No reservation exists return good status */
8430                         mtx_unlock(&lun->lun_lock);
8431                         goto done;
8432                 }
8433                 /*
8434                  * Is this nexus a reservation holder?
8435                  */
8436                 if (lun->pr_res_idx != residx
8437                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8438                         /*
8439                          * not a res holder return good status but
8440                          * do nothing
8441                          */
8442                         mtx_unlock(&lun->lun_lock);
8443                         goto done;
8444                 }
8445
8446                 if (lun->pr_res_type != type) {
8447                         mtx_unlock(&lun->lun_lock);
8448                         free(ctsio->kern_data_ptr, M_CTL);
8449                         ctl_set_illegal_pr_release(ctsio);
8450                         ctl_done((union ctl_io *)ctsio);
8451                         return (CTL_RETVAL_COMPLETE);
8452                 }
8453
8454                 /* okay to release */
8455                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8456                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8457                 lun->pr_res_type = 0;
8458
8459                 /*
8460                  * If this isn't an exclusive access reservation and NUAR
8461                  * is not set, generate UA for all other registrants.
8462                  */
8463                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8464                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8465                         for (i = softc->init_min; i < softc->init_max; i++) {
8466                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8467                                         continue;
8468                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8469                         }
8470                 }
8471                 mtx_unlock(&lun->lun_lock);
8472
8473                 /* Send msg to other side */
8474                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8475                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8476                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8477                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8478                      sizeof(persis_io.pr), M_WAITOK);
8479                 break;
8480
8481         case SPRO_CLEAR:
8482                 /* send msg to other side */
8483
8484                 mtx_lock(&lun->lun_lock);
8485                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8486                 lun->pr_res_type = 0;
8487                 lun->pr_key_count = 0;
8488                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8489
8490                 ctl_clr_prkey(lun, residx);
8491                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8492                         if (ctl_get_prkey(lun, i) != 0) {
8493                                 ctl_clr_prkey(lun, i);
8494                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8495                         }
8496                 lun->pr_generation++;
8497                 mtx_unlock(&lun->lun_lock);
8498
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_CLEAR;
8502                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8503                      sizeof(persis_io.pr), M_WAITOK);
8504                 break;
8505
8506         case SPRO_PREEMPT:
8507         case SPRO_PRE_ABO: {
8508                 int nretval;
8509
8510                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8511                                           residx, ctsio, cdb, param);
8512                 if (nretval != 0)
8513                         return (CTL_RETVAL_COMPLETE);
8514                 break;
8515         }
8516         default:
8517                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8518         }
8519
8520 done:
8521         free(ctsio->kern_data_ptr, M_CTL);
8522         ctl_set_success(ctsio);
8523         ctl_done((union ctl_io *)ctsio);
8524
8525         return (retval);
8526 }
8527
8528 /*
8529  * This routine is for handling a message from the other SC pertaining to
8530  * persistent reserve out. All the error checking will have been done
8531  * so only perorming the action need be done here to keep the two
8532  * in sync.
8533  */
8534 static void
8535 ctl_hndl_per_res_out_on_other_sc(union ctl_io *io)
8536 {
8537         struct ctl_softc *softc = CTL_SOFTC(io);
8538         union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg;
8539         struct ctl_lun *lun;
8540         int i;
8541         uint32_t residx, targ_lun;
8542
8543         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8544         mtx_lock(&softc->ctl_lock);
8545         if (targ_lun >= ctl_max_luns ||
8546             (lun = softc->ctl_luns[targ_lun]) == NULL) {
8547                 mtx_unlock(&softc->ctl_lock);
8548                 return;
8549         }
8550         mtx_lock(&lun->lun_lock);
8551         mtx_unlock(&softc->ctl_lock);
8552         if (lun->flags & CTL_LUN_DISABLED) {
8553                 mtx_unlock(&lun->lun_lock);
8554                 return;
8555         }
8556         residx = ctl_get_initindex(&msg->hdr.nexus);
8557         switch(msg->pr.pr_info.action) {
8558         case CTL_PR_REG_KEY:
8559                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8560                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8561                         lun->pr_key_count++;
8562                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8563                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8564                 lun->pr_generation++;
8565                 break;
8566
8567         case CTL_PR_UNREG_KEY:
8568                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8569                 lun->pr_key_count--;
8570
8571                 /* XXX Need to see if the reservation has been released */
8572                 /* if so do we need to generate UA? */
8573                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8574                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8575                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8576
8577                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8578                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8579                             lun->pr_key_count) {
8580                                 /*
8581                                  * If the reservation is a registrants
8582                                  * only type we need to generate a UA
8583                                  * for other registered inits.  The
8584                                  * sense code should be RESERVATIONS
8585                                  * RELEASED
8586                                  */
8587
8588                                 for (i = softc->init_min; i < softc->init_max; i++) {
8589                                         if (ctl_get_prkey(lun, i) == 0)
8590                                                 continue;
8591
8592                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8593                                 }
8594                         }
8595                         lun->pr_res_type = 0;
8596                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8597                         if (lun->pr_key_count==0) {
8598                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8599                                 lun->pr_res_type = 0;
8600                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8601                         }
8602                 }
8603                 lun->pr_generation++;
8604                 break;
8605
8606         case CTL_PR_RESERVE:
8607                 lun->flags |= CTL_LUN_PR_RESERVED;
8608                 lun->pr_res_type = msg->pr.pr_info.res_type;
8609                 lun->pr_res_idx = msg->pr.pr_info.residx;
8610
8611                 break;
8612
8613         case CTL_PR_RELEASE:
8614                 /*
8615                  * If this isn't an exclusive access reservation and NUAR
8616                  * is not set, generate UA for all other registrants.
8617                  */
8618                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8619                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8620                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8621                         for (i = softc->init_min; i < softc->init_max; i++)
8622                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8623                                         continue;
8624                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8625                 }
8626
8627                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8628                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8629                 lun->pr_res_type = 0;
8630                 break;
8631
8632         case CTL_PR_PREEMPT:
8633                 ctl_pro_preempt_other(lun, msg);
8634                 break;
8635         case CTL_PR_CLEAR:
8636                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8637                 lun->pr_res_type = 0;
8638                 lun->pr_key_count = 0;
8639                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8640
8641                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8642                         if (ctl_get_prkey(lun, i) == 0)
8643                                 continue;
8644                         ctl_clr_prkey(lun, i);
8645                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8646                 }
8647                 lun->pr_generation++;
8648                 break;
8649         }
8650
8651         mtx_unlock(&lun->lun_lock);
8652 }
8653
8654 int
8655 ctl_read_write(struct ctl_scsiio *ctsio)
8656 {
8657         struct ctl_lun *lun = CTL_LUN(ctsio);
8658         struct ctl_lba_len_flags *lbalen;
8659         uint64_t lba;
8660         uint32_t num_blocks;
8661         int flags, retval;
8662         int isread;
8663
8664         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8665
8666         flags = 0;
8667         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8668               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8669         switch (ctsio->cdb[0]) {
8670         case READ_6:
8671         case WRITE_6: {
8672                 struct scsi_rw_6 *cdb;
8673
8674                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8675
8676                 lba = scsi_3btoul(cdb->addr);
8677                 /* only 5 bits are valid in the most significant address byte */
8678                 lba &= 0x1fffff;
8679                 num_blocks = cdb->length;
8680                 /*
8681                  * This is correct according to SBC-2.
8682                  */
8683                 if (num_blocks == 0)
8684                         num_blocks = 256;
8685                 break;
8686         }
8687         case READ_10:
8688         case WRITE_10: {
8689                 struct scsi_rw_10 *cdb;
8690
8691                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8692                 if (cdb->byte2 & SRW10_FUA)
8693                         flags |= CTL_LLF_FUA;
8694                 if (cdb->byte2 & SRW10_DPO)
8695                         flags |= CTL_LLF_DPO;
8696                 lba = scsi_4btoul(cdb->addr);
8697                 num_blocks = scsi_2btoul(cdb->length);
8698                 break;
8699         }
8700         case WRITE_VERIFY_10: {
8701                 struct scsi_write_verify_10 *cdb;
8702
8703                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8704                 flags |= CTL_LLF_FUA;
8705                 if (cdb->byte2 & SWV_DPO)
8706                         flags |= CTL_LLF_DPO;
8707                 lba = scsi_4btoul(cdb->addr);
8708                 num_blocks = scsi_2btoul(cdb->length);
8709                 break;
8710         }
8711         case READ_12:
8712         case WRITE_12: {
8713                 struct scsi_rw_12 *cdb;
8714
8715                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8716                 if (cdb->byte2 & SRW12_FUA)
8717                         flags |= CTL_LLF_FUA;
8718                 if (cdb->byte2 & SRW12_DPO)
8719                         flags |= CTL_LLF_DPO;
8720                 lba = scsi_4btoul(cdb->addr);
8721                 num_blocks = scsi_4btoul(cdb->length);
8722                 break;
8723         }
8724         case WRITE_VERIFY_12: {
8725                 struct scsi_write_verify_12 *cdb;
8726
8727                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8728                 flags |= CTL_LLF_FUA;
8729                 if (cdb->byte2 & SWV_DPO)
8730                         flags |= CTL_LLF_DPO;
8731                 lba = scsi_4btoul(cdb->addr);
8732                 num_blocks = scsi_4btoul(cdb->length);
8733                 break;
8734         }
8735         case READ_16:
8736         case WRITE_16: {
8737                 struct scsi_rw_16 *cdb;
8738
8739                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8740                 if (cdb->byte2 & SRW12_FUA)
8741                         flags |= CTL_LLF_FUA;
8742                 if (cdb->byte2 & SRW12_DPO)
8743                         flags |= CTL_LLF_DPO;
8744                 lba = scsi_8btou64(cdb->addr);
8745                 num_blocks = scsi_4btoul(cdb->length);
8746                 break;
8747         }
8748         case WRITE_ATOMIC_16: {
8749                 struct scsi_write_atomic_16 *cdb;
8750
8751                 if (lun->be_lun->atomicblock == 0) {
8752                         ctl_set_invalid_opcode(ctsio);
8753                         ctl_done((union ctl_io *)ctsio);
8754                         return (CTL_RETVAL_COMPLETE);
8755                 }
8756
8757                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8758                 if (cdb->byte2 & SRW12_FUA)
8759                         flags |= CTL_LLF_FUA;
8760                 if (cdb->byte2 & SRW12_DPO)
8761                         flags |= CTL_LLF_DPO;
8762                 lba = scsi_8btou64(cdb->addr);
8763                 num_blocks = scsi_2btoul(cdb->length);
8764                 if (num_blocks > lun->be_lun->atomicblock) {
8765                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8766                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8767                             /*bit*/ 0);
8768                         ctl_done((union ctl_io *)ctsio);
8769                         return (CTL_RETVAL_COMPLETE);
8770                 }
8771                 break;
8772         }
8773         case WRITE_VERIFY_16: {
8774                 struct scsi_write_verify_16 *cdb;
8775
8776                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8777                 flags |= CTL_LLF_FUA;
8778                 if (cdb->byte2 & SWV_DPO)
8779                         flags |= CTL_LLF_DPO;
8780                 lba = scsi_8btou64(cdb->addr);
8781                 num_blocks = scsi_4btoul(cdb->length);
8782                 break;
8783         }
8784         default:
8785                 /*
8786                  * We got a command we don't support.  This shouldn't
8787                  * happen, commands should be filtered out above us.
8788                  */
8789                 ctl_set_invalid_opcode(ctsio);
8790                 ctl_done((union ctl_io *)ctsio);
8791
8792                 return (CTL_RETVAL_COMPLETE);
8793                 break; /* NOTREACHED */
8794         }
8795
8796         /*
8797          * The first check is to make sure we're in bounds, the second
8798          * check is to catch wrap-around problems.  If the lba + num blocks
8799          * is less than the lba, then we've wrapped around and the block
8800          * range is invalid anyway.
8801          */
8802         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8803          || ((lba + num_blocks) < lba)) {
8804                 ctl_set_lba_out_of_range(ctsio,
8805                     MAX(lba, lun->be_lun->maxlba + 1));
8806                 ctl_done((union ctl_io *)ctsio);
8807                 return (CTL_RETVAL_COMPLETE);
8808         }
8809
8810         /*
8811          * According to SBC-3, a transfer length of 0 is not an error.
8812          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8813          * translates to 256 blocks for those commands.
8814          */
8815         if (num_blocks == 0) {
8816                 ctl_set_success(ctsio);
8817                 ctl_done((union ctl_io *)ctsio);
8818                 return (CTL_RETVAL_COMPLETE);
8819         }
8820
8821         /* Set FUA and/or DPO if caches are disabled. */
8822         if (isread) {
8823                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8824                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8825         } else {
8826                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8827                         flags |= CTL_LLF_FUA;
8828         }
8829
8830         lbalen = (struct ctl_lba_len_flags *)
8831             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8832         lbalen->lba = lba;
8833         lbalen->len = num_blocks;
8834         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8835
8836         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8837         ctsio->kern_rel_offset = 0;
8838
8839         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8840
8841         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8842         return (retval);
8843 }
8844
8845 static int
8846 ctl_cnw_cont(union ctl_io *io)
8847 {
8848         struct ctl_lun *lun = CTL_LUN(io);
8849         struct ctl_scsiio *ctsio;
8850         struct ctl_lba_len_flags *lbalen;
8851         int retval;
8852
8853         ctsio = &io->scsiio;
8854         ctsio->io_hdr.status = CTL_STATUS_NONE;
8855         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8856         lbalen = (struct ctl_lba_len_flags *)
8857             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8858         lbalen->flags &= ~CTL_LLF_COMPARE;
8859         lbalen->flags |= CTL_LLF_WRITE;
8860
8861         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8862         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8863         return (retval);
8864 }
8865
8866 int
8867 ctl_cnw(struct ctl_scsiio *ctsio)
8868 {
8869         struct ctl_lun *lun = CTL_LUN(ctsio);
8870         struct ctl_lba_len_flags *lbalen;
8871         uint64_t lba;
8872         uint32_t num_blocks;
8873         int flags, retval;
8874
8875         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8876
8877         flags = 0;
8878         switch (ctsio->cdb[0]) {
8879         case COMPARE_AND_WRITE: {
8880                 struct scsi_compare_and_write *cdb;
8881
8882                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8883                 if (cdb->byte2 & SRW10_FUA)
8884                         flags |= CTL_LLF_FUA;
8885                 if (cdb->byte2 & SRW10_DPO)
8886                         flags |= CTL_LLF_DPO;
8887                 lba = scsi_8btou64(cdb->addr);
8888                 num_blocks = cdb->length;
8889                 break;
8890         }
8891         default:
8892                 /*
8893                  * We got a command we don't support.  This shouldn't
8894                  * happen, commands should be filtered out above us.
8895                  */
8896                 ctl_set_invalid_opcode(ctsio);
8897                 ctl_done((union ctl_io *)ctsio);
8898
8899                 return (CTL_RETVAL_COMPLETE);
8900                 break; /* NOTREACHED */
8901         }
8902
8903         /*
8904          * The first check is to make sure we're in bounds, the second
8905          * check is to catch wrap-around problems.  If the lba + num blocks
8906          * is less than the lba, then we've wrapped around and the block
8907          * range is invalid anyway.
8908          */
8909         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8910          || ((lba + num_blocks) < lba)) {
8911                 ctl_set_lba_out_of_range(ctsio,
8912                     MAX(lba, lun->be_lun->maxlba + 1));
8913                 ctl_done((union ctl_io *)ctsio);
8914                 return (CTL_RETVAL_COMPLETE);
8915         }
8916
8917         /*
8918          * According to SBC-3, a transfer length of 0 is not an error.
8919          */
8920         if (num_blocks == 0) {
8921                 ctl_set_success(ctsio);
8922                 ctl_done((union ctl_io *)ctsio);
8923                 return (CTL_RETVAL_COMPLETE);
8924         }
8925
8926         /* Set FUA if write cache is disabled. */
8927         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8928                 flags |= CTL_LLF_FUA;
8929
8930         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8931         ctsio->kern_rel_offset = 0;
8932
8933         /*
8934          * Set the IO_CONT flag, so that if this I/O gets passed to
8935          * ctl_data_submit_done(), it'll get passed back to
8936          * ctl_ctl_cnw_cont() for further processing.
8937          */
8938         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8939         ctsio->io_cont = ctl_cnw_cont;
8940
8941         lbalen = (struct ctl_lba_len_flags *)
8942             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8943         lbalen->lba = lba;
8944         lbalen->len = num_blocks;
8945         lbalen->flags = CTL_LLF_COMPARE | flags;
8946
8947         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8948         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8949         return (retval);
8950 }
8951
8952 int
8953 ctl_verify(struct ctl_scsiio *ctsio)
8954 {
8955         struct ctl_lun *lun = CTL_LUN(ctsio);
8956         struct ctl_lba_len_flags *lbalen;
8957         uint64_t lba;
8958         uint32_t num_blocks;
8959         int bytchk, flags;
8960         int retval;
8961
8962         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8963
8964         bytchk = 0;
8965         flags = CTL_LLF_FUA;
8966         switch (ctsio->cdb[0]) {
8967         case VERIFY_10: {
8968                 struct scsi_verify_10 *cdb;
8969
8970                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8971                 if (cdb->byte2 & SVFY_BYTCHK)
8972                         bytchk = 1;
8973                 if (cdb->byte2 & SVFY_DPO)
8974                         flags |= CTL_LLF_DPO;
8975                 lba = scsi_4btoul(cdb->addr);
8976                 num_blocks = scsi_2btoul(cdb->length);
8977                 break;
8978         }
8979         case VERIFY_12: {
8980                 struct scsi_verify_12 *cdb;
8981
8982                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8983                 if (cdb->byte2 & SVFY_BYTCHK)
8984                         bytchk = 1;
8985                 if (cdb->byte2 & SVFY_DPO)
8986                         flags |= CTL_LLF_DPO;
8987                 lba = scsi_4btoul(cdb->addr);
8988                 num_blocks = scsi_4btoul(cdb->length);
8989                 break;
8990         }
8991         case VERIFY_16: {
8992                 struct scsi_rw_16 *cdb;
8993
8994                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8995                 if (cdb->byte2 & SVFY_BYTCHK)
8996                         bytchk = 1;
8997                 if (cdb->byte2 & SVFY_DPO)
8998                         flags |= CTL_LLF_DPO;
8999                 lba = scsi_8btou64(cdb->addr);
9000                 num_blocks = scsi_4btoul(cdb->length);
9001                 break;
9002         }
9003         default:
9004                 /*
9005                  * We got a command we don't support.  This shouldn't
9006                  * happen, commands should be filtered out above us.
9007                  */
9008                 ctl_set_invalid_opcode(ctsio);
9009                 ctl_done((union ctl_io *)ctsio);
9010                 return (CTL_RETVAL_COMPLETE);
9011         }
9012
9013         /*
9014          * The first check is to make sure we're in bounds, the second
9015          * check is to catch wrap-around problems.  If the lba + num blocks
9016          * is less than the lba, then we've wrapped around and the block
9017          * range is invalid anyway.
9018          */
9019         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9020          || ((lba + num_blocks) < lba)) {
9021                 ctl_set_lba_out_of_range(ctsio,
9022                     MAX(lba, lun->be_lun->maxlba + 1));
9023                 ctl_done((union ctl_io *)ctsio);
9024                 return (CTL_RETVAL_COMPLETE);
9025         }
9026
9027         /*
9028          * According to SBC-3, a transfer length of 0 is not an error.
9029          */
9030         if (num_blocks == 0) {
9031                 ctl_set_success(ctsio);
9032                 ctl_done((union ctl_io *)ctsio);
9033                 return (CTL_RETVAL_COMPLETE);
9034         }
9035
9036         lbalen = (struct ctl_lba_len_flags *)
9037             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9038         lbalen->lba = lba;
9039         lbalen->len = num_blocks;
9040         if (bytchk) {
9041                 lbalen->flags = CTL_LLF_COMPARE | flags;
9042                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9043         } else {
9044                 lbalen->flags = CTL_LLF_VERIFY | flags;
9045                 ctsio->kern_total_len = 0;
9046         }
9047         ctsio->kern_rel_offset = 0;
9048
9049         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9050         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9051         return (retval);
9052 }
9053
9054 int
9055 ctl_report_luns(struct ctl_scsiio *ctsio)
9056 {
9057         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9058         struct ctl_port *port = CTL_PORT(ctsio);
9059         struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio);
9060         struct scsi_report_luns *cdb;
9061         struct scsi_report_luns_data *lun_data;
9062         int num_filled, num_luns, num_port_luns, retval;
9063         uint32_t alloc_len, lun_datalen;
9064         uint32_t initidx, targ_lun_id, lun_id;
9065
9066         retval = CTL_RETVAL_COMPLETE;
9067         cdb = (struct scsi_report_luns *)ctsio->cdb;
9068
9069         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9070
9071         num_luns = 0;
9072         num_port_luns = port->lun_map ? port->lun_map_size : ctl_max_luns;
9073         mtx_lock(&softc->ctl_lock);
9074         for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) {
9075                 if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX)
9076                         num_luns++;
9077         }
9078         mtx_unlock(&softc->ctl_lock);
9079
9080         switch (cdb->select_report) {
9081         case RPL_REPORT_DEFAULT:
9082         case RPL_REPORT_ALL:
9083         case RPL_REPORT_NONSUBSID:
9084                 break;
9085         case RPL_REPORT_WELLKNOWN:
9086         case RPL_REPORT_ADMIN:
9087         case RPL_REPORT_CONGLOM:
9088                 num_luns = 0;
9089                 break;
9090         default:
9091                 ctl_set_invalid_field(ctsio,
9092                                       /*sks_valid*/ 1,
9093                                       /*command*/ 1,
9094                                       /*field*/ 2,
9095                                       /*bit_valid*/ 0,
9096                                       /*bit*/ 0);
9097                 ctl_done((union ctl_io *)ctsio);
9098                 return (retval);
9099                 break; /* NOTREACHED */
9100         }
9101
9102         alloc_len = scsi_4btoul(cdb->length);
9103         /*
9104          * The initiator has to allocate at least 16 bytes for this request,
9105          * so he can at least get the header and the first LUN.  Otherwise
9106          * we reject the request (per SPC-3 rev 14, section 6.21).
9107          */
9108         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9109             sizeof(struct scsi_report_luns_lundata))) {
9110                 ctl_set_invalid_field(ctsio,
9111                                       /*sks_valid*/ 1,
9112                                       /*command*/ 1,
9113                                       /*field*/ 6,
9114                                       /*bit_valid*/ 0,
9115                                       /*bit*/ 0);
9116                 ctl_done((union ctl_io *)ctsio);
9117                 return (retval);
9118         }
9119
9120         lun_datalen = sizeof(*lun_data) +
9121                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9122
9123         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9124         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9125         ctsio->kern_sg_entries = 0;
9126
9127         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9128
9129         mtx_lock(&softc->ctl_lock);
9130         for (targ_lun_id = 0, num_filled = 0;
9131             targ_lun_id < num_port_luns && num_filled < num_luns;
9132             targ_lun_id++) {
9133                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9134                 if (lun_id == UINT32_MAX)
9135                         continue;
9136                 lun = softc->ctl_luns[lun_id];
9137                 if (lun == NULL)
9138                         continue;
9139
9140                 be64enc(lun_data->luns[num_filled++].lundata,
9141                     ctl_encode_lun(targ_lun_id));
9142
9143                 /*
9144                  * According to SPC-3, rev 14 section 6.21:
9145                  *
9146                  * "The execution of a REPORT LUNS command to any valid and
9147                  * installed logical unit shall clear the REPORTED LUNS DATA
9148                  * HAS CHANGED unit attention condition for all logical
9149                  * units of that target with respect to the requesting
9150                  * initiator. A valid and installed logical unit is one
9151                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9152                  * INQUIRY data (see 6.4.2)."
9153                  *
9154                  * If request_lun is NULL, the LUN this report luns command
9155                  * was issued to is either disabled or doesn't exist. In that
9156                  * case, we shouldn't clear any pending lun change unit
9157                  * attention.
9158                  */
9159                 if (request_lun != NULL) {
9160                         mtx_lock(&lun->lun_lock);
9161                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9162                         mtx_unlock(&lun->lun_lock);
9163                 }
9164         }
9165         mtx_unlock(&softc->ctl_lock);
9166
9167         /*
9168          * It's quite possible that we've returned fewer LUNs than we allocated
9169          * space for.  Trim it.
9170          */
9171         lun_datalen = sizeof(*lun_data) +
9172                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9173         ctsio->kern_rel_offset = 0;
9174         ctsio->kern_sg_entries = 0;
9175         ctsio->kern_data_len = min(lun_datalen, alloc_len);
9176         ctsio->kern_total_len = ctsio->kern_data_len;
9177
9178         /*
9179          * We set this to the actual data length, regardless of how much
9180          * space we actually have to return results.  If the user looks at
9181          * this value, he'll know whether or not he allocated enough space
9182          * and reissue the command if necessary.  We don't support well
9183          * known logical units, so if the user asks for that, return none.
9184          */
9185         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9186
9187         /*
9188          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9189          * this request.
9190          */
9191         ctl_set_success(ctsio);
9192         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9193         ctsio->be_move_done = ctl_config_move_done;
9194         ctl_datamove((union ctl_io *)ctsio);
9195         return (retval);
9196 }
9197
9198 int
9199 ctl_request_sense(struct ctl_scsiio *ctsio)
9200 {
9201         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9202         struct ctl_lun *lun = CTL_LUN(ctsio);
9203         struct scsi_request_sense *cdb;
9204         struct scsi_sense_data *sense_ptr, *ps;
9205         uint32_t initidx;
9206         int have_error;
9207         u_int sense_len = SSD_FULL_SIZE;
9208         scsi_sense_data_type sense_format;
9209         ctl_ua_type ua_type;
9210         uint8_t asc = 0, ascq = 0;
9211
9212         cdb = (struct scsi_request_sense *)ctsio->cdb;
9213
9214         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9215
9216         /*
9217          * Determine which sense format the user wants.
9218          */
9219         if (cdb->byte2 & SRS_DESC)
9220                 sense_format = SSD_TYPE_DESC;
9221         else
9222                 sense_format = SSD_TYPE_FIXED;
9223
9224         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9225         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9226         ctsio->kern_sg_entries = 0;
9227         ctsio->kern_rel_offset = 0;
9228
9229         /*
9230          * struct scsi_sense_data, which is currently set to 256 bytes, is
9231          * larger than the largest allowed value for the length field in the
9232          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9233          */
9234         ctsio->kern_data_len = cdb->length;
9235         ctsio->kern_total_len = cdb->length;
9236
9237         /*
9238          * If we don't have a LUN, we don't have any pending sense.
9239          */
9240         if (lun == NULL ||
9241             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9242              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9243                 /* "Logical unit not supported" */
9244                 ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9245                     /*current_error*/ 1,
9246                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9247                     /*asc*/ 0x25,
9248                     /*ascq*/ 0x00,
9249                     SSD_ELEM_NONE);
9250                 goto send;
9251         }
9252
9253         have_error = 0;
9254         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9255         /*
9256          * Check for pending sense, and then for pending unit attentions.
9257          * Pending sense gets returned first, then pending unit attentions.
9258          */
9259         mtx_lock(&lun->lun_lock);
9260         ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
9261         if (ps != NULL)
9262                 ps += initidx % CTL_MAX_INIT_PER_PORT;
9263         if (ps != NULL && ps->error_code != 0) {
9264                 scsi_sense_data_type stored_format;
9265
9266                 /*
9267                  * Check to see which sense format was used for the stored
9268                  * sense data.
9269                  */
9270                 stored_format = scsi_sense_type(ps);
9271
9272                 /*
9273                  * If the user requested a different sense format than the
9274                  * one we stored, then we need to convert it to the other
9275                  * format.  If we're going from descriptor to fixed format
9276                  * sense data, we may lose things in translation, depending
9277                  * on what options were used.
9278                  *
9279                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9280                  * for some reason we'll just copy it out as-is.
9281                  */
9282                 if ((stored_format == SSD_TYPE_FIXED)
9283                  && (sense_format == SSD_TYPE_DESC))
9284                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9285                             ps, (struct scsi_sense_data_desc *)sense_ptr);
9286                 else if ((stored_format == SSD_TYPE_DESC)
9287                       && (sense_format == SSD_TYPE_FIXED))
9288                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9289                             ps, (struct scsi_sense_data_fixed *)sense_ptr);
9290                 else
9291                         memcpy(sense_ptr, ps, sizeof(*sense_ptr));
9292
9293                 ps->error_code = 0;
9294                 have_error = 1;
9295         } else {
9296                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9297                     sense_format);
9298                 if (ua_type != CTL_UA_NONE)
9299                         have_error = 1;
9300         }
9301         if (have_error == 0) {
9302                 /*
9303                  * Report informational exception if have one and allowed.
9304                  */
9305                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9306                         asc = lun->ie_asc;
9307                         ascq = lun->ie_ascq;
9308                 }
9309                 ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9310                     /*current_error*/ 1,
9311                     /*sense_key*/ SSD_KEY_NO_SENSE,
9312                     /*asc*/ asc,
9313                     /*ascq*/ ascq,
9314                     SSD_ELEM_NONE);
9315         }
9316         mtx_unlock(&lun->lun_lock);
9317
9318 send:
9319         /*
9320          * We report the SCSI status as OK, since the status of the command
9321          * itself is OK.  We're reporting sense as parameter data.
9322          */
9323         ctl_set_success(ctsio);
9324         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9325         ctsio->be_move_done = ctl_config_move_done;
9326         ctl_datamove((union ctl_io *)ctsio);
9327         return (CTL_RETVAL_COMPLETE);
9328 }
9329
9330 int
9331 ctl_tur(struct ctl_scsiio *ctsio)
9332 {
9333
9334         CTL_DEBUG_PRINT(("ctl_tur\n"));
9335
9336         ctl_set_success(ctsio);
9337         ctl_done((union ctl_io *)ctsio);
9338
9339         return (CTL_RETVAL_COMPLETE);
9340 }
9341
9342 /*
9343  * SCSI VPD page 0x00, the Supported VPD Pages page.
9344  */
9345 static int
9346 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9347 {
9348         struct ctl_lun *lun = CTL_LUN(ctsio);
9349         struct scsi_vpd_supported_pages *pages;
9350         int sup_page_size;
9351         int p;
9352
9353         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9354             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9355         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9356         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9357         ctsio->kern_rel_offset = 0;
9358         ctsio->kern_sg_entries = 0;
9359         ctsio->kern_data_len = min(sup_page_size, alloc_len);
9360         ctsio->kern_total_len = ctsio->kern_data_len;
9361
9362         /*
9363          * The control device is always connected.  The disk device, on the
9364          * other hand, may not be online all the time.  Need to change this
9365          * to figure out whether the disk device is actually online or not.
9366          */
9367         if (lun != NULL)
9368                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9369                                 lun->be_lun->lun_type;
9370         else
9371                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9372
9373         p = 0;
9374         /* Supported VPD pages */
9375         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9376         /* Serial Number */
9377         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9378         /* Device Identification */
9379         pages->page_list[p++] = SVPD_DEVICE_ID;
9380         /* Extended INQUIRY Data */
9381         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9382         /* Mode Page Policy */
9383         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9384         /* SCSI Ports */
9385         pages->page_list[p++] = SVPD_SCSI_PORTS;
9386         /* Third-party Copy */
9387         pages->page_list[p++] = SVPD_SCSI_TPC;
9388         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9389                 /* Block limits */
9390                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9391                 /* Block Device Characteristics */
9392                 pages->page_list[p++] = SVPD_BDC;
9393                 /* Logical Block Provisioning */
9394                 pages->page_list[p++] = SVPD_LBP;
9395         }
9396         pages->length = p;
9397
9398         ctl_set_success(ctsio);
9399         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9400         ctsio->be_move_done = ctl_config_move_done;
9401         ctl_datamove((union ctl_io *)ctsio);
9402         return (CTL_RETVAL_COMPLETE);
9403 }
9404
9405 /*
9406  * SCSI VPD page 0x80, the Unit Serial Number page.
9407  */
9408 static int
9409 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9410 {
9411         struct ctl_lun *lun = CTL_LUN(ctsio);
9412         struct scsi_vpd_unit_serial_number *sn_ptr;
9413         int data_len;
9414
9415         data_len = 4 + CTL_SN_LEN;
9416         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9417         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9418         ctsio->kern_rel_offset = 0;
9419         ctsio->kern_sg_entries = 0;
9420         ctsio->kern_data_len = min(data_len, alloc_len);
9421         ctsio->kern_total_len = ctsio->kern_data_len;
9422
9423         /*
9424          * The control device is always connected.  The disk device, on the
9425          * other hand, may not be online all the time.  Need to change this
9426          * to figure out whether the disk device is actually online or not.
9427          */
9428         if (lun != NULL)
9429                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9430                                   lun->be_lun->lun_type;
9431         else
9432                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9433
9434         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9435         sn_ptr->length = CTL_SN_LEN;
9436         /*
9437          * If we don't have a LUN, we just leave the serial number as
9438          * all spaces.
9439          */
9440         if (lun != NULL) {
9441                 strncpy((char *)sn_ptr->serial_num,
9442                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9443         } else
9444                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9445
9446         ctl_set_success(ctsio);
9447         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9448         ctsio->be_move_done = ctl_config_move_done;
9449         ctl_datamove((union ctl_io *)ctsio);
9450         return (CTL_RETVAL_COMPLETE);
9451 }
9452
9453
9454 /*
9455  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9456  */
9457 static int
9458 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9459 {
9460         struct ctl_lun *lun = CTL_LUN(ctsio);
9461         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9462         int data_len;
9463
9464         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9465         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9466         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9467         ctsio->kern_sg_entries = 0;
9468         ctsio->kern_rel_offset = 0;
9469         ctsio->kern_data_len = min(data_len, alloc_len);
9470         ctsio->kern_total_len = ctsio->kern_data_len;
9471
9472         /*
9473          * The control device is always connected.  The disk device, on the
9474          * other hand, may not be online all the time.
9475          */
9476         if (lun != NULL)
9477                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9478                                      lun->be_lun->lun_type;
9479         else
9480                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9481         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9482         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9483         /*
9484          * We support head of queue, ordered and simple tags.
9485          */
9486         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9487         /*
9488          * Volatile cache supported.
9489          */
9490         eid_ptr->flags3 = SVPD_EID_V_SUP;
9491
9492         /*
9493          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9494          * attention for a particular IT nexus on all LUNs once we report
9495          * it to that nexus once.  This bit is required as of SPC-4.
9496          */
9497         eid_ptr->flags4 = SVPD_EID_LUICLR;
9498
9499         /*
9500          * We support revert to defaults (RTD) bit in MODE SELECT.
9501          */
9502         eid_ptr->flags5 = SVPD_EID_RTD_SUP;
9503
9504         /*
9505          * XXX KDM in order to correctly answer this, we would need
9506          * information from the SIM to determine how much sense data it
9507          * can send.  So this would really be a path inquiry field, most
9508          * likely.  This can be set to a maximum of 252 according to SPC-4,
9509          * but the hardware may or may not be able to support that much.
9510          * 0 just means that the maximum sense data length is not reported.
9511          */
9512         eid_ptr->max_sense_length = 0;
9513
9514         ctl_set_success(ctsio);
9515         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9516         ctsio->be_move_done = ctl_config_move_done;
9517         ctl_datamove((union ctl_io *)ctsio);
9518         return (CTL_RETVAL_COMPLETE);
9519 }
9520
9521 static int
9522 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9523 {
9524         struct ctl_lun *lun = CTL_LUN(ctsio);
9525         struct scsi_vpd_mode_page_policy *mpp_ptr;
9526         int data_len;
9527
9528         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9529             sizeof(struct scsi_vpd_mode_page_policy_descr);
9530
9531         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9532         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9533         ctsio->kern_rel_offset = 0;
9534         ctsio->kern_sg_entries = 0;
9535         ctsio->kern_data_len = min(data_len, alloc_len);
9536         ctsio->kern_total_len = ctsio->kern_data_len;
9537
9538         /*
9539          * The control device is always connected.  The disk device, on the
9540          * other hand, may not be online all the time.
9541          */
9542         if (lun != NULL)
9543                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9544                                      lun->be_lun->lun_type;
9545         else
9546                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9547         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9548         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9549         mpp_ptr->descr[0].page_code = 0x3f;
9550         mpp_ptr->descr[0].subpage_code = 0xff;
9551         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9552
9553         ctl_set_success(ctsio);
9554         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9555         ctsio->be_move_done = ctl_config_move_done;
9556         ctl_datamove((union ctl_io *)ctsio);
9557         return (CTL_RETVAL_COMPLETE);
9558 }
9559
9560 /*
9561  * SCSI VPD page 0x83, the Device Identification page.
9562  */
9563 static int
9564 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9565 {
9566         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9567         struct ctl_port *port = CTL_PORT(ctsio);
9568         struct ctl_lun *lun = CTL_LUN(ctsio);
9569         struct scsi_vpd_device_id *devid_ptr;
9570         struct scsi_vpd_id_descriptor *desc;
9571         int data_len, g;
9572         uint8_t proto;
9573
9574         data_len = sizeof(struct scsi_vpd_device_id) +
9575             sizeof(struct scsi_vpd_id_descriptor) +
9576                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9577             sizeof(struct scsi_vpd_id_descriptor) +
9578                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9579         if (lun && lun->lun_devid)
9580                 data_len += lun->lun_devid->len;
9581         if (port && port->port_devid)
9582                 data_len += port->port_devid->len;
9583         if (port && port->target_devid)
9584                 data_len += port->target_devid->len;
9585
9586         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9587         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9588         ctsio->kern_sg_entries = 0;
9589         ctsio->kern_rel_offset = 0;
9590         ctsio->kern_sg_entries = 0;
9591         ctsio->kern_data_len = min(data_len, alloc_len);
9592         ctsio->kern_total_len = ctsio->kern_data_len;
9593
9594         /*
9595          * The control device is always connected.  The disk device, on the
9596          * other hand, may not be online all the time.
9597          */
9598         if (lun != NULL)
9599                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9600                                      lun->be_lun->lun_type;
9601         else
9602                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9603         devid_ptr->page_code = SVPD_DEVICE_ID;
9604         scsi_ulto2b(data_len - 4, devid_ptr->length);
9605
9606         if (port && port->port_type == CTL_PORT_FC)
9607                 proto = SCSI_PROTO_FC << 4;
9608         else if (port && port->port_type == CTL_PORT_SAS)
9609                 proto = SCSI_PROTO_SAS << 4;
9610         else if (port && port->port_type == CTL_PORT_ISCSI)
9611                 proto = SCSI_PROTO_ISCSI << 4;
9612         else
9613                 proto = SCSI_PROTO_SPI << 4;
9614         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9615
9616         /*
9617          * We're using a LUN association here.  i.e., this device ID is a
9618          * per-LUN identifier.
9619          */
9620         if (lun && lun->lun_devid) {
9621                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9622                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9623                     lun->lun_devid->len);
9624         }
9625
9626         /*
9627          * This is for the WWPN which is a port association.
9628          */
9629         if (port && port->port_devid) {
9630                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9631                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9632                     port->port_devid->len);
9633         }
9634
9635         /*
9636          * This is for the Relative Target Port(type 4h) identifier
9637          */
9638         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9639         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9640             SVPD_ID_TYPE_RELTARG;
9641         desc->length = 4;
9642         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9643         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9644             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9645
9646         /*
9647          * This is for the Target Port Group(type 5h) identifier
9648          */
9649         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9650         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9651             SVPD_ID_TYPE_TPORTGRP;
9652         desc->length = 4;
9653         if (softc->is_single ||
9654             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9655                 g = 1;
9656         else
9657                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9658         scsi_ulto2b(g, &desc->identifier[2]);
9659         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9660             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9661
9662         /*
9663          * This is for the Target identifier
9664          */
9665         if (port && port->target_devid) {
9666                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9667         }
9668
9669         ctl_set_success(ctsio);
9670         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9671         ctsio->be_move_done = ctl_config_move_done;
9672         ctl_datamove((union ctl_io *)ctsio);
9673         return (CTL_RETVAL_COMPLETE);
9674 }
9675
9676 static int
9677 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9678 {
9679         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9680         struct ctl_lun *lun = CTL_LUN(ctsio);
9681         struct scsi_vpd_scsi_ports *sp;
9682         struct scsi_vpd_port_designation *pd;
9683         struct scsi_vpd_port_designation_cont *pdc;
9684         struct ctl_port *port;
9685         int data_len, num_target_ports, iid_len, id_len;
9686
9687         num_target_ports = 0;
9688         iid_len = 0;
9689         id_len = 0;
9690         mtx_lock(&softc->ctl_lock);
9691         STAILQ_FOREACH(port, &softc->port_list, links) {
9692                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9693                         continue;
9694                 if (lun != NULL &&
9695                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9696                         continue;
9697                 num_target_ports++;
9698                 if (port->init_devid)
9699                         iid_len += port->init_devid->len;
9700                 if (port->port_devid)
9701                         id_len += port->port_devid->len;
9702         }
9703         mtx_unlock(&softc->ctl_lock);
9704
9705         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9706             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9707              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9708         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9709         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9710         ctsio->kern_sg_entries = 0;
9711         ctsio->kern_rel_offset = 0;
9712         ctsio->kern_sg_entries = 0;
9713         ctsio->kern_data_len = min(data_len, alloc_len);
9714         ctsio->kern_total_len = ctsio->kern_data_len;
9715
9716         /*
9717          * The control device is always connected.  The disk device, on the
9718          * other hand, may not be online all the time.  Need to change this
9719          * to figure out whether the disk device is actually online or not.
9720          */
9721         if (lun != NULL)
9722                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9723                                   lun->be_lun->lun_type;
9724         else
9725                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9726
9727         sp->page_code = SVPD_SCSI_PORTS;
9728         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9729             sp->page_length);
9730         pd = &sp->design[0];
9731
9732         mtx_lock(&softc->ctl_lock);
9733         STAILQ_FOREACH(port, &softc->port_list, links) {
9734                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9735                         continue;
9736                 if (lun != NULL &&
9737                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9738                         continue;
9739                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9740                 if (port->init_devid) {
9741                         iid_len = port->init_devid->len;
9742                         memcpy(pd->initiator_transportid,
9743                             port->init_devid->data, port->init_devid->len);
9744                 } else
9745                         iid_len = 0;
9746                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9747                 pdc = (struct scsi_vpd_port_designation_cont *)
9748                     (&pd->initiator_transportid[iid_len]);
9749                 if (port->port_devid) {
9750                         id_len = port->port_devid->len;
9751                         memcpy(pdc->target_port_descriptors,
9752                             port->port_devid->data, port->port_devid->len);
9753                 } else
9754                         id_len = 0;
9755                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9756                 pd = (struct scsi_vpd_port_designation *)
9757                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9758         }
9759         mtx_unlock(&softc->ctl_lock);
9760
9761         ctl_set_success(ctsio);
9762         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9763         ctsio->be_move_done = ctl_config_move_done;
9764         ctl_datamove((union ctl_io *)ctsio);
9765         return (CTL_RETVAL_COMPLETE);
9766 }
9767
9768 static int
9769 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9770 {
9771         struct ctl_lun *lun = CTL_LUN(ctsio);
9772         struct scsi_vpd_block_limits *bl_ptr;
9773         uint64_t ival;
9774
9775         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9776         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9777         ctsio->kern_sg_entries = 0;
9778         ctsio->kern_rel_offset = 0;
9779         ctsio->kern_sg_entries = 0;
9780         ctsio->kern_data_len = min(sizeof(*bl_ptr), alloc_len);
9781         ctsio->kern_total_len = ctsio->kern_data_len;
9782
9783         /*
9784          * The control device is always connected.  The disk device, on the
9785          * other hand, may not be online all the time.  Need to change this
9786          * to figure out whether the disk device is actually online or not.
9787          */
9788         if (lun != NULL)
9789                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9790                                   lun->be_lun->lun_type;
9791         else
9792                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9793
9794         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9795         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9796         bl_ptr->max_cmp_write_len = 0xff;
9797         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9798         if (lun != NULL) {
9799                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9800                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9801                         ival = 0xffffffff;
9802                         ctl_get_opt_number(&lun->be_lun->options,
9803                             "unmap_max_lba", &ival);
9804                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9805                         ival = 0xffffffff;
9806                         ctl_get_opt_number(&lun->be_lun->options,
9807                             "unmap_max_descr", &ival);
9808                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9809                         if (lun->be_lun->ublockexp != 0) {
9810                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9811                                     bl_ptr->opt_unmap_grain);
9812                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9813                                     bl_ptr->unmap_grain_align);
9814                         }
9815                 }
9816                 scsi_ulto4b(lun->be_lun->atomicblock,
9817                     bl_ptr->max_atomic_transfer_length);
9818                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9819                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9820                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9821                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9822                 ival = UINT64_MAX;
9823                 ctl_get_opt_number(&lun->be_lun->options, "write_same_max_lba", &ival);
9824                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9825         }
9826
9827         ctl_set_success(ctsio);
9828         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9829         ctsio->be_move_done = ctl_config_move_done;
9830         ctl_datamove((union ctl_io *)ctsio);
9831         return (CTL_RETVAL_COMPLETE);
9832 }
9833
9834 static int
9835 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9836 {
9837         struct ctl_lun *lun = CTL_LUN(ctsio);
9838         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9839         const char *value;
9840         u_int i;
9841
9842         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9843         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9844         ctsio->kern_sg_entries = 0;
9845         ctsio->kern_rel_offset = 0;
9846         ctsio->kern_data_len = min(sizeof(*bdc_ptr), alloc_len);
9847         ctsio->kern_total_len = ctsio->kern_data_len;
9848
9849         /*
9850          * The control device is always connected.  The disk device, on the
9851          * other hand, may not be online all the time.  Need to change this
9852          * to figure out whether the disk device is actually online or not.
9853          */
9854         if (lun != NULL)
9855                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9856                                   lun->be_lun->lun_type;
9857         else
9858                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9859         bdc_ptr->page_code = SVPD_BDC;
9860         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9861         if (lun != NULL &&
9862             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9863                 i = strtol(value, NULL, 0);
9864         else
9865                 i = CTL_DEFAULT_ROTATION_RATE;
9866         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9867         if (lun != NULL &&
9868             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9869                 i = strtol(value, NULL, 0);
9870         else
9871                 i = 0;
9872         bdc_ptr->wab_wac_ff = (i & 0x0f);
9873         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9874
9875         ctl_set_success(ctsio);
9876         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9877         ctsio->be_move_done = ctl_config_move_done;
9878         ctl_datamove((union ctl_io *)ctsio);
9879         return (CTL_RETVAL_COMPLETE);
9880 }
9881
9882 static int
9883 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9884 {
9885         struct ctl_lun *lun = CTL_LUN(ctsio);
9886         struct scsi_vpd_logical_block_prov *lbp_ptr;
9887         const char *value;
9888
9889         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9890         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9891         ctsio->kern_sg_entries = 0;
9892         ctsio->kern_rel_offset = 0;
9893         ctsio->kern_data_len = min(sizeof(*lbp_ptr), alloc_len);
9894         ctsio->kern_total_len = ctsio->kern_data_len;
9895
9896         /*
9897          * The control device is always connected.  The disk device, on the
9898          * other hand, may not be online all the time.  Need to change this
9899          * to figure out whether the disk device is actually online or not.
9900          */
9901         if (lun != NULL)
9902                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9903                                   lun->be_lun->lun_type;
9904         else
9905                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9906
9907         lbp_ptr->page_code = SVPD_LBP;
9908         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9909         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
9910         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9911                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
9912                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
9913                 value = ctl_get_opt(&lun->be_lun->options, "provisioning_type");
9914                 if (value != NULL) {
9915                         if (strcmp(value, "resource") == 0)
9916                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
9917                         else if (strcmp(value, "thin") == 0)
9918                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
9919                 } else
9920                         lbp_ptr->prov_type = SVPD_LBP_THIN;
9921         }
9922
9923         ctl_set_success(ctsio);
9924         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9925         ctsio->be_move_done = ctl_config_move_done;
9926         ctl_datamove((union ctl_io *)ctsio);
9927         return (CTL_RETVAL_COMPLETE);
9928 }
9929
9930 /*
9931  * INQUIRY with the EVPD bit set.
9932  */
9933 static int
9934 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9935 {
9936         struct ctl_lun *lun = CTL_LUN(ctsio);
9937         struct scsi_inquiry *cdb;
9938         int alloc_len, retval;
9939
9940         cdb = (struct scsi_inquiry *)ctsio->cdb;
9941         alloc_len = scsi_2btoul(cdb->length);
9942
9943         switch (cdb->page_code) {
9944         case SVPD_SUPPORTED_PAGES:
9945                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9946                 break;
9947         case SVPD_UNIT_SERIAL_NUMBER:
9948                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9949                 break;
9950         case SVPD_DEVICE_ID:
9951                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9952                 break;
9953         case SVPD_EXTENDED_INQUIRY_DATA:
9954                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
9955                 break;
9956         case SVPD_MODE_PAGE_POLICY:
9957                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
9958                 break;
9959         case SVPD_SCSI_PORTS:
9960                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
9961                 break;
9962         case SVPD_SCSI_TPC:
9963                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
9964                 break;
9965         case SVPD_BLOCK_LIMITS:
9966                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9967                         goto err;
9968                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9969                 break;
9970         case SVPD_BDC:
9971                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9972                         goto err;
9973                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
9974                 break;
9975         case SVPD_LBP:
9976                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9977                         goto err;
9978                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9979                 break;
9980         default:
9981 err:
9982                 ctl_set_invalid_field(ctsio,
9983                                       /*sks_valid*/ 1,
9984                                       /*command*/ 1,
9985                                       /*field*/ 2,
9986                                       /*bit_valid*/ 0,
9987                                       /*bit*/ 0);
9988                 ctl_done((union ctl_io *)ctsio);
9989                 retval = CTL_RETVAL_COMPLETE;
9990                 break;
9991         }
9992
9993         return (retval);
9994 }
9995
9996 /*
9997  * Standard INQUIRY data.
9998  */
9999 static int
10000 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10001 {
10002         struct ctl_softc *softc = CTL_SOFTC(ctsio);
10003         struct ctl_port *port = CTL_PORT(ctsio);
10004         struct ctl_lun *lun = CTL_LUN(ctsio);
10005         struct scsi_inquiry_data *inq_ptr;
10006         struct scsi_inquiry *cdb;
10007         char *val;
10008         uint32_t alloc_len, data_len;
10009         ctl_port_type port_type;
10010
10011         port_type = port->port_type;
10012         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10013                 port_type = CTL_PORT_SCSI;
10014
10015         cdb = (struct scsi_inquiry *)ctsio->cdb;
10016         alloc_len = scsi_2btoul(cdb->length);
10017
10018         /*
10019          * We malloc the full inquiry data size here and fill it
10020          * in.  If the user only asks for less, we'll give him
10021          * that much.
10022          */
10023         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10024         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10025         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10026         ctsio->kern_sg_entries = 0;
10027         ctsio->kern_rel_offset = 0;
10028         ctsio->kern_data_len = min(data_len, alloc_len);
10029         ctsio->kern_total_len = ctsio->kern_data_len;
10030
10031         if (lun != NULL) {
10032                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10033                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10034                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10035                             lun->be_lun->lun_type;
10036                 } else {
10037                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10038                             lun->be_lun->lun_type;
10039                 }
10040                 if (lun->flags & CTL_LUN_REMOVABLE)
10041                         inq_ptr->dev_qual2 |= SID_RMB;
10042         } else
10043                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10044
10045         /* RMB in byte 2 is 0 */
10046         inq_ptr->version = SCSI_REV_SPC5;
10047
10048         /*
10049          * According to SAM-3, even if a device only supports a single
10050          * level of LUN addressing, it should still set the HISUP bit:
10051          *
10052          * 4.9.1 Logical unit numbers overview
10053          *
10054          * All logical unit number formats described in this standard are
10055          * hierarchical in structure even when only a single level in that
10056          * hierarchy is used. The HISUP bit shall be set to one in the
10057          * standard INQUIRY data (see SPC-2) when any logical unit number
10058          * format described in this standard is used.  Non-hierarchical
10059          * formats are outside the scope of this standard.
10060          *
10061          * Therefore we set the HiSup bit here.
10062          *
10063          * The response format is 2, per SPC-3.
10064          */
10065         inq_ptr->response_format = SID_HiSup | 2;
10066
10067         inq_ptr->additional_length = data_len -
10068             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10069         CTL_DEBUG_PRINT(("additional_length = %d\n",
10070                          inq_ptr->additional_length));
10071
10072         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10073         if (port_type == CTL_PORT_SCSI)
10074                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10075         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10076         inq_ptr->flags = SID_CmdQue;
10077         if (port_type == CTL_PORT_SCSI)
10078                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10079
10080         /*
10081          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10082          * We have 8 bytes for the vendor name, and 16 bytes for the device
10083          * name and 4 bytes for the revision.
10084          */
10085         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10086             "vendor")) == NULL) {
10087                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10088         } else {
10089                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10090                 strncpy(inq_ptr->vendor, val,
10091                     min(sizeof(inq_ptr->vendor), strlen(val)));
10092         }
10093         if (lun == NULL) {
10094                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10095                     sizeof(inq_ptr->product));
10096         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10097                 switch (lun->be_lun->lun_type) {
10098                 case T_DIRECT:
10099                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10100                             sizeof(inq_ptr->product));
10101                         break;
10102                 case T_PROCESSOR:
10103                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10104                             sizeof(inq_ptr->product));
10105                         break;
10106                 case T_CDROM:
10107                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10108                             sizeof(inq_ptr->product));
10109                         break;
10110                 default:
10111                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10112                             sizeof(inq_ptr->product));
10113                         break;
10114                 }
10115         } else {
10116                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10117                 strncpy(inq_ptr->product, val,
10118                     min(sizeof(inq_ptr->product), strlen(val)));
10119         }
10120
10121         /*
10122          * XXX make this a macro somewhere so it automatically gets
10123          * incremented when we make changes.
10124          */
10125         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10126             "revision")) == NULL) {
10127                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10128         } else {
10129                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10130                 strncpy(inq_ptr->revision, val,
10131                     min(sizeof(inq_ptr->revision), strlen(val)));
10132         }
10133
10134         /*
10135          * For parallel SCSI, we support double transition and single
10136          * transition clocking.  We also support QAS (Quick Arbitration
10137          * and Selection) and Information Unit transfers on both the
10138          * control and array devices.
10139          */
10140         if (port_type == CTL_PORT_SCSI)
10141                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10142                                     SID_SPI_IUS;
10143
10144         /* SAM-6 (no version claimed) */
10145         scsi_ulto2b(0x00C0, inq_ptr->version1);
10146         /* SPC-5 (no version claimed) */
10147         scsi_ulto2b(0x05C0, inq_ptr->version2);
10148         if (port_type == CTL_PORT_FC) {
10149                 /* FCP-2 ANSI INCITS.350:2003 */
10150                 scsi_ulto2b(0x0917, inq_ptr->version3);
10151         } else if (port_type == CTL_PORT_SCSI) {
10152                 /* SPI-4 ANSI INCITS.362:200x */
10153                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10154         } else if (port_type == CTL_PORT_ISCSI) {
10155                 /* iSCSI (no version claimed) */
10156                 scsi_ulto2b(0x0960, inq_ptr->version3);
10157         } else if (port_type == CTL_PORT_SAS) {
10158                 /* SAS (no version claimed) */
10159                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10160         } else if (port_type == CTL_PORT_UMASS) {
10161                 /* USB Mass Storage Class Bulk-Only Transport, Revision 1.0 */
10162                 scsi_ulto2b(0x1730, inq_ptr->version3);
10163         }
10164
10165         if (lun == NULL) {
10166                 /* SBC-4 (no version claimed) */
10167                 scsi_ulto2b(0x0600, inq_ptr->version4);
10168         } else {
10169                 switch (lun->be_lun->lun_type) {
10170                 case T_DIRECT:
10171                         /* SBC-4 (no version claimed) */
10172                         scsi_ulto2b(0x0600, inq_ptr->version4);
10173                         break;
10174                 case T_PROCESSOR:
10175                         break;
10176                 case T_CDROM:
10177                         /* MMC-6 (no version claimed) */
10178                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10179                         break;
10180                 default:
10181                         break;
10182                 }
10183         }
10184
10185         ctl_set_success(ctsio);
10186         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10187         ctsio->be_move_done = ctl_config_move_done;
10188         ctl_datamove((union ctl_io *)ctsio);
10189         return (CTL_RETVAL_COMPLETE);
10190 }
10191
10192 int
10193 ctl_inquiry(struct ctl_scsiio *ctsio)
10194 {
10195         struct scsi_inquiry *cdb;
10196         int retval;
10197
10198         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10199
10200         cdb = (struct scsi_inquiry *)ctsio->cdb;
10201         if (cdb->byte2 & SI_EVPD)
10202                 retval = ctl_inquiry_evpd(ctsio);
10203         else if (cdb->page_code == 0)
10204                 retval = ctl_inquiry_std(ctsio);
10205         else {
10206                 ctl_set_invalid_field(ctsio,
10207                                       /*sks_valid*/ 1,
10208                                       /*command*/ 1,
10209                                       /*field*/ 2,
10210                                       /*bit_valid*/ 0,
10211                                       /*bit*/ 0);
10212                 ctl_done((union ctl_io *)ctsio);
10213                 return (CTL_RETVAL_COMPLETE);
10214         }
10215
10216         return (retval);
10217 }
10218
10219 int
10220 ctl_get_config(struct ctl_scsiio *ctsio)
10221 {
10222         struct ctl_lun *lun = CTL_LUN(ctsio);
10223         struct scsi_get_config_header *hdr;
10224         struct scsi_get_config_feature *feature;
10225         struct scsi_get_config *cdb;
10226         uint32_t alloc_len, data_len;
10227         int rt, starting;
10228
10229         cdb = (struct scsi_get_config *)ctsio->cdb;
10230         rt = (cdb->rt & SGC_RT_MASK);
10231         starting = scsi_2btoul(cdb->starting_feature);
10232         alloc_len = scsi_2btoul(cdb->length);
10233
10234         data_len = sizeof(struct scsi_get_config_header) +
10235             sizeof(struct scsi_get_config_feature) + 8 +
10236             sizeof(struct scsi_get_config_feature) + 8 +
10237             sizeof(struct scsi_get_config_feature) + 4 +
10238             sizeof(struct scsi_get_config_feature) + 4 +
10239             sizeof(struct scsi_get_config_feature) + 8 +
10240             sizeof(struct scsi_get_config_feature) +
10241             sizeof(struct scsi_get_config_feature) + 4 +
10242             sizeof(struct scsi_get_config_feature) + 4 +
10243             sizeof(struct scsi_get_config_feature) + 4 +
10244             sizeof(struct scsi_get_config_feature) + 4 +
10245             sizeof(struct scsi_get_config_feature) + 4 +
10246             sizeof(struct scsi_get_config_feature) + 4;
10247         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10248         ctsio->kern_sg_entries = 0;
10249         ctsio->kern_rel_offset = 0;
10250
10251         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10252         if (lun->flags & CTL_LUN_NO_MEDIA)
10253                 scsi_ulto2b(0x0000, hdr->current_profile);
10254         else
10255                 scsi_ulto2b(0x0010, hdr->current_profile);
10256         feature = (struct scsi_get_config_feature *)(hdr + 1);
10257
10258         if (starting > 0x003b)
10259                 goto done;
10260         if (starting > 0x003a)
10261                 goto f3b;
10262         if (starting > 0x002b)
10263                 goto f3a;
10264         if (starting > 0x002a)
10265                 goto f2b;
10266         if (starting > 0x001f)
10267                 goto f2a;
10268         if (starting > 0x001e)
10269                 goto f1f;
10270         if (starting > 0x001d)
10271                 goto f1e;
10272         if (starting > 0x0010)
10273                 goto f1d;
10274         if (starting > 0x0003)
10275                 goto f10;
10276         if (starting > 0x0002)
10277                 goto f3;
10278         if (starting > 0x0001)
10279                 goto f2;
10280         if (starting > 0x0000)
10281                 goto f1;
10282
10283         /* Profile List */
10284         scsi_ulto2b(0x0000, feature->feature_code);
10285         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10286         feature->add_length = 8;
10287         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10288         feature->feature_data[2] = 0x00;
10289         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10290         feature->feature_data[6] = 0x01;
10291         feature = (struct scsi_get_config_feature *)
10292             &feature->feature_data[feature->add_length];
10293
10294 f1:     /* Core */
10295         scsi_ulto2b(0x0001, feature->feature_code);
10296         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10297         feature->add_length = 8;
10298         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10299         feature->feature_data[4] = 0x03;
10300         feature = (struct scsi_get_config_feature *)
10301             &feature->feature_data[feature->add_length];
10302
10303 f2:     /* Morphing */
10304         scsi_ulto2b(0x0002, feature->feature_code);
10305         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10306         feature->add_length = 4;
10307         feature->feature_data[0] = 0x02;
10308         feature = (struct scsi_get_config_feature *)
10309             &feature->feature_data[feature->add_length];
10310
10311 f3:     /* Removable Medium */
10312         scsi_ulto2b(0x0003, feature->feature_code);
10313         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10314         feature->add_length = 4;
10315         feature->feature_data[0] = 0x39;
10316         feature = (struct scsi_get_config_feature *)
10317             &feature->feature_data[feature->add_length];
10318
10319         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10320                 goto done;
10321
10322 f10:    /* Random Read */
10323         scsi_ulto2b(0x0010, feature->feature_code);
10324         feature->flags = 0x00;
10325         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10326                 feature->flags |= SGC_F_CURRENT;
10327         feature->add_length = 8;
10328         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10329         scsi_ulto2b(1, &feature->feature_data[4]);
10330         feature->feature_data[6] = 0x00;
10331         feature = (struct scsi_get_config_feature *)
10332             &feature->feature_data[feature->add_length];
10333
10334 f1d:    /* Multi-Read */
10335         scsi_ulto2b(0x001D, feature->feature_code);
10336         feature->flags = 0x00;
10337         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10338                 feature->flags |= SGC_F_CURRENT;
10339         feature->add_length = 0;
10340         feature = (struct scsi_get_config_feature *)
10341             &feature->feature_data[feature->add_length];
10342
10343 f1e:    /* CD Read */
10344         scsi_ulto2b(0x001E, feature->feature_code);
10345         feature->flags = 0x00;
10346         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10347                 feature->flags |= SGC_F_CURRENT;
10348         feature->add_length = 4;
10349         feature->feature_data[0] = 0x00;
10350         feature = (struct scsi_get_config_feature *)
10351             &feature->feature_data[feature->add_length];
10352
10353 f1f:    /* DVD Read */
10354         scsi_ulto2b(0x001F, feature->feature_code);
10355         feature->flags = 0x08;
10356         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10357                 feature->flags |= SGC_F_CURRENT;
10358         feature->add_length = 4;
10359         feature->feature_data[0] = 0x01;
10360         feature->feature_data[2] = 0x03;
10361         feature = (struct scsi_get_config_feature *)
10362             &feature->feature_data[feature->add_length];
10363
10364 f2a:    /* DVD+RW */
10365         scsi_ulto2b(0x002A, feature->feature_code);
10366         feature->flags = 0x04;
10367         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10368                 feature->flags |= SGC_F_CURRENT;
10369         feature->add_length = 4;
10370         feature->feature_data[0] = 0x00;
10371         feature->feature_data[1] = 0x00;
10372         feature = (struct scsi_get_config_feature *)
10373             &feature->feature_data[feature->add_length];
10374
10375 f2b:    /* DVD+R */
10376         scsi_ulto2b(0x002B, feature->feature_code);
10377         feature->flags = 0x00;
10378         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10379                 feature->flags |= SGC_F_CURRENT;
10380         feature->add_length = 4;
10381         feature->feature_data[0] = 0x00;
10382         feature = (struct scsi_get_config_feature *)
10383             &feature->feature_data[feature->add_length];
10384
10385 f3a:    /* DVD+RW Dual Layer */
10386         scsi_ulto2b(0x003A, feature->feature_code);
10387         feature->flags = 0x00;
10388         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10389                 feature->flags |= SGC_F_CURRENT;
10390         feature->add_length = 4;
10391         feature->feature_data[0] = 0x00;
10392         feature->feature_data[1] = 0x00;
10393         feature = (struct scsi_get_config_feature *)
10394             &feature->feature_data[feature->add_length];
10395
10396 f3b:    /* DVD+R Dual Layer */
10397         scsi_ulto2b(0x003B, feature->feature_code);
10398         feature->flags = 0x00;
10399         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10400                 feature->flags |= SGC_F_CURRENT;
10401         feature->add_length = 4;
10402         feature->feature_data[0] = 0x00;
10403         feature = (struct scsi_get_config_feature *)
10404             &feature->feature_data[feature->add_length];
10405
10406 done:
10407         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10408         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10409                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10410                 if (scsi_2btoul(feature->feature_code) == starting)
10411                         feature = (struct scsi_get_config_feature *)
10412                             &feature->feature_data[feature->add_length];
10413                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10414         }
10415         scsi_ulto4b(data_len - 4, hdr->data_length);
10416         ctsio->kern_data_len = min(data_len, alloc_len);
10417         ctsio->kern_total_len = ctsio->kern_data_len;
10418
10419         ctl_set_success(ctsio);
10420         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10421         ctsio->be_move_done = ctl_config_move_done;
10422         ctl_datamove((union ctl_io *)ctsio);
10423         return (CTL_RETVAL_COMPLETE);
10424 }
10425
10426 int
10427 ctl_get_event_status(struct ctl_scsiio *ctsio)
10428 {
10429         struct scsi_get_event_status_header *hdr;
10430         struct scsi_get_event_status *cdb;
10431         uint32_t alloc_len, data_len;
10432         int notif_class;
10433
10434         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10435         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10436                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10437                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10438                 ctl_done((union ctl_io *)ctsio);
10439                 return (CTL_RETVAL_COMPLETE);
10440         }
10441         notif_class = cdb->notif_class;
10442         alloc_len = scsi_2btoul(cdb->length);
10443
10444         data_len = sizeof(struct scsi_get_event_status_header);
10445         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10446         ctsio->kern_sg_entries = 0;
10447         ctsio->kern_rel_offset = 0;
10448         ctsio->kern_data_len = min(data_len, alloc_len);
10449         ctsio->kern_total_len = ctsio->kern_data_len;
10450
10451         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10452         scsi_ulto2b(0, hdr->descr_length);
10453         hdr->nea_class = SGESN_NEA;
10454         hdr->supported_class = 0;
10455
10456         ctl_set_success(ctsio);
10457         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10458         ctsio->be_move_done = ctl_config_move_done;
10459         ctl_datamove((union ctl_io *)ctsio);
10460         return (CTL_RETVAL_COMPLETE);
10461 }
10462
10463 int
10464 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10465 {
10466         struct scsi_mechanism_status_header *hdr;
10467         struct scsi_mechanism_status *cdb;
10468         uint32_t alloc_len, data_len;
10469
10470         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10471         alloc_len = scsi_2btoul(cdb->length);
10472
10473         data_len = sizeof(struct scsi_mechanism_status_header);
10474         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10475         ctsio->kern_sg_entries = 0;
10476         ctsio->kern_rel_offset = 0;
10477         ctsio->kern_data_len = min(data_len, alloc_len);
10478         ctsio->kern_total_len = ctsio->kern_data_len;
10479
10480         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10481         hdr->state1 = 0x00;
10482         hdr->state2 = 0xe0;
10483         scsi_ulto3b(0, hdr->lba);
10484         hdr->slots_num = 0;
10485         scsi_ulto2b(0, hdr->slots_length);
10486
10487         ctl_set_success(ctsio);
10488         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10489         ctsio->be_move_done = ctl_config_move_done;
10490         ctl_datamove((union ctl_io *)ctsio);
10491         return (CTL_RETVAL_COMPLETE);
10492 }
10493
10494 static void
10495 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10496 {
10497
10498         lba += 150;
10499         buf[0] = 0;
10500         buf[1] = bin2bcd((lba / 75) / 60);
10501         buf[2] = bin2bcd((lba / 75) % 60);
10502         buf[3] = bin2bcd(lba % 75);
10503 }
10504
10505 int
10506 ctl_read_toc(struct ctl_scsiio *ctsio)
10507 {
10508         struct ctl_lun *lun = CTL_LUN(ctsio);
10509         struct scsi_read_toc_hdr *hdr;
10510         struct scsi_read_toc_type01_descr *descr;
10511         struct scsi_read_toc *cdb;
10512         uint32_t alloc_len, data_len;
10513         int format, msf;
10514
10515         cdb = (struct scsi_read_toc *)ctsio->cdb;
10516         msf = (cdb->byte2 & CD_MSF) != 0;
10517         format = cdb->format;
10518         alloc_len = scsi_2btoul(cdb->data_len);
10519
10520         data_len = sizeof(struct scsi_read_toc_hdr);
10521         if (format == 0)
10522                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10523         else
10524                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10525         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10526         ctsio->kern_sg_entries = 0;
10527         ctsio->kern_rel_offset = 0;
10528         ctsio->kern_data_len = min(data_len, alloc_len);
10529         ctsio->kern_total_len = ctsio->kern_data_len;
10530
10531         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10532         if (format == 0) {
10533                 scsi_ulto2b(0x12, hdr->data_length);
10534                 hdr->first = 1;
10535                 hdr->last = 1;
10536                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10537                 descr->addr_ctl = 0x14;
10538                 descr->track_number = 1;
10539                 if (msf)
10540                         ctl_ultomsf(0, descr->track_start);
10541                 else
10542                         scsi_ulto4b(0, descr->track_start);
10543                 descr++;
10544                 descr->addr_ctl = 0x14;
10545                 descr->track_number = 0xaa;
10546                 if (msf)
10547                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10548                 else
10549                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10550         } else {
10551                 scsi_ulto2b(0x0a, hdr->data_length);
10552                 hdr->first = 1;
10553                 hdr->last = 1;
10554                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10555                 descr->addr_ctl = 0x14;
10556                 descr->track_number = 1;
10557                 if (msf)
10558                         ctl_ultomsf(0, descr->track_start);
10559                 else
10560                         scsi_ulto4b(0, descr->track_start);
10561         }
10562
10563         ctl_set_success(ctsio);
10564         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10565         ctsio->be_move_done = ctl_config_move_done;
10566         ctl_datamove((union ctl_io *)ctsio);
10567         return (CTL_RETVAL_COMPLETE);
10568 }
10569
10570 /*
10571  * For known CDB types, parse the LBA and length.
10572  */
10573 static int
10574 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10575 {
10576         if (io->io_hdr.io_type != CTL_IO_SCSI)
10577                 return (1);
10578
10579         switch (io->scsiio.cdb[0]) {
10580         case COMPARE_AND_WRITE: {
10581                 struct scsi_compare_and_write *cdb;
10582
10583                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10584
10585                 *lba = scsi_8btou64(cdb->addr);
10586                 *len = cdb->length;
10587                 break;
10588         }
10589         case READ_6:
10590         case WRITE_6: {
10591                 struct scsi_rw_6 *cdb;
10592
10593                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10594
10595                 *lba = scsi_3btoul(cdb->addr);
10596                 /* only 5 bits are valid in the most significant address byte */
10597                 *lba &= 0x1fffff;
10598                 *len = cdb->length;
10599                 break;
10600         }
10601         case READ_10:
10602         case WRITE_10: {
10603                 struct scsi_rw_10 *cdb;
10604
10605                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10606
10607                 *lba = scsi_4btoul(cdb->addr);
10608                 *len = scsi_2btoul(cdb->length);
10609                 break;
10610         }
10611         case WRITE_VERIFY_10: {
10612                 struct scsi_write_verify_10 *cdb;
10613
10614                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10615
10616                 *lba = scsi_4btoul(cdb->addr);
10617                 *len = scsi_2btoul(cdb->length);
10618                 break;
10619         }
10620         case READ_12:
10621         case WRITE_12: {
10622                 struct scsi_rw_12 *cdb;
10623
10624                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10625
10626                 *lba = scsi_4btoul(cdb->addr);
10627                 *len = scsi_4btoul(cdb->length);
10628                 break;
10629         }
10630         case WRITE_VERIFY_12: {
10631                 struct scsi_write_verify_12 *cdb;
10632
10633                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10634
10635                 *lba = scsi_4btoul(cdb->addr);
10636                 *len = scsi_4btoul(cdb->length);
10637                 break;
10638         }
10639         case READ_16:
10640         case WRITE_16: {
10641                 struct scsi_rw_16 *cdb;
10642
10643                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10644
10645                 *lba = scsi_8btou64(cdb->addr);
10646                 *len = scsi_4btoul(cdb->length);
10647                 break;
10648         }
10649         case WRITE_ATOMIC_16: {
10650                 struct scsi_write_atomic_16 *cdb;
10651
10652                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10653
10654                 *lba = scsi_8btou64(cdb->addr);
10655                 *len = scsi_2btoul(cdb->length);
10656                 break;
10657         }
10658         case WRITE_VERIFY_16: {
10659                 struct scsi_write_verify_16 *cdb;
10660
10661                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10662
10663                 *lba = scsi_8btou64(cdb->addr);
10664                 *len = scsi_4btoul(cdb->length);
10665                 break;
10666         }
10667         case WRITE_SAME_10: {
10668                 struct scsi_write_same_10 *cdb;
10669
10670                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10671
10672                 *lba = scsi_4btoul(cdb->addr);
10673                 *len = scsi_2btoul(cdb->length);
10674                 break;
10675         }
10676         case WRITE_SAME_16: {
10677                 struct scsi_write_same_16 *cdb;
10678
10679                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10680
10681                 *lba = scsi_8btou64(cdb->addr);
10682                 *len = scsi_4btoul(cdb->length);
10683                 break;
10684         }
10685         case VERIFY_10: {
10686                 struct scsi_verify_10 *cdb;
10687
10688                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10689
10690                 *lba = scsi_4btoul(cdb->addr);
10691                 *len = scsi_2btoul(cdb->length);
10692                 break;
10693         }
10694         case VERIFY_12: {
10695                 struct scsi_verify_12 *cdb;
10696
10697                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10698
10699                 *lba = scsi_4btoul(cdb->addr);
10700                 *len = scsi_4btoul(cdb->length);
10701                 break;
10702         }
10703         case VERIFY_16: {
10704                 struct scsi_verify_16 *cdb;
10705
10706                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10707
10708                 *lba = scsi_8btou64(cdb->addr);
10709                 *len = scsi_4btoul(cdb->length);
10710                 break;
10711         }
10712         case UNMAP: {
10713                 *lba = 0;
10714                 *len = UINT64_MAX;
10715                 break;
10716         }
10717         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10718                 struct scsi_get_lba_status *cdb;
10719
10720                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10721                 *lba = scsi_8btou64(cdb->addr);
10722                 *len = UINT32_MAX;
10723                 break;
10724         }
10725         default:
10726                 return (1);
10727                 break; /* NOTREACHED */
10728         }
10729
10730         return (0);
10731 }
10732
10733 static ctl_action
10734 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10735     bool seq)
10736 {
10737         uint64_t endlba1, endlba2;
10738
10739         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10740         endlba2 = lba2 + len2 - 1;
10741
10742         if ((endlba1 < lba2) || (endlba2 < lba1))
10743                 return (CTL_ACTION_PASS);
10744         else
10745                 return (CTL_ACTION_BLOCK);
10746 }
10747
10748 static int
10749 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10750 {
10751         struct ctl_ptr_len_flags *ptrlen;
10752         struct scsi_unmap_desc *buf, *end, *range;
10753         uint64_t lba;
10754         uint32_t len;
10755
10756         /* If not UNMAP -- go other way. */
10757         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10758             io->scsiio.cdb[0] != UNMAP)
10759                 return (CTL_ACTION_ERROR);
10760
10761         /* If UNMAP without data -- block and wait for data. */
10762         ptrlen = (struct ctl_ptr_len_flags *)
10763             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10764         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10765             ptrlen->ptr == NULL)
10766                 return (CTL_ACTION_BLOCK);
10767
10768         /* UNMAP with data -- check for collision. */
10769         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10770         end = buf + ptrlen->len / sizeof(*buf);
10771         for (range = buf; range < end; range++) {
10772                 lba = scsi_8btou64(range->lba);
10773                 len = scsi_4btoul(range->length);
10774                 if ((lba < lba2 + len2) && (lba + len > lba2))
10775                         return (CTL_ACTION_BLOCK);
10776         }
10777         return (CTL_ACTION_PASS);
10778 }
10779
10780 static ctl_action
10781 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10782 {
10783         uint64_t lba1, lba2;
10784         uint64_t len1, len2;
10785         int retval;
10786
10787         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10788                 return (CTL_ACTION_ERROR);
10789
10790         retval = ctl_extent_check_unmap(io1, lba2, len2);
10791         if (retval != CTL_ACTION_ERROR)
10792                 return (retval);
10793
10794         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10795                 return (CTL_ACTION_ERROR);
10796
10797         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10798                 seq = FALSE;
10799         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10800 }
10801
10802 static ctl_action
10803 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10804 {
10805         uint64_t lba1, lba2;
10806         uint64_t len1, len2;
10807
10808         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10809                 return (CTL_ACTION_PASS);
10810         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10811                 return (CTL_ACTION_ERROR);
10812         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10813                 return (CTL_ACTION_ERROR);
10814
10815         if (lba1 + len1 == lba2)
10816                 return (CTL_ACTION_BLOCK);
10817         return (CTL_ACTION_PASS);
10818 }
10819
10820 static ctl_action
10821 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10822     union ctl_io *ooa_io)
10823 {
10824         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10825         const ctl_serialize_action *serialize_row;
10826
10827         /*
10828          * The initiator attempted multiple untagged commands at the same
10829          * time.  Can't do that.
10830          */
10831         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10832          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10833          && ((pending_io->io_hdr.nexus.targ_port ==
10834               ooa_io->io_hdr.nexus.targ_port)
10835           && (pending_io->io_hdr.nexus.initid ==
10836               ooa_io->io_hdr.nexus.initid))
10837          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10838               CTL_FLAG_STATUS_SENT)) == 0))
10839                 return (CTL_ACTION_OVERLAP);
10840
10841         /*
10842          * The initiator attempted to send multiple tagged commands with
10843          * the same ID.  (It's fine if different initiators have the same
10844          * tag ID.)
10845          *
10846          * Even if all of those conditions are true, we don't kill the I/O
10847          * if the command ahead of us has been aborted.  We won't end up
10848          * sending it to the FETD, and it's perfectly legal to resend a
10849          * command with the same tag number as long as the previous
10850          * instance of this tag number has been aborted somehow.
10851          */
10852         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10853          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10854          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10855          && ((pending_io->io_hdr.nexus.targ_port ==
10856               ooa_io->io_hdr.nexus.targ_port)
10857           && (pending_io->io_hdr.nexus.initid ==
10858               ooa_io->io_hdr.nexus.initid))
10859          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10860               CTL_FLAG_STATUS_SENT)) == 0))
10861                 return (CTL_ACTION_OVERLAP_TAG);
10862
10863         /*
10864          * If we get a head of queue tag, SAM-3 says that we should
10865          * immediately execute it.
10866          *
10867          * What happens if this command would normally block for some other
10868          * reason?  e.g. a request sense with a head of queue tag
10869          * immediately after a write.  Normally that would block, but this
10870          * will result in its getting executed immediately...
10871          *
10872          * We currently return "pass" instead of "skip", so we'll end up
10873          * going through the rest of the queue to check for overlapped tags.
10874          *
10875          * XXX KDM check for other types of blockage first??
10876          */
10877         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10878                 return (CTL_ACTION_PASS);
10879
10880         /*
10881          * Ordered tags have to block until all items ahead of them
10882          * have completed.  If we get called with an ordered tag, we always
10883          * block, if something else is ahead of us in the queue.
10884          */
10885         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10886                 return (CTL_ACTION_BLOCK);
10887
10888         /*
10889          * Simple tags get blocked until all head of queue and ordered tags
10890          * ahead of them have completed.  I'm lumping untagged commands in
10891          * with simple tags here.  XXX KDM is that the right thing to do?
10892          */
10893         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10894           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10895          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10896           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10897                 return (CTL_ACTION_BLOCK);
10898
10899         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10900         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
10901             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
10902              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
10903              pending_io->scsiio.cdb[1], pending_io));
10904         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10905         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
10906                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
10907         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
10908             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
10909              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
10910              ooa_io->scsiio.cdb[1], ooa_io));
10911
10912         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10913
10914         switch (serialize_row[pending_entry->seridx]) {
10915         case CTL_SER_BLOCK:
10916                 return (CTL_ACTION_BLOCK);
10917         case CTL_SER_EXTENT:
10918                 return (ctl_extent_check(ooa_io, pending_io,
10919                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10920         case CTL_SER_EXTENTOPT:
10921                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10922                     SCP_QUEUE_ALG_UNRESTRICTED)
10923                         return (ctl_extent_check(ooa_io, pending_io,
10924                             (lun->be_lun &&
10925                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10926                 return (CTL_ACTION_PASS);
10927         case CTL_SER_EXTENTSEQ:
10928                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10929                         return (ctl_extent_check_seq(ooa_io, pending_io));
10930                 return (CTL_ACTION_PASS);
10931         case CTL_SER_PASS:
10932                 return (CTL_ACTION_PASS);
10933         case CTL_SER_BLOCKOPT:
10934                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10935                     SCP_QUEUE_ALG_UNRESTRICTED)
10936                         return (CTL_ACTION_BLOCK);
10937                 return (CTL_ACTION_PASS);
10938         case CTL_SER_SKIP:
10939                 return (CTL_ACTION_SKIP);
10940         default:
10941                 panic("%s: Invalid serialization value %d for %d => %d",
10942                     __func__, serialize_row[pending_entry->seridx],
10943                     pending_entry->seridx, ooa_entry->seridx);
10944         }
10945
10946         return (CTL_ACTION_ERROR);
10947 }
10948
10949 /*
10950  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10951  * Assumptions:
10952  * - pending_io is generally either incoming, or on the blocked queue
10953  * - starting I/O is the I/O we want to start the check with.
10954  */
10955 static ctl_action
10956 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10957               union ctl_io *starting_io)
10958 {
10959         union ctl_io *ooa_io;
10960         ctl_action action;
10961
10962         mtx_assert(&lun->lun_lock, MA_OWNED);
10963
10964         /*
10965          * Run back along the OOA queue, starting with the current
10966          * blocked I/O and going through every I/O before it on the
10967          * queue.  If starting_io is NULL, we'll just end up returning
10968          * CTL_ACTION_PASS.
10969          */
10970         for (ooa_io = starting_io; ooa_io != NULL;
10971              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10972              ooa_links)){
10973
10974                 /*
10975                  * This routine just checks to see whether
10976                  * cur_blocked is blocked by ooa_io, which is ahead
10977                  * of it in the queue.  It doesn't queue/dequeue
10978                  * cur_blocked.
10979                  */
10980                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10981                 switch (action) {
10982                 case CTL_ACTION_BLOCK:
10983                 case CTL_ACTION_OVERLAP:
10984                 case CTL_ACTION_OVERLAP_TAG:
10985                 case CTL_ACTION_SKIP:
10986                 case CTL_ACTION_ERROR:
10987                         return (action);
10988                         break; /* NOTREACHED */
10989                 case CTL_ACTION_PASS:
10990                         break;
10991                 default:
10992                         panic("%s: Invalid action %d\n", __func__, action);
10993                 }
10994         }
10995
10996         return (CTL_ACTION_PASS);
10997 }
10998
10999 /*
11000  * Assumptions:
11001  * - An I/O has just completed, and has been removed from the per-LUN OOA
11002  *   queue, so some items on the blocked queue may now be unblocked.
11003  */
11004 static int
11005 ctl_check_blocked(struct ctl_lun *lun)
11006 {
11007         struct ctl_softc *softc = lun->ctl_softc;
11008         union ctl_io *cur_blocked, *next_blocked;
11009
11010         mtx_assert(&lun->lun_lock, MA_OWNED);
11011
11012         /*
11013          * Run forward from the head of the blocked queue, checking each
11014          * entry against the I/Os prior to it on the OOA queue to see if
11015          * there is still any blockage.
11016          *
11017          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11018          * with our removing a variable on it while it is traversing the
11019          * list.
11020          */
11021         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11022              cur_blocked != NULL; cur_blocked = next_blocked) {
11023                 union ctl_io *prev_ooa;
11024                 ctl_action action;
11025
11026                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11027                                                           blocked_links);
11028
11029                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11030                                                       ctl_ooaq, ooa_links);
11031
11032                 /*
11033                  * If cur_blocked happens to be the first item in the OOA
11034                  * queue now, prev_ooa will be NULL, and the action
11035                  * returned will just be CTL_ACTION_PASS.
11036                  */
11037                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11038
11039                 switch (action) {
11040                 case CTL_ACTION_BLOCK:
11041                         /* Nothing to do here, still blocked */
11042                         break;
11043                 case CTL_ACTION_OVERLAP:
11044                 case CTL_ACTION_OVERLAP_TAG:
11045                         /*
11046                          * This shouldn't happen!  In theory we've already
11047                          * checked this command for overlap...
11048                          */
11049                         break;
11050                 case CTL_ACTION_PASS:
11051                 case CTL_ACTION_SKIP: {
11052                         const struct ctl_cmd_entry *entry;
11053
11054                         /*
11055                          * The skip case shouldn't happen, this transaction
11056                          * should have never made it onto the blocked queue.
11057                          */
11058                         /*
11059                          * This I/O is no longer blocked, we can remove it
11060                          * from the blocked queue.  Since this is a TAILQ
11061                          * (doubly linked list), we can do O(1) removals
11062                          * from any place on the list.
11063                          */
11064                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11065                                      blocked_links);
11066                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11067
11068                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11069                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11070                                 /*
11071                                  * Need to send IO back to original side to
11072                                  * run
11073                                  */
11074                                 union ctl_ha_msg msg_info;
11075
11076                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11077                                 msg_info.hdr.original_sc =
11078                                         cur_blocked->io_hdr.original_sc;
11079                                 msg_info.hdr.serializing_sc = cur_blocked;
11080                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11081                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11082                                     sizeof(msg_info.hdr), M_NOWAIT);
11083                                 break;
11084                         }
11085                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11086
11087                         /*
11088                          * Check this I/O for LUN state changes that may
11089                          * have happened while this command was blocked.
11090                          * The LUN state may have been changed by a command
11091                          * ahead of us in the queue, so we need to re-check
11092                          * for any states that can be caused by SCSI
11093                          * commands.
11094                          */
11095                         if (ctl_scsiio_lun_check(lun, entry,
11096                                                  &cur_blocked->scsiio) == 0) {
11097                                 cur_blocked->io_hdr.flags |=
11098                                                       CTL_FLAG_IS_WAS_ON_RTR;
11099                                 ctl_enqueue_rtr(cur_blocked);
11100                         } else
11101                                 ctl_done(cur_blocked);
11102                         break;
11103                 }
11104                 default:
11105                         /*
11106                          * This probably shouldn't happen -- we shouldn't
11107                          * get CTL_ACTION_ERROR, or anything else.
11108                          */
11109                         break;
11110                 }
11111         }
11112
11113         return (CTL_RETVAL_COMPLETE);
11114 }
11115
11116 /*
11117  * This routine (with one exception) checks LUN flags that can be set by
11118  * commands ahead of us in the OOA queue.  These flags have to be checked
11119  * when a command initially comes in, and when we pull a command off the
11120  * blocked queue and are preparing to execute it.  The reason we have to
11121  * check these flags for commands on the blocked queue is that the LUN
11122  * state may have been changed by a command ahead of us while we're on the
11123  * blocked queue.
11124  *
11125  * Ordering is somewhat important with these checks, so please pay
11126  * careful attention to the placement of any new checks.
11127  */
11128 static int
11129 ctl_scsiio_lun_check(struct ctl_lun *lun,
11130     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11131 {
11132         struct ctl_softc *softc = lun->ctl_softc;
11133         int retval;
11134         uint32_t residx;
11135
11136         retval = 0;
11137
11138         mtx_assert(&lun->lun_lock, MA_OWNED);
11139
11140         /*
11141          * If this shelf is a secondary shelf controller, we may have to
11142          * reject some commands disallowed by HA mode and link state.
11143          */
11144         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11145                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11146                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11147                         ctl_set_lun_unavail(ctsio);
11148                         retval = 1;
11149                         goto bailout;
11150                 }
11151                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11152                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11153                         ctl_set_lun_transit(ctsio);
11154                         retval = 1;
11155                         goto bailout;
11156                 }
11157                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11158                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11159                         ctl_set_lun_standby(ctsio);
11160                         retval = 1;
11161                         goto bailout;
11162                 }
11163
11164                 /* The rest of checks are only done on executing side */
11165                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11166                         goto bailout;
11167         }
11168
11169         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11170                 if (lun->be_lun &&
11171                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11172                         ctl_set_hw_write_protected(ctsio);
11173                         retval = 1;
11174                         goto bailout;
11175                 }
11176                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11177                         ctl_set_sense(ctsio, /*current_error*/ 1,
11178                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11179                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11180                         retval = 1;
11181                         goto bailout;
11182                 }
11183         }
11184
11185         /*
11186          * Check for a reservation conflict.  If this command isn't allowed
11187          * even on reserved LUNs, and if this initiator isn't the one who
11188          * reserved us, reject the command with a reservation conflict.
11189          */
11190         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11191         if ((lun->flags & CTL_LUN_RESERVED)
11192          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11193                 if (lun->res_idx != residx) {
11194                         ctl_set_reservation_conflict(ctsio);
11195                         retval = 1;
11196                         goto bailout;
11197                 }
11198         }
11199
11200         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11201             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11202                 /* No reservation or command is allowed. */;
11203         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11204             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11205              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11206              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11207                 /* The command is allowed for Write Exclusive resv. */;
11208         } else {
11209                 /*
11210                  * if we aren't registered or it's a res holder type
11211                  * reservation and this isn't the res holder then set a
11212                  * conflict.
11213                  */
11214                 if (ctl_get_prkey(lun, residx) == 0 ||
11215                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11216                         ctl_set_reservation_conflict(ctsio);
11217                         retval = 1;
11218                         goto bailout;
11219                 }
11220         }
11221
11222         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11223                 if (lun->flags & CTL_LUN_EJECTED)
11224                         ctl_set_lun_ejected(ctsio);
11225                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11226                         if (lun->flags & CTL_LUN_REMOVABLE)
11227                                 ctl_set_lun_no_media(ctsio);
11228                         else
11229                                 ctl_set_lun_int_reqd(ctsio);
11230                 } else if (lun->flags & CTL_LUN_STOPPED)
11231                         ctl_set_lun_stopped(ctsio);
11232                 else
11233                         goto bailout;
11234                 retval = 1;
11235                 goto bailout;
11236         }
11237
11238 bailout:
11239         return (retval);
11240 }
11241
11242 static void
11243 ctl_failover_io(union ctl_io *io, int have_lock)
11244 {
11245         ctl_set_busy(&io->scsiio);
11246         ctl_done(io);
11247 }
11248
11249 static void
11250 ctl_failover_lun(union ctl_io *rio)
11251 {
11252         struct ctl_softc *softc = CTL_SOFTC(rio);
11253         struct ctl_lun *lun;
11254         struct ctl_io_hdr *io, *next_io;
11255         uint32_t targ_lun;
11256
11257         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11258         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11259
11260         /* Find and lock the LUN. */
11261         mtx_lock(&softc->ctl_lock);
11262         if (targ_lun > ctl_max_luns ||
11263             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11264                 mtx_unlock(&softc->ctl_lock);
11265                 return;
11266         }
11267         mtx_lock(&lun->lun_lock);
11268         mtx_unlock(&softc->ctl_lock);
11269         if (lun->flags & CTL_LUN_DISABLED) {
11270                 mtx_unlock(&lun->lun_lock);
11271                 return;
11272         }
11273
11274         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11275                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11276                         /* We are master */
11277                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11278                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11279                                         io->flags |= CTL_FLAG_ABORT;
11280                                         io->flags |= CTL_FLAG_FAILOVER;
11281                                 } else { /* This can be only due to DATAMOVE */
11282                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11283                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11284                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11285                                         io->port_status = 31340;
11286                                         ctl_enqueue_isc((union ctl_io *)io);
11287                                 }
11288                         }
11289                         /* We are slave */
11290                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11291                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11292                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11293                                         io->flags |= CTL_FLAG_FAILOVER;
11294                                 } else {
11295                                         ctl_set_busy(&((union ctl_io *)io)->
11296                                             scsiio);
11297                                         ctl_done((union ctl_io *)io);
11298                                 }
11299                         }
11300                 }
11301         } else { /* SERIALIZE modes */
11302                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11303                     next_io) {
11304                         /* We are master */
11305                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11306                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11307                                     blocked_links);
11308                                 io->flags &= ~CTL_FLAG_BLOCKED;
11309                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11310                                 ctl_free_io((union ctl_io *)io);
11311                         }
11312                 }
11313                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11314                         /* We are master */
11315                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11316                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11317                                 ctl_free_io((union ctl_io *)io);
11318                         }
11319                         /* We are slave */
11320                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11321                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11322                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11323                                         ctl_set_busy(&((union ctl_io *)io)->
11324                                             scsiio);
11325                                         ctl_done((union ctl_io *)io);
11326                                 }
11327                         }
11328                 }
11329                 ctl_check_blocked(lun);
11330         }
11331         mtx_unlock(&lun->lun_lock);
11332 }
11333
11334 static int
11335 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11336 {
11337         struct ctl_lun *lun;
11338         const struct ctl_cmd_entry *entry;
11339         uint32_t initidx, targ_lun;
11340         int retval = 0;
11341
11342         lun = NULL;
11343         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11344         if (targ_lun < ctl_max_luns)
11345                 lun = softc->ctl_luns[targ_lun];
11346         if (lun) {
11347                 /*
11348                  * If the LUN is invalid, pretend that it doesn't exist.
11349                  * It will go away as soon as all pending I/O has been
11350                  * completed.
11351                  */
11352                 mtx_lock(&lun->lun_lock);
11353                 if (lun->flags & CTL_LUN_DISABLED) {
11354                         mtx_unlock(&lun->lun_lock);
11355                         lun = NULL;
11356                 }
11357         }
11358         CTL_LUN(ctsio) = lun;
11359         if (lun) {
11360                 CTL_BACKEND_LUN(ctsio) = lun->be_lun;
11361
11362                 /*
11363                  * Every I/O goes into the OOA queue for a particular LUN,
11364                  * and stays there until completion.
11365                  */
11366 #ifdef CTL_TIME_IO
11367                 if (TAILQ_EMPTY(&lun->ooa_queue))
11368                         lun->idle_time += getsbinuptime() - lun->last_busy;
11369 #endif
11370                 TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11371         }
11372
11373         /* Get command entry and return error if it is unsuppotyed. */
11374         entry = ctl_validate_command(ctsio);
11375         if (entry == NULL) {
11376                 if (lun)
11377                         mtx_unlock(&lun->lun_lock);
11378                 return (retval);
11379         }
11380
11381         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11382         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11383
11384         /*
11385          * Check to see whether we can send this command to LUNs that don't
11386          * exist.  This should pretty much only be the case for inquiry
11387          * and request sense.  Further checks, below, really require having
11388          * a LUN, so we can't really check the command anymore.  Just put
11389          * it on the rtr queue.
11390          */
11391         if (lun == NULL) {
11392                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11393                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11394                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11395                         return (retval);
11396                 }
11397
11398                 ctl_set_unsupported_lun(ctsio);
11399                 ctl_done((union ctl_io *)ctsio);
11400                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11401                 return (retval);
11402         } else {
11403                 /*
11404                  * Make sure we support this particular command on this LUN.
11405                  * e.g., we don't support writes to the control LUN.
11406                  */
11407                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11408                         mtx_unlock(&lun->lun_lock);
11409                         ctl_set_invalid_opcode(ctsio);
11410                         ctl_done((union ctl_io *)ctsio);
11411                         return (retval);
11412                 }
11413         }
11414
11415         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11416
11417         /*
11418          * If we've got a request sense, it'll clear the contingent
11419          * allegiance condition.  Otherwise, if we have a CA condition for
11420          * this initiator, clear it, because it sent down a command other
11421          * than request sense.
11422          */
11423         if (ctsio->cdb[0] != REQUEST_SENSE) {
11424                 struct scsi_sense_data *ps;
11425
11426                 ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
11427                 if (ps != NULL)
11428                         ps[initidx % CTL_MAX_INIT_PER_PORT].error_code = 0;
11429         }
11430
11431         /*
11432          * If the command has this flag set, it handles its own unit
11433          * attention reporting, we shouldn't do anything.  Otherwise we
11434          * check for any pending unit attentions, and send them back to the
11435          * initiator.  We only do this when a command initially comes in,
11436          * not when we pull it off the blocked queue.
11437          *
11438          * According to SAM-3, section 5.3.2, the order that things get
11439          * presented back to the host is basically unit attentions caused
11440          * by some sort of reset event, busy status, reservation conflicts
11441          * or task set full, and finally any other status.
11442          *
11443          * One issue here is that some of the unit attentions we report
11444          * don't fall into the "reset" category (e.g. "reported luns data
11445          * has changed").  So reporting it here, before the reservation
11446          * check, may be technically wrong.  I guess the only thing to do
11447          * would be to check for and report the reset events here, and then
11448          * check for the other unit attention types after we check for a
11449          * reservation conflict.
11450          *
11451          * XXX KDM need to fix this
11452          */
11453         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11454                 ctl_ua_type ua_type;
11455                 u_int sense_len = 0;
11456
11457                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11458                     &sense_len, SSD_TYPE_NONE);
11459                 if (ua_type != CTL_UA_NONE) {
11460                         mtx_unlock(&lun->lun_lock);
11461                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11462                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11463                         ctsio->sense_len = sense_len;
11464                         ctl_done((union ctl_io *)ctsio);
11465                         return (retval);
11466                 }
11467         }
11468
11469
11470         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11471                 mtx_unlock(&lun->lun_lock);
11472                 ctl_done((union ctl_io *)ctsio);
11473                 return (retval);
11474         }
11475
11476         /*
11477          * XXX CHD this is where we want to send IO to other side if
11478          * this LUN is secondary on this SC. We will need to make a copy
11479          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11480          * the copy we send as FROM_OTHER.
11481          * We also need to stuff the address of the original IO so we can
11482          * find it easily. Something similar will need be done on the other
11483          * side so when we are done we can find the copy.
11484          */
11485         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11486             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11487             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11488                 union ctl_ha_msg msg_info;
11489                 int isc_retval;
11490
11491                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11492                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11493                 mtx_unlock(&lun->lun_lock);
11494
11495                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11496                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11497                 msg_info.hdr.serializing_sc = NULL;
11498                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11499                 msg_info.scsi.tag_num = ctsio->tag_num;
11500                 msg_info.scsi.tag_type = ctsio->tag_type;
11501                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11502                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11503
11504                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11505                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11506                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11507                         ctl_set_busy(ctsio);
11508                         ctl_done((union ctl_io *)ctsio);
11509                         return (retval);
11510                 }
11511                 return (retval);
11512         }
11513
11514         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11515                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11516                               ctl_ooaq, ooa_links))) {
11517         case CTL_ACTION_BLOCK:
11518                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11519                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11520                                   blocked_links);
11521                 mtx_unlock(&lun->lun_lock);
11522                 return (retval);
11523         case CTL_ACTION_PASS:
11524         case CTL_ACTION_SKIP:
11525                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11526                 mtx_unlock(&lun->lun_lock);
11527                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11528                 break;
11529         case CTL_ACTION_OVERLAP:
11530                 mtx_unlock(&lun->lun_lock);
11531                 ctl_set_overlapped_cmd(ctsio);
11532                 ctl_done((union ctl_io *)ctsio);
11533                 break;
11534         case CTL_ACTION_OVERLAP_TAG:
11535                 mtx_unlock(&lun->lun_lock);
11536                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11537                 ctl_done((union ctl_io *)ctsio);
11538                 break;
11539         case CTL_ACTION_ERROR:
11540         default:
11541                 mtx_unlock(&lun->lun_lock);
11542                 ctl_set_internal_failure(ctsio,
11543                                          /*sks_valid*/ 0,
11544                                          /*retry_count*/ 0);
11545                 ctl_done((union ctl_io *)ctsio);
11546                 break;
11547         }
11548         return (retval);
11549 }
11550
11551 const struct ctl_cmd_entry *
11552 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11553 {
11554         const struct ctl_cmd_entry *entry;
11555         int service_action;
11556
11557         entry = &ctl_cmd_table[ctsio->cdb[0]];
11558         if (sa)
11559                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11560         if (entry->flags & CTL_CMD_FLAG_SA5) {
11561                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11562                 entry = &((const struct ctl_cmd_entry *)
11563                     entry->execute)[service_action];
11564         }
11565         return (entry);
11566 }
11567
11568 const struct ctl_cmd_entry *
11569 ctl_validate_command(struct ctl_scsiio *ctsio)
11570 {
11571         const struct ctl_cmd_entry *entry;
11572         int i, sa;
11573         uint8_t diff;
11574
11575         entry = ctl_get_cmd_entry(ctsio, &sa);
11576         if (entry->execute == NULL) {
11577                 if (sa)
11578                         ctl_set_invalid_field(ctsio,
11579                                               /*sks_valid*/ 1,
11580                                               /*command*/ 1,
11581                                               /*field*/ 1,
11582                                               /*bit_valid*/ 1,
11583                                               /*bit*/ 4);
11584                 else
11585                         ctl_set_invalid_opcode(ctsio);
11586                 ctl_done((union ctl_io *)ctsio);
11587                 return (NULL);
11588         }
11589         KASSERT(entry->length > 0,
11590             ("Not defined length for command 0x%02x/0x%02x",
11591              ctsio->cdb[0], ctsio->cdb[1]));
11592         for (i = 1; i < entry->length; i++) {
11593                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11594                 if (diff == 0)
11595                         continue;
11596                 ctl_set_invalid_field(ctsio,
11597                                       /*sks_valid*/ 1,
11598                                       /*command*/ 1,
11599                                       /*field*/ i,
11600                                       /*bit_valid*/ 1,
11601                                       /*bit*/ fls(diff) - 1);
11602                 ctl_done((union ctl_io *)ctsio);
11603                 return (NULL);
11604         }
11605         return (entry);
11606 }
11607
11608 static int
11609 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11610 {
11611
11612         switch (lun_type) {
11613         case T_DIRECT:
11614                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11615                         return (0);
11616                 break;
11617         case T_PROCESSOR:
11618                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11619                         return (0);
11620                 break;
11621         case T_CDROM:
11622                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11623                         return (0);
11624                 break;
11625         default:
11626                 return (0);
11627         }
11628         return (1);
11629 }
11630
11631 static int
11632 ctl_scsiio(struct ctl_scsiio *ctsio)
11633 {
11634         int retval;
11635         const struct ctl_cmd_entry *entry;
11636
11637         retval = CTL_RETVAL_COMPLETE;
11638
11639         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11640
11641         entry = ctl_get_cmd_entry(ctsio, NULL);
11642
11643         /*
11644          * If this I/O has been aborted, just send it straight to
11645          * ctl_done() without executing it.
11646          */
11647         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11648                 ctl_done((union ctl_io *)ctsio);
11649                 goto bailout;
11650         }
11651
11652         /*
11653          * All the checks should have been handled by ctl_scsiio_precheck().
11654          * We should be clear now to just execute the I/O.
11655          */
11656         retval = entry->execute(ctsio);
11657
11658 bailout:
11659         return (retval);
11660 }
11661
11662 static int
11663 ctl_target_reset(union ctl_io *io)
11664 {
11665         struct ctl_softc *softc = CTL_SOFTC(io);
11666         struct ctl_port *port = CTL_PORT(io);
11667         struct ctl_lun *lun;
11668         uint32_t initidx;
11669         ctl_ua_type ua_type;
11670
11671         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11672                 union ctl_ha_msg msg_info;
11673
11674                 msg_info.hdr.nexus = io->io_hdr.nexus;
11675                 msg_info.task.task_action = io->taskio.task_action;
11676                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11677                 msg_info.hdr.original_sc = NULL;
11678                 msg_info.hdr.serializing_sc = NULL;
11679                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11680                     sizeof(msg_info.task), M_WAITOK);
11681         }
11682
11683         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11684         if (io->taskio.task_action == CTL_TASK_TARGET_RESET)
11685                 ua_type = CTL_UA_TARG_RESET;
11686         else
11687                 ua_type = CTL_UA_BUS_RESET;
11688         mtx_lock(&softc->ctl_lock);
11689         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11690                 if (port != NULL &&
11691                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
11692                         continue;
11693                 ctl_do_lun_reset(lun, initidx, ua_type);
11694         }
11695         mtx_unlock(&softc->ctl_lock);
11696         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11697         return (0);
11698 }
11699
11700 /*
11701  * The LUN should always be set.  The I/O is optional, and is used to
11702  * distinguish between I/Os sent by this initiator, and by other
11703  * initiators.  We set unit attention for initiators other than this one.
11704  * SAM-3 is vague on this point.  It does say that a unit attention should
11705  * be established for other initiators when a LUN is reset (see section
11706  * 5.7.3), but it doesn't specifically say that the unit attention should
11707  * be established for this particular initiator when a LUN is reset.  Here
11708  * is the relevant text, from SAM-3 rev 8:
11709  *
11710  * 5.7.2 When a SCSI initiator port aborts its own tasks
11711  *
11712  * When a SCSI initiator port causes its own task(s) to be aborted, no
11713  * notification that the task(s) have been aborted shall be returned to
11714  * the SCSI initiator port other than the completion response for the
11715  * command or task management function action that caused the task(s) to
11716  * be aborted and notification(s) associated with related effects of the
11717  * action (e.g., a reset unit attention condition).
11718  *
11719  * XXX KDM for now, we're setting unit attention for all initiators.
11720  */
11721 static void
11722 ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua_type)
11723 {
11724         union ctl_io *xio;
11725         int i;
11726
11727         mtx_lock(&lun->lun_lock);
11728         /* Abort tasks. */
11729         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11730              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11731                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11732         }
11733         /* Clear CA. */
11734         for (i = 0; i < ctl_max_ports; i++) {
11735                 free(lun->pending_sense[i], M_CTL);
11736                 lun->pending_sense[i] = NULL;
11737         }
11738         /* Clear reservation. */
11739         lun->flags &= ~CTL_LUN_RESERVED;
11740         /* Clear prevent media removal. */
11741         if (lun->prevent) {
11742                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
11743                         ctl_clear_mask(lun->prevent, i);
11744                 lun->prevent_count = 0;
11745         }
11746         /* Clear TPC status */
11747         ctl_tpc_lun_clear(lun, -1);
11748         /* Establish UA. */
11749 #if 0
11750         ctl_est_ua_all(lun, initidx, ua_type);
11751 #else
11752         ctl_est_ua_all(lun, -1, ua_type);
11753 #endif
11754         mtx_unlock(&lun->lun_lock);
11755 }
11756
11757 static int
11758 ctl_lun_reset(union ctl_io *io)
11759 {
11760         struct ctl_softc *softc = CTL_SOFTC(io);
11761         struct ctl_lun *lun;
11762         uint32_t targ_lun, initidx;
11763
11764         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11765         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11766         mtx_lock(&softc->ctl_lock);
11767         if (targ_lun >= ctl_max_luns ||
11768             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11769                 mtx_unlock(&softc->ctl_lock);
11770                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11771                 return (1);
11772         }
11773         ctl_do_lun_reset(lun, initidx, CTL_UA_LUN_RESET);
11774         mtx_unlock(&softc->ctl_lock);
11775         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11776
11777         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11778                 union ctl_ha_msg msg_info;
11779
11780                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11781                 msg_info.hdr.nexus = io->io_hdr.nexus;
11782                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11783                 msg_info.hdr.original_sc = NULL;
11784                 msg_info.hdr.serializing_sc = NULL;
11785                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11786                     sizeof(msg_info.task), M_WAITOK);
11787         }
11788         return (0);
11789 }
11790
11791 static void
11792 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11793     int other_sc)
11794 {
11795         union ctl_io *xio;
11796
11797         mtx_assert(&lun->lun_lock, MA_OWNED);
11798
11799         /*
11800          * Run through the OOA queue and attempt to find the given I/O.
11801          * The target port, initiator ID, tag type and tag number have to
11802          * match the values that we got from the initiator.  If we have an
11803          * untagged command to abort, simply abort the first untagged command
11804          * we come to.  We only allow one untagged command at a time of course.
11805          */
11806         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11807              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11808
11809                 if ((targ_port == UINT32_MAX ||
11810                      targ_port == xio->io_hdr.nexus.targ_port) &&
11811                     (init_id == UINT32_MAX ||
11812                      init_id == xio->io_hdr.nexus.initid)) {
11813                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11814                             init_id != xio->io_hdr.nexus.initid)
11815                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11816                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11817                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11818                                 union ctl_ha_msg msg_info;
11819
11820                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11821                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11822                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11823                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11824                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11825                                 msg_info.hdr.original_sc = NULL;
11826                                 msg_info.hdr.serializing_sc = NULL;
11827                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11828                                     sizeof(msg_info.task), M_NOWAIT);
11829                         }
11830                 }
11831         }
11832 }
11833
11834 static int
11835 ctl_abort_task_set(union ctl_io *io)
11836 {
11837         struct ctl_softc *softc = CTL_SOFTC(io);
11838         struct ctl_lun *lun;
11839         uint32_t targ_lun;
11840
11841         /*
11842          * Look up the LUN.
11843          */
11844         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11845         mtx_lock(&softc->ctl_lock);
11846         if (targ_lun >= ctl_max_luns ||
11847             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11848                 mtx_unlock(&softc->ctl_lock);
11849                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11850                 return (1);
11851         }
11852
11853         mtx_lock(&lun->lun_lock);
11854         mtx_unlock(&softc->ctl_lock);
11855         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11856                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11857                     io->io_hdr.nexus.initid,
11858                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11859         } else { /* CTL_TASK_CLEAR_TASK_SET */
11860                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11861                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11862         }
11863         mtx_unlock(&lun->lun_lock);
11864         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11865         return (0);
11866 }
11867
11868 static void
11869 ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
11870     ctl_ua_type ua_type)
11871 {
11872         struct ctl_lun *lun;
11873         struct scsi_sense_data *ps;
11874         uint32_t p, i;
11875
11876         p = initidx / CTL_MAX_INIT_PER_PORT;
11877         i = initidx % CTL_MAX_INIT_PER_PORT;
11878         mtx_lock(&softc->ctl_lock);
11879         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11880                 mtx_lock(&lun->lun_lock);
11881                 /* Abort tasks. */
11882                 ctl_abort_tasks_lun(lun, p, i, 1);
11883                 /* Clear CA. */
11884                 ps = lun->pending_sense[p];
11885                 if (ps != NULL)
11886                         ps[i].error_code = 0;
11887                 /* Clear reservation. */
11888                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11889                         lun->flags &= ~CTL_LUN_RESERVED;
11890                 /* Clear prevent media removal. */
11891                 if (lun->prevent && ctl_is_set(lun->prevent, initidx)) {
11892                         ctl_clear_mask(lun->prevent, initidx);
11893                         lun->prevent_count--;
11894                 }
11895                 /* Clear TPC status */
11896                 ctl_tpc_lun_clear(lun, initidx);
11897                 /* Establish UA. */
11898                 ctl_est_ua(lun, initidx, ua_type);
11899                 mtx_unlock(&lun->lun_lock);
11900         }
11901         mtx_unlock(&softc->ctl_lock);
11902 }
11903
11904 static int
11905 ctl_i_t_nexus_reset(union ctl_io *io)
11906 {
11907         struct ctl_softc *softc = CTL_SOFTC(io);
11908         uint32_t initidx;
11909
11910         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11911                 union ctl_ha_msg msg_info;
11912
11913                 msg_info.hdr.nexus = io->io_hdr.nexus;
11914                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
11915                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11916                 msg_info.hdr.original_sc = NULL;
11917                 msg_info.hdr.serializing_sc = NULL;
11918                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11919                     sizeof(msg_info.task), M_WAITOK);
11920         }
11921
11922         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11923         ctl_i_t_nexus_loss(softc, initidx, CTL_UA_I_T_NEXUS_LOSS);
11924         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11925         return (0);
11926 }
11927
11928 static int
11929 ctl_abort_task(union ctl_io *io)
11930 {
11931         struct ctl_softc *softc = CTL_SOFTC(io);
11932         union ctl_io *xio;
11933         struct ctl_lun *lun;
11934 #if 0
11935         struct sbuf sb;
11936         char printbuf[128];
11937 #endif
11938         int found;
11939         uint32_t targ_lun;
11940
11941         found = 0;
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 #if 0
11956         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11957                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11958 #endif
11959
11960         mtx_lock(&lun->lun_lock);
11961         mtx_unlock(&softc->ctl_lock);
11962         /*
11963          * Run through the OOA queue and attempt to find the given I/O.
11964          * The target port, initiator ID, tag type and tag number have to
11965          * match the values that we got from the initiator.  If we have an
11966          * untagged command to abort, simply abort the first untagged command
11967          * we come to.  We only allow one untagged command at a time of course.
11968          */
11969         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11970              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11971 #if 0
11972                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11973
11974                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11975                             lun->lun, xio->scsiio.tag_num,
11976                             xio->scsiio.tag_type,
11977                             (xio->io_hdr.blocked_links.tqe_prev
11978                             == NULL) ? "" : " BLOCKED",
11979                             (xio->io_hdr.flags &
11980                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11981                             (xio->io_hdr.flags &
11982                             CTL_FLAG_ABORT) ? " ABORT" : "",
11983                             (xio->io_hdr.flags &
11984                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11985                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11986                 sbuf_finish(&sb);
11987                 printf("%s\n", sbuf_data(&sb));
11988 #endif
11989
11990                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11991                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11992                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11993                         continue;
11994
11995                 /*
11996                  * If the abort says that the task is untagged, the
11997                  * task in the queue must be untagged.  Otherwise,
11998                  * we just check to see whether the tag numbers
11999                  * match.  This is because the QLogic firmware
12000                  * doesn't pass back the tag type in an abort
12001                  * request.
12002                  */
12003 #if 0
12004                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12005                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12006                  || (xio->scsiio.tag_num == io->taskio.tag_num))
12007 #endif
12008                 /*
12009                  * XXX KDM we've got problems with FC, because it
12010                  * doesn't send down a tag type with aborts.  So we
12011                  * can only really go by the tag number...
12012                  * This may cause problems with parallel SCSI.
12013                  * Need to figure that out!!
12014                  */
12015                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
12016                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12017                         found = 1;
12018                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12019                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12020                                 union ctl_ha_msg msg_info;
12021
12022                                 msg_info.hdr.nexus = io->io_hdr.nexus;
12023                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12024                                 msg_info.task.tag_num = io->taskio.tag_num;
12025                                 msg_info.task.tag_type = io->taskio.tag_type;
12026                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12027                                 msg_info.hdr.original_sc = NULL;
12028                                 msg_info.hdr.serializing_sc = NULL;
12029 #if 0
12030                                 printf("Sent Abort to other side\n");
12031 #endif
12032                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12033                                     sizeof(msg_info.task), M_NOWAIT);
12034                         }
12035 #if 0
12036                         printf("ctl_abort_task: found I/O to abort\n");
12037 #endif
12038                 }
12039         }
12040         mtx_unlock(&lun->lun_lock);
12041
12042         if (found == 0) {
12043                 /*
12044                  * This isn't really an error.  It's entirely possible for
12045                  * the abort and command completion to cross on the wire.
12046                  * This is more of an informative/diagnostic error.
12047                  */
12048 #if 0
12049                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12050                        "%u:%u:%u tag %d type %d\n",
12051                        io->io_hdr.nexus.initid,
12052                        io->io_hdr.nexus.targ_port,
12053                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12054                        io->taskio.tag_type);
12055 #endif
12056         }
12057         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12058         return (0);
12059 }
12060
12061 static int
12062 ctl_query_task(union ctl_io *io, int task_set)
12063 {
12064         struct ctl_softc *softc = CTL_SOFTC(io);
12065         union ctl_io *xio;
12066         struct ctl_lun *lun;
12067         int found = 0;
12068         uint32_t targ_lun;
12069
12070         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12071         mtx_lock(&softc->ctl_lock);
12072         if (targ_lun >= ctl_max_luns ||
12073             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12074                 mtx_unlock(&softc->ctl_lock);
12075                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12076                 return (1);
12077         }
12078         mtx_lock(&lun->lun_lock);
12079         mtx_unlock(&softc->ctl_lock);
12080         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12081              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12082
12083                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12084                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12085                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12086                         continue;
12087
12088                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12089                         found = 1;
12090                         break;
12091                 }
12092         }
12093         mtx_unlock(&lun->lun_lock);
12094         if (found)
12095                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12096         else
12097                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12098         return (0);
12099 }
12100
12101 static int
12102 ctl_query_async_event(union ctl_io *io)
12103 {
12104         struct ctl_softc *softc = CTL_SOFTC(io);
12105         struct ctl_lun *lun;
12106         ctl_ua_type ua;
12107         uint32_t targ_lun, initidx;
12108
12109         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12110         mtx_lock(&softc->ctl_lock);
12111         if (targ_lun >= ctl_max_luns ||
12112             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12113                 mtx_unlock(&softc->ctl_lock);
12114                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12115                 return (1);
12116         }
12117         mtx_lock(&lun->lun_lock);
12118         mtx_unlock(&softc->ctl_lock);
12119         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12120         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12121         mtx_unlock(&lun->lun_lock);
12122         if (ua != CTL_UA_NONE)
12123                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12124         else
12125                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12126         return (0);
12127 }
12128
12129 static void
12130 ctl_run_task(union ctl_io *io)
12131 {
12132         int retval = 1;
12133
12134         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12135         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12136             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12137         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12138         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12139         switch (io->taskio.task_action) {
12140         case CTL_TASK_ABORT_TASK:
12141                 retval = ctl_abort_task(io);
12142                 break;
12143         case CTL_TASK_ABORT_TASK_SET:
12144         case CTL_TASK_CLEAR_TASK_SET:
12145                 retval = ctl_abort_task_set(io);
12146                 break;
12147         case CTL_TASK_CLEAR_ACA:
12148                 break;
12149         case CTL_TASK_I_T_NEXUS_RESET:
12150                 retval = ctl_i_t_nexus_reset(io);
12151                 break;
12152         case CTL_TASK_LUN_RESET:
12153                 retval = ctl_lun_reset(io);
12154                 break;
12155         case CTL_TASK_TARGET_RESET:
12156         case CTL_TASK_BUS_RESET:
12157                 retval = ctl_target_reset(io);
12158                 break;
12159         case CTL_TASK_PORT_LOGIN:
12160                 break;
12161         case CTL_TASK_PORT_LOGOUT:
12162                 break;
12163         case CTL_TASK_QUERY_TASK:
12164                 retval = ctl_query_task(io, 0);
12165                 break;
12166         case CTL_TASK_QUERY_TASK_SET:
12167                 retval = ctl_query_task(io, 1);
12168                 break;
12169         case CTL_TASK_QUERY_ASYNC_EVENT:
12170                 retval = ctl_query_async_event(io);
12171                 break;
12172         default:
12173                 printf("%s: got unknown task management event %d\n",
12174                        __func__, io->taskio.task_action);
12175                 break;
12176         }
12177         if (retval == 0)
12178                 io->io_hdr.status = CTL_SUCCESS;
12179         else
12180                 io->io_hdr.status = CTL_ERROR;
12181         ctl_done(io);
12182 }
12183
12184 /*
12185  * For HA operation.  Handle commands that come in from the other
12186  * controller.
12187  */
12188 static void
12189 ctl_handle_isc(union ctl_io *io)
12190 {
12191         struct ctl_softc *softc = CTL_SOFTC(io);
12192         struct ctl_lun *lun;
12193         const struct ctl_cmd_entry *entry;
12194         uint32_t targ_lun;
12195
12196         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12197         switch (io->io_hdr.msg_type) {
12198         case CTL_MSG_SERIALIZE:
12199                 ctl_serialize_other_sc_cmd(&io->scsiio);
12200                 break;
12201         case CTL_MSG_R2R:               /* Only used in SER_ONLY mode. */
12202                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12203                 if (targ_lun >= ctl_max_luns ||
12204                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12205                         ctl_done(io);
12206                         break;
12207                 }
12208                 mtx_lock(&lun->lun_lock);
12209                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
12210                         mtx_unlock(&lun->lun_lock);
12211                         ctl_done(io);
12212                         break;
12213                 }
12214                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12215                 mtx_unlock(&lun->lun_lock);
12216                 ctl_enqueue_rtr(io);
12217                 break;
12218         case CTL_MSG_FINISH_IO:
12219                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12220                         ctl_done(io);
12221                         break;
12222                 }
12223                 if (targ_lun >= ctl_max_luns ||
12224                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12225                         ctl_free_io(io);
12226                         break;
12227                 }
12228                 mtx_lock(&lun->lun_lock);
12229                 TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12230                 ctl_check_blocked(lun);
12231                 mtx_unlock(&lun->lun_lock);
12232                 ctl_free_io(io);
12233                 break;
12234         case CTL_MSG_PERS_ACTION:
12235                 ctl_hndl_per_res_out_on_other_sc(io);
12236                 ctl_free_io(io);
12237                 break;
12238         case CTL_MSG_BAD_JUJU:
12239                 ctl_done(io);
12240                 break;
12241         case CTL_MSG_DATAMOVE:          /* Only used in XFER mode */
12242                 ctl_datamove_remote(io);
12243                 break;
12244         case CTL_MSG_DATAMOVE_DONE:     /* Only used in XFER mode */
12245                 io->scsiio.be_move_done(io);
12246                 break;
12247         case CTL_MSG_FAILOVER:
12248                 ctl_failover_lun(io);
12249                 ctl_free_io(io);
12250                 break;
12251         default:
12252                 printf("%s: Invalid message type %d\n",
12253                        __func__, io->io_hdr.msg_type);
12254                 ctl_free_io(io);
12255                 break;
12256         }
12257
12258 }
12259
12260
12261 /*
12262  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12263  * there is no match.
12264  */
12265 static ctl_lun_error_pattern
12266 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12267 {
12268         const struct ctl_cmd_entry *entry;
12269         ctl_lun_error_pattern filtered_pattern, pattern;
12270
12271         pattern = desc->error_pattern;
12272
12273         /*
12274          * XXX KDM we need more data passed into this function to match a
12275          * custom pattern, and we actually need to implement custom pattern
12276          * matching.
12277          */
12278         if (pattern & CTL_LUN_PAT_CMD)
12279                 return (CTL_LUN_PAT_CMD);
12280
12281         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12282                 return (CTL_LUN_PAT_ANY);
12283
12284         entry = ctl_get_cmd_entry(ctsio, NULL);
12285
12286         filtered_pattern = entry->pattern & pattern;
12287
12288         /*
12289          * If the user requested specific flags in the pattern (e.g.
12290          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12291          * flags.
12292          *
12293          * If the user did not specify any flags, it doesn't matter whether
12294          * or not the command supports the flags.
12295          */
12296         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12297              (pattern & ~CTL_LUN_PAT_MASK))
12298                 return (CTL_LUN_PAT_NONE);
12299
12300         /*
12301          * If the user asked for a range check, see if the requested LBA
12302          * range overlaps with this command's LBA range.
12303          */
12304         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12305                 uint64_t lba1;
12306                 uint64_t len1;
12307                 ctl_action action;
12308                 int retval;
12309
12310                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12311                 if (retval != 0)
12312                         return (CTL_LUN_PAT_NONE);
12313
12314                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12315                                               desc->lba_range.len, FALSE);
12316                 /*
12317                  * A "pass" means that the LBA ranges don't overlap, so
12318                  * this doesn't match the user's range criteria.
12319                  */
12320                 if (action == CTL_ACTION_PASS)
12321                         return (CTL_LUN_PAT_NONE);
12322         }
12323
12324         return (filtered_pattern);
12325 }
12326
12327 static void
12328 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12329 {
12330         struct ctl_error_desc *desc, *desc2;
12331
12332         mtx_assert(&lun->lun_lock, MA_OWNED);
12333
12334         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12335                 ctl_lun_error_pattern pattern;
12336                 /*
12337                  * Check to see whether this particular command matches
12338                  * the pattern in the descriptor.
12339                  */
12340                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12341                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12342                         continue;
12343
12344                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12345                 case CTL_LUN_INJ_ABORTED:
12346                         ctl_set_aborted(&io->scsiio);
12347                         break;
12348                 case CTL_LUN_INJ_MEDIUM_ERR:
12349                         ctl_set_medium_error(&io->scsiio,
12350                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12351                              CTL_FLAG_DATA_OUT);
12352                         break;
12353                 case CTL_LUN_INJ_UA:
12354                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12355                          * OCCURRED */
12356                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12357                         break;
12358                 case CTL_LUN_INJ_CUSTOM:
12359                         /*
12360                          * We're assuming the user knows what he is doing.
12361                          * Just copy the sense information without doing
12362                          * checks.
12363                          */
12364                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12365                               MIN(sizeof(desc->custom_sense),
12366                                   sizeof(io->scsiio.sense_data)));
12367                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12368                         io->scsiio.sense_len = SSD_FULL_SIZE;
12369                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12370                         break;
12371                 case CTL_LUN_INJ_NONE:
12372                 default:
12373                         /*
12374                          * If this is an error injection type we don't know
12375                          * about, clear the continuous flag (if it is set)
12376                          * so it will get deleted below.
12377                          */
12378                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12379                         break;
12380                 }
12381                 /*
12382                  * By default, each error injection action is a one-shot
12383                  */
12384                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12385                         continue;
12386
12387                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12388
12389                 free(desc, M_CTL);
12390         }
12391 }
12392
12393 #ifdef CTL_IO_DELAY
12394 static void
12395 ctl_datamove_timer_wakeup(void *arg)
12396 {
12397         union ctl_io *io;
12398
12399         io = (union ctl_io *)arg;
12400
12401         ctl_datamove(io);
12402 }
12403 #endif /* CTL_IO_DELAY */
12404
12405 void
12406 ctl_datamove(union ctl_io *io)
12407 {
12408         void (*fe_datamove)(union ctl_io *io);
12409
12410         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12411
12412         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12413
12414         /* No data transferred yet.  Frontend must update this when done. */
12415         io->scsiio.kern_data_resid = io->scsiio.kern_data_len;
12416
12417 #ifdef CTL_TIME_IO
12418         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12419                 char str[256];
12420                 char path_str[64];
12421                 struct sbuf sb;
12422
12423                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12424                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12425
12426                 sbuf_cat(&sb, path_str);
12427                 switch (io->io_hdr.io_type) {
12428                 case CTL_IO_SCSI:
12429                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12430                         sbuf_printf(&sb, "\n");
12431                         sbuf_cat(&sb, path_str);
12432                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12433                                     io->scsiio.tag_num, io->scsiio.tag_type);
12434                         break;
12435                 case CTL_IO_TASK:
12436                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12437                                     "Tag Type: %d\n", io->taskio.task_action,
12438                                     io->taskio.tag_num, io->taskio.tag_type);
12439                         break;
12440                 default:
12441                         panic("%s: Invalid CTL I/O type %d\n",
12442                             __func__, io->io_hdr.io_type);
12443                 }
12444                 sbuf_cat(&sb, path_str);
12445                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12446                             (intmax_t)time_uptime - io->io_hdr.start_time);
12447                 sbuf_finish(&sb);
12448                 printf("%s", sbuf_data(&sb));
12449         }
12450 #endif /* CTL_TIME_IO */
12451
12452 #ifdef CTL_IO_DELAY
12453         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12454                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12455         } else {
12456                 struct ctl_lun *lun;
12457
12458                 lun = CTL_LUN(io);
12459                 if ((lun != NULL)
12460                  && (lun->delay_info.datamove_delay > 0)) {
12461
12462                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12463                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12464                         callout_reset(&io->io_hdr.delay_callout,
12465                                       lun->delay_info.datamove_delay * hz,
12466                                       ctl_datamove_timer_wakeup, io);
12467                         if (lun->delay_info.datamove_type ==
12468                             CTL_DELAY_TYPE_ONESHOT)
12469                                 lun->delay_info.datamove_delay = 0;
12470                         return;
12471                 }
12472         }
12473 #endif
12474
12475         /*
12476          * This command has been aborted.  Set the port status, so we fail
12477          * the data move.
12478          */
12479         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12480                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12481                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12482                        io->io_hdr.nexus.targ_port,
12483                        io->io_hdr.nexus.targ_lun);
12484                 io->io_hdr.port_status = 31337;
12485                 /*
12486                  * Note that the backend, in this case, will get the
12487                  * callback in its context.  In other cases it may get
12488                  * called in the frontend's interrupt thread context.
12489                  */
12490                 io->scsiio.be_move_done(io);
12491                 return;
12492         }
12493
12494         /* Don't confuse frontend with zero length data move. */
12495         if (io->scsiio.kern_data_len == 0) {
12496                 io->scsiio.be_move_done(io);
12497                 return;
12498         }
12499
12500         fe_datamove = CTL_PORT(io)->fe_datamove;
12501         fe_datamove(io);
12502 }
12503
12504 static void
12505 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12506 {
12507         union ctl_ha_msg msg;
12508 #ifdef CTL_TIME_IO
12509         struct bintime cur_bt;
12510 #endif
12511
12512         memset(&msg, 0, sizeof(msg));
12513         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12514         msg.hdr.original_sc = io;
12515         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12516         msg.hdr.nexus = io->io_hdr.nexus;
12517         msg.hdr.status = io->io_hdr.status;
12518         msg.scsi.kern_data_resid = io->scsiio.kern_data_resid;
12519         msg.scsi.tag_num = io->scsiio.tag_num;
12520         msg.scsi.tag_type = io->scsiio.tag_type;
12521         msg.scsi.scsi_status = io->scsiio.scsi_status;
12522         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12523                io->scsiio.sense_len);
12524         msg.scsi.sense_len = io->scsiio.sense_len;
12525         msg.scsi.port_status = io->io_hdr.port_status;
12526         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12527         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12528                 ctl_failover_io(io, /*have_lock*/ have_lock);
12529                 return;
12530         }
12531         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12532             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12533             msg.scsi.sense_len, M_WAITOK);
12534
12535 #ifdef CTL_TIME_IO
12536         getbinuptime(&cur_bt);
12537         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12538         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12539 #endif
12540         io->io_hdr.num_dmas++;
12541 }
12542
12543 /*
12544  * The DMA to the remote side is done, now we need to tell the other side
12545  * we're done so it can continue with its data movement.
12546  */
12547 static void
12548 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12549 {
12550         union ctl_io *io;
12551         uint32_t i;
12552
12553         io = rq->context;
12554
12555         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12556                 printf("%s: ISC DMA write failed with error %d", __func__,
12557                        rq->ret);
12558                 ctl_set_internal_failure(&io->scsiio,
12559                                          /*sks_valid*/ 1,
12560                                          /*retry_count*/ rq->ret);
12561         }
12562
12563         ctl_dt_req_free(rq);
12564
12565         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12566                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12567         free(io->io_hdr.remote_sglist, M_CTL);
12568         io->io_hdr.remote_sglist = NULL;
12569         io->io_hdr.local_sglist = NULL;
12570
12571         /*
12572          * The data is in local and remote memory, so now we need to send
12573          * status (good or back) back to the other side.
12574          */
12575         ctl_send_datamove_done(io, /*have_lock*/ 0);
12576 }
12577
12578 /*
12579  * We've moved the data from the host/controller into local memory.  Now we
12580  * need to push it over to the remote controller's memory.
12581  */
12582 static int
12583 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12584 {
12585         int retval;
12586
12587         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12588                                           ctl_datamove_remote_write_cb);
12589         return (retval);
12590 }
12591
12592 static void
12593 ctl_datamove_remote_write(union ctl_io *io)
12594 {
12595         int retval;
12596         void (*fe_datamove)(union ctl_io *io);
12597
12598         /*
12599          * - Get the data from the host/HBA into local memory.
12600          * - DMA memory from the local controller to the remote controller.
12601          * - Send status back to the remote controller.
12602          */
12603
12604         retval = ctl_datamove_remote_sgl_setup(io);
12605         if (retval != 0)
12606                 return;
12607
12608         /* Switch the pointer over so the FETD knows what to do */
12609         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12610
12611         /*
12612          * Use a custom move done callback, since we need to send completion
12613          * back to the other controller, not to the backend on this side.
12614          */
12615         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12616
12617         fe_datamove = CTL_PORT(io)->fe_datamove;
12618         fe_datamove(io);
12619 }
12620
12621 static int
12622 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12623 {
12624 #if 0
12625         char str[256];
12626         char path_str[64];
12627         struct sbuf sb;
12628 #endif
12629         uint32_t i;
12630
12631         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12632                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12633         free(io->io_hdr.remote_sglist, M_CTL);
12634         io->io_hdr.remote_sglist = NULL;
12635         io->io_hdr.local_sglist = NULL;
12636
12637 #if 0
12638         scsi_path_string(io, path_str, sizeof(path_str));
12639         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12640         sbuf_cat(&sb, path_str);
12641         scsi_command_string(&io->scsiio, NULL, &sb);
12642         sbuf_printf(&sb, "\n");
12643         sbuf_cat(&sb, path_str);
12644         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12645                     io->scsiio.tag_num, io->scsiio.tag_type);
12646         sbuf_cat(&sb, path_str);
12647         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12648                     io->io_hdr.flags, io->io_hdr.status);
12649         sbuf_finish(&sb);
12650         printk("%s", sbuf_data(&sb));
12651 #endif
12652
12653
12654         /*
12655          * The read is done, now we need to send status (good or bad) back
12656          * to the other side.
12657          */
12658         ctl_send_datamove_done(io, /*have_lock*/ 0);
12659
12660         return (0);
12661 }
12662
12663 static void
12664 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12665 {
12666         union ctl_io *io;
12667         void (*fe_datamove)(union ctl_io *io);
12668
12669         io = rq->context;
12670
12671         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12672                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12673                        rq->ret);
12674                 ctl_set_internal_failure(&io->scsiio,
12675                                          /*sks_valid*/ 1,
12676                                          /*retry_count*/ rq->ret);
12677         }
12678
12679         ctl_dt_req_free(rq);
12680
12681         /* Switch the pointer over so the FETD knows what to do */
12682         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12683
12684         /*
12685          * Use a custom move done callback, since we need to send completion
12686          * back to the other controller, not to the backend on this side.
12687          */
12688         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12689
12690         /* XXX KDM add checks like the ones in ctl_datamove? */
12691
12692         fe_datamove = CTL_PORT(io)->fe_datamove;
12693         fe_datamove(io);
12694 }
12695
12696 static int
12697 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12698 {
12699         struct ctl_sg_entry *local_sglist;
12700         uint32_t len_to_go;
12701         int retval;
12702         int i;
12703
12704         retval = 0;
12705         local_sglist = io->io_hdr.local_sglist;
12706         len_to_go = io->scsiio.kern_data_len;
12707
12708         /*
12709          * The difficult thing here is that the size of the various
12710          * S/G segments may be different than the size from the
12711          * remote controller.  That'll make it harder when DMAing
12712          * the data back to the other side.
12713          */
12714         for (i = 0; len_to_go > 0; i++) {
12715                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12716                 local_sglist[i].addr =
12717                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12718
12719                 len_to_go -= local_sglist[i].len;
12720         }
12721         /*
12722          * Reset the number of S/G entries accordingly.  The original
12723          * number of S/G entries is available in rem_sg_entries.
12724          */
12725         io->scsiio.kern_sg_entries = i;
12726
12727 #if 0
12728         printf("%s: kern_sg_entries = %d\n", __func__,
12729                io->scsiio.kern_sg_entries);
12730         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12731                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12732                        local_sglist[i].addr, local_sglist[i].len);
12733 #endif
12734
12735         return (retval);
12736 }
12737
12738 static int
12739 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12740                          ctl_ha_dt_cb callback)
12741 {
12742         struct ctl_ha_dt_req *rq;
12743         struct ctl_sg_entry *remote_sglist, *local_sglist;
12744         uint32_t local_used, remote_used, total_used;
12745         int i, j, isc_ret;
12746
12747         rq = ctl_dt_req_alloc();
12748
12749         /*
12750          * If we failed to allocate the request, and if the DMA didn't fail
12751          * anyway, set busy status.  This is just a resource allocation
12752          * failure.
12753          */
12754         if ((rq == NULL)
12755          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12756              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12757                 ctl_set_busy(&io->scsiio);
12758
12759         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12760             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12761
12762                 if (rq != NULL)
12763                         ctl_dt_req_free(rq);
12764
12765                 /*
12766                  * The data move failed.  We need to return status back
12767                  * to the other controller.  No point in trying to DMA
12768                  * data to the remote controller.
12769                  */
12770
12771                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12772
12773                 return (1);
12774         }
12775
12776         local_sglist = io->io_hdr.local_sglist;
12777         remote_sglist = io->io_hdr.remote_sglist;
12778         local_used = 0;
12779         remote_used = 0;
12780         total_used = 0;
12781
12782         /*
12783          * Pull/push the data over the wire from/to the other controller.
12784          * This takes into account the possibility that the local and
12785          * remote sglists may not be identical in terms of the size of
12786          * the elements and the number of elements.
12787          *
12788          * One fundamental assumption here is that the length allocated for
12789          * both the local and remote sglists is identical.  Otherwise, we've
12790          * essentially got a coding error of some sort.
12791          */
12792         isc_ret = CTL_HA_STATUS_SUCCESS;
12793         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12794                 uint32_t cur_len;
12795                 uint8_t *tmp_ptr;
12796
12797                 rq->command = command;
12798                 rq->context = io;
12799
12800                 /*
12801                  * Both pointers should be aligned.  But it is possible
12802                  * that the allocation length is not.  They should both
12803                  * also have enough slack left over at the end, though,
12804                  * to round up to the next 8 byte boundary.
12805                  */
12806                 cur_len = MIN(local_sglist[i].len - local_used,
12807                               remote_sglist[j].len - remote_used);
12808                 rq->size = cur_len;
12809
12810                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12811                 tmp_ptr += local_used;
12812
12813 #if 0
12814                 /* Use physical addresses when talking to ISC hardware */
12815                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12816                         /* XXX KDM use busdma */
12817                         rq->local = vtophys(tmp_ptr);
12818                 } else
12819                         rq->local = tmp_ptr;
12820 #else
12821                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12822                     ("HA does not support BUS_ADDR"));
12823                 rq->local = tmp_ptr;
12824 #endif
12825
12826                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12827                 tmp_ptr += remote_used;
12828                 rq->remote = tmp_ptr;
12829
12830                 rq->callback = NULL;
12831
12832                 local_used += cur_len;
12833                 if (local_used >= local_sglist[i].len) {
12834                         i++;
12835                         local_used = 0;
12836                 }
12837
12838                 remote_used += cur_len;
12839                 if (remote_used >= remote_sglist[j].len) {
12840                         j++;
12841                         remote_used = 0;
12842                 }
12843                 total_used += cur_len;
12844
12845                 if (total_used >= io->scsiio.kern_data_len)
12846                         rq->callback = callback;
12847
12848 #if 0
12849                 printf("%s: %s: local %p remote %p size %d\n", __func__,
12850                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12851                        rq->local, rq->remote, rq->size);
12852 #endif
12853
12854                 isc_ret = ctl_dt_single(rq);
12855                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
12856                         break;
12857         }
12858         if (isc_ret != CTL_HA_STATUS_WAIT) {
12859                 rq->ret = isc_ret;
12860                 callback(rq);
12861         }
12862
12863         return (0);
12864 }
12865
12866 static void
12867 ctl_datamove_remote_read(union ctl_io *io)
12868 {
12869         int retval;
12870         uint32_t i;
12871
12872         /*
12873          * This will send an error to the other controller in the case of a
12874          * failure.
12875          */
12876         retval = ctl_datamove_remote_sgl_setup(io);
12877         if (retval != 0)
12878                 return;
12879
12880         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12881                                           ctl_datamove_remote_read_cb);
12882         if (retval != 0) {
12883                 /*
12884                  * Make sure we free memory if there was an error..  The
12885                  * ctl_datamove_remote_xfer() function will send the
12886                  * datamove done message, or call the callback with an
12887                  * error if there is a problem.
12888                  */
12889                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12890                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12891                 free(io->io_hdr.remote_sglist, M_CTL);
12892                 io->io_hdr.remote_sglist = NULL;
12893                 io->io_hdr.local_sglist = NULL;
12894         }
12895 }
12896
12897 /*
12898  * Process a datamove request from the other controller.  This is used for
12899  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12900  * first.  Once that is complete, the data gets DMAed into the remote
12901  * controller's memory.  For reads, we DMA from the remote controller's
12902  * memory into our memory first, and then move it out to the FETD.
12903  */
12904 static void
12905 ctl_datamove_remote(union ctl_io *io)
12906 {
12907
12908         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12909
12910         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12911                 ctl_failover_io(io, /*have_lock*/ 0);
12912                 return;
12913         }
12914
12915         /*
12916          * Note that we look for an aborted I/O here, but don't do some of
12917          * the other checks that ctl_datamove() normally does.
12918          * We don't need to run the datamove delay code, since that should
12919          * have been done if need be on the other controller.
12920          */
12921         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12922                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
12923                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12924                        io->io_hdr.nexus.targ_port,
12925                        io->io_hdr.nexus.targ_lun);
12926                 io->io_hdr.port_status = 31338;
12927                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12928                 return;
12929         }
12930
12931         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
12932                 ctl_datamove_remote_write(io);
12933         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
12934                 ctl_datamove_remote_read(io);
12935         else {
12936                 io->io_hdr.port_status = 31339;
12937                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12938         }
12939 }
12940
12941 static void
12942 ctl_process_done(union ctl_io *io)
12943 {
12944         struct ctl_softc *softc = CTL_SOFTC(io);
12945         struct ctl_port *port = CTL_PORT(io);
12946         struct ctl_lun *lun = CTL_LUN(io);
12947         void (*fe_done)(union ctl_io *io);
12948         union ctl_ha_msg msg;
12949
12950         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12951         fe_done = port->fe_done;
12952
12953 #ifdef CTL_TIME_IO
12954         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12955                 char str[256];
12956                 char path_str[64];
12957                 struct sbuf sb;
12958
12959                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12960                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12961
12962                 sbuf_cat(&sb, path_str);
12963                 switch (io->io_hdr.io_type) {
12964                 case CTL_IO_SCSI:
12965                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12966                         sbuf_printf(&sb, "\n");
12967                         sbuf_cat(&sb, path_str);
12968                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12969                                     io->scsiio.tag_num, io->scsiio.tag_type);
12970                         break;
12971                 case CTL_IO_TASK:
12972                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12973                                     "Tag Type: %d\n", io->taskio.task_action,
12974                                     io->taskio.tag_num, io->taskio.tag_type);
12975                         break;
12976                 default:
12977                         panic("%s: Invalid CTL I/O type %d\n",
12978                             __func__, io->io_hdr.io_type);
12979                 }
12980                 sbuf_cat(&sb, path_str);
12981                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12982                             (intmax_t)time_uptime - io->io_hdr.start_time);
12983                 sbuf_finish(&sb);
12984                 printf("%s", sbuf_data(&sb));
12985         }
12986 #endif /* CTL_TIME_IO */
12987
12988         switch (io->io_hdr.io_type) {
12989         case CTL_IO_SCSI:
12990                 break;
12991         case CTL_IO_TASK:
12992                 if (ctl_debug & CTL_DEBUG_INFO)
12993                         ctl_io_error_print(io, NULL);
12994                 fe_done(io);
12995                 return;
12996         default:
12997                 panic("%s: Invalid CTL I/O type %d\n",
12998                     __func__, io->io_hdr.io_type);
12999         }
13000
13001         if (lun == NULL) {
13002                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13003                                  io->io_hdr.nexus.targ_mapped_lun));
13004                 goto bailout;
13005         }
13006
13007         mtx_lock(&lun->lun_lock);
13008
13009         /*
13010          * Check to see if we have any informational exception and status
13011          * of this command can be modified to report it in form of either
13012          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
13013          */
13014         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
13015             io->io_hdr.status == CTL_SUCCESS &&
13016             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
13017                 uint8_t mrie = lun->MODE_IE.mrie;
13018                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
13019                     (lun->MODE_VER.byte3 & SMS_VER_PER));
13020                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
13021                      mrie == SIEP_MRIE_REC_UNCOND ||
13022                      mrie == SIEP_MRIE_NO_SENSE) &&
13023                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
13024                      CTL_CMD_FLAG_NO_SENSE) == 0) {
13025                         ctl_set_sense(&io->scsiio,
13026                               /*current_error*/ 1,
13027                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
13028                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
13029                               /*asc*/ lun->ie_asc,
13030                               /*ascq*/ lun->ie_ascq,
13031                               SSD_ELEM_NONE);
13032                         lun->ie_reported = 1;
13033                 }
13034         } else if (lun->ie_reported < 0)
13035                 lun->ie_reported = 0;
13036
13037         /*
13038          * Check to see if we have any errors to inject here.  We only
13039          * inject errors for commands that don't already have errors set.
13040          */
13041         if (!STAILQ_EMPTY(&lun->error_list) &&
13042             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13043             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13044                 ctl_inject_error(lun, io);
13045
13046         /*
13047          * XXX KDM how do we treat commands that aren't completed
13048          * successfully?
13049          *
13050          * XXX KDM should we also track I/O latency?
13051          */
13052         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13053             io->io_hdr.io_type == CTL_IO_SCSI) {
13054                 int type;
13055 #ifdef CTL_TIME_IO
13056                 struct bintime bt;
13057
13058                 getbinuptime(&bt);
13059                 bintime_sub(&bt, &io->io_hdr.start_bt);
13060 #endif
13061                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13062                     CTL_FLAG_DATA_IN)
13063                         type = CTL_STATS_READ;
13064                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13065                     CTL_FLAG_DATA_OUT)
13066                         type = CTL_STATS_WRITE;
13067                 else
13068                         type = CTL_STATS_NO_IO;
13069
13070 #ifdef CTL_LEGACY_STATS
13071                 uint32_t targ_port = port->targ_port;
13072                 lun->legacy_stats.ports[targ_port].bytes[type] +=
13073                     io->scsiio.kern_total_len;
13074                 lun->legacy_stats.ports[targ_port].operations[type] ++;
13075                 lun->legacy_stats.ports[targ_port].num_dmas[type] +=
13076                     io->io_hdr.num_dmas;
13077 #ifdef CTL_TIME_IO
13078                 bintime_add(&lun->legacy_stats.ports[targ_port].dma_time[type],
13079                    &io->io_hdr.dma_bt);
13080                 bintime_add(&lun->legacy_stats.ports[targ_port].time[type],
13081                     &bt);
13082 #endif
13083 #endif /* CTL_LEGACY_STATS */
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          * Remove this from the OOA queue.
13106          */
13107         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13108 #ifdef CTL_TIME_IO
13109         if (TAILQ_EMPTY(&lun->ooa_queue))
13110                 lun->last_busy = getsbinuptime();
13111 #endif
13112
13113         /*
13114          * Run through the blocked queue on this LUN and see if anything
13115          * has become unblocked, now that this transaction is done.
13116          */
13117         ctl_check_blocked(lun);
13118
13119         /*
13120          * If the LUN has been invalidated, free it if there is nothing
13121          * left on its OOA queue.
13122          */
13123         if ((lun->flags & CTL_LUN_INVALID)
13124          && TAILQ_EMPTY(&lun->ooa_queue)) {
13125                 mtx_unlock(&lun->lun_lock);
13126                 ctl_free_lun(lun);
13127         } else
13128                 mtx_unlock(&lun->lun_lock);
13129
13130 bailout:
13131
13132         /*
13133          * If this command has been aborted, make sure we set the status
13134          * properly.  The FETD is responsible for freeing the I/O and doing
13135          * whatever it needs to do to clean up its state.
13136          */
13137         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13138                 ctl_set_task_aborted(&io->scsiio);
13139
13140         /*
13141          * If enabled, print command error status.
13142          */
13143         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13144             (ctl_debug & CTL_DEBUG_INFO) != 0)
13145                 ctl_io_error_print(io, NULL);
13146
13147         /*
13148          * Tell the FETD or the other shelf controller we're done with this
13149          * command.  Note that only SCSI commands get to this point.  Task
13150          * management commands are completed above.
13151          */
13152         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13153             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13154                 memset(&msg, 0, sizeof(msg));
13155                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13156                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13157                 msg.hdr.nexus = io->io_hdr.nexus;
13158                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13159                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13160                     M_WAITOK);
13161         }
13162
13163         fe_done(io);
13164 }
13165
13166 /*
13167  * Front end should call this if it doesn't do autosense.  When the request
13168  * sense comes back in from the initiator, we'll dequeue this and send it.
13169  */
13170 int
13171 ctl_queue_sense(union ctl_io *io)
13172 {
13173         struct ctl_softc *softc = CTL_SOFTC(io);
13174         struct ctl_port *port = CTL_PORT(io);
13175         struct ctl_lun *lun;
13176         struct scsi_sense_data *ps;
13177         uint32_t initidx, p, targ_lun;
13178
13179         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13180
13181         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13182
13183         /*
13184          * LUN lookup will likely move to the ctl_work_thread() once we
13185          * have our new queueing infrastructure (that doesn't put things on
13186          * a per-LUN queue initially).  That is so that we can handle
13187          * things like an INQUIRY to a LUN that we don't have enabled.  We
13188          * can't deal with that right now.
13189          * If we don't have a LUN for this, just toss the sense information.
13190          */
13191         mtx_lock(&softc->ctl_lock);
13192         if (targ_lun >= ctl_max_luns ||
13193             (lun = softc->ctl_luns[targ_lun]) == NULL) {
13194                 mtx_unlock(&softc->ctl_lock);
13195                 goto bailout;
13196         }
13197         mtx_lock(&lun->lun_lock);
13198         mtx_unlock(&softc->ctl_lock);
13199
13200         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13201         p = initidx / CTL_MAX_INIT_PER_PORT;
13202         if (lun->pending_sense[p] == NULL) {
13203                 lun->pending_sense[p] = malloc(sizeof(*ps) * CTL_MAX_INIT_PER_PORT,
13204                     M_CTL, M_NOWAIT | M_ZERO);
13205         }
13206         if ((ps = lun->pending_sense[p]) != NULL) {
13207                 ps += initidx % CTL_MAX_INIT_PER_PORT;
13208                 memset(ps, 0, sizeof(*ps));
13209                 memcpy(ps, &io->scsiio.sense_data, io->scsiio.sense_len);
13210         }
13211         mtx_unlock(&lun->lun_lock);
13212
13213 bailout:
13214         ctl_free_io(io);
13215         return (CTL_RETVAL_COMPLETE);
13216 }
13217
13218 /*
13219  * Primary command inlet from frontend ports.  All SCSI and task I/O
13220  * requests must go through this function.
13221  */
13222 int
13223 ctl_queue(union ctl_io *io)
13224 {
13225         struct ctl_port *port = CTL_PORT(io);
13226
13227         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13228
13229 #ifdef CTL_TIME_IO
13230         io->io_hdr.start_time = time_uptime;
13231         getbinuptime(&io->io_hdr.start_bt);
13232 #endif /* CTL_TIME_IO */
13233
13234         /* Map FE-specific LUN ID into global one. */
13235         io->io_hdr.nexus.targ_mapped_lun =
13236             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13237
13238         switch (io->io_hdr.io_type) {
13239         case CTL_IO_SCSI:
13240         case CTL_IO_TASK:
13241                 if (ctl_debug & CTL_DEBUG_CDB)
13242                         ctl_io_print(io);
13243                 ctl_enqueue_incoming(io);
13244                 break;
13245         default:
13246                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13247                 return (EINVAL);
13248         }
13249
13250         return (CTL_RETVAL_COMPLETE);
13251 }
13252
13253 #ifdef CTL_IO_DELAY
13254 static void
13255 ctl_done_timer_wakeup(void *arg)
13256 {
13257         union ctl_io *io;
13258
13259         io = (union ctl_io *)arg;
13260         ctl_done(io);
13261 }
13262 #endif /* CTL_IO_DELAY */
13263
13264 void
13265 ctl_serseq_done(union ctl_io *io)
13266 {
13267         struct ctl_lun *lun = CTL_LUN(io);;
13268
13269         if (lun->be_lun == NULL ||
13270             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13271                 return;
13272         mtx_lock(&lun->lun_lock);
13273         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13274         ctl_check_blocked(lun);
13275         mtx_unlock(&lun->lun_lock);
13276 }
13277
13278 void
13279 ctl_done(union ctl_io *io)
13280 {
13281
13282         /*
13283          * Enable this to catch duplicate completion issues.
13284          */
13285 #if 0
13286         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13287                 printf("%s: type %d msg %d cdb %x iptl: "
13288                        "%u:%u:%u tag 0x%04x "
13289                        "flag %#x status %x\n",
13290                         __func__,
13291                         io->io_hdr.io_type,
13292                         io->io_hdr.msg_type,
13293                         io->scsiio.cdb[0],
13294                         io->io_hdr.nexus.initid,
13295                         io->io_hdr.nexus.targ_port,
13296                         io->io_hdr.nexus.targ_lun,
13297                         (io->io_hdr.io_type ==
13298                         CTL_IO_TASK) ?
13299                         io->taskio.tag_num :
13300                         io->scsiio.tag_num,
13301                         io->io_hdr.flags,
13302                         io->io_hdr.status);
13303         } else
13304                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13305 #endif
13306
13307         /*
13308          * This is an internal copy of an I/O, and should not go through
13309          * the normal done processing logic.
13310          */
13311         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13312                 return;
13313
13314 #ifdef CTL_IO_DELAY
13315         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13316                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13317         } else {
13318                 struct ctl_lun *lun = CTL_LUN(io);
13319
13320                 if ((lun != NULL)
13321                  && (lun->delay_info.done_delay > 0)) {
13322
13323                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13324                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13325                         callout_reset(&io->io_hdr.delay_callout,
13326                                       lun->delay_info.done_delay * hz,
13327                                       ctl_done_timer_wakeup, io);
13328                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13329                                 lun->delay_info.done_delay = 0;
13330                         return;
13331                 }
13332         }
13333 #endif /* CTL_IO_DELAY */
13334
13335         ctl_enqueue_done(io);
13336 }
13337
13338 static void
13339 ctl_work_thread(void *arg)
13340 {
13341         struct ctl_thread *thr = (struct ctl_thread *)arg;
13342         struct ctl_softc *softc = thr->ctl_softc;
13343         union ctl_io *io;
13344         int retval;
13345
13346         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13347
13348         while (!softc->shutdown) {
13349                 /*
13350                  * We handle the queues in this order:
13351                  * - ISC
13352                  * - done queue (to free up resources, unblock other commands)
13353                  * - RtR queue
13354                  * - incoming queue
13355                  *
13356                  * If those queues are empty, we break out of the loop and
13357                  * go to sleep.
13358                  */
13359                 mtx_lock(&thr->queue_lock);
13360                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13361                 if (io != NULL) {
13362                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13363                         mtx_unlock(&thr->queue_lock);
13364                         ctl_handle_isc(io);
13365                         continue;
13366                 }
13367                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13368                 if (io != NULL) {
13369                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13370                         /* clear any blocked commands, call fe_done */
13371                         mtx_unlock(&thr->queue_lock);
13372                         ctl_process_done(io);
13373                         continue;
13374                 }
13375                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13376                 if (io != NULL) {
13377                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13378                         mtx_unlock(&thr->queue_lock);
13379                         if (io->io_hdr.io_type == CTL_IO_TASK)
13380                                 ctl_run_task(io);
13381                         else
13382                                 ctl_scsiio_precheck(softc, &io->scsiio);
13383                         continue;
13384                 }
13385                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13386                 if (io != NULL) {
13387                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13388                         mtx_unlock(&thr->queue_lock);
13389                         retval = ctl_scsiio(&io->scsiio);
13390                         if (retval != CTL_RETVAL_COMPLETE)
13391                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13392                         continue;
13393                 }
13394
13395                 /* Sleep until we have something to do. */
13396                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13397         }
13398         thr->thread = NULL;
13399         kthread_exit();
13400 }
13401
13402 static void
13403 ctl_lun_thread(void *arg)
13404 {
13405         struct ctl_softc *softc = (struct ctl_softc *)arg;
13406         struct ctl_be_lun *be_lun;
13407
13408         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13409
13410         while (!softc->shutdown) {
13411                 mtx_lock(&softc->ctl_lock);
13412                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13413                 if (be_lun != NULL) {
13414                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13415                         mtx_unlock(&softc->ctl_lock);
13416                         ctl_create_lun(be_lun);
13417                         continue;
13418                 }
13419
13420                 /* Sleep until we have something to do. */
13421                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13422                     PDROP | PRIBIO, "-", 0);
13423         }
13424         softc->lun_thread = NULL;
13425         kthread_exit();
13426 }
13427
13428 static void
13429 ctl_thresh_thread(void *arg)
13430 {
13431         struct ctl_softc *softc = (struct ctl_softc *)arg;
13432         struct ctl_lun *lun;
13433         struct ctl_logical_block_provisioning_page *page;
13434         const char *attr;
13435         union ctl_ha_msg msg;
13436         uint64_t thres, val;
13437         int i, e, set;
13438
13439         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13440
13441         while (!softc->shutdown) {
13442                 mtx_lock(&softc->ctl_lock);
13443                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13444                         if ((lun->flags & CTL_LUN_DISABLED) ||
13445                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13446                             lun->backend->lun_attr == NULL)
13447                                 continue;
13448                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13449                             softc->ha_mode == CTL_HA_MODE_XFER)
13450                                 continue;
13451                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13452                                 continue;
13453                         e = 0;
13454                         page = &lun->MODE_LBP;
13455                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13456                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13457                                         continue;
13458                                 thres = scsi_4btoul(page->descr[i].count);
13459                                 thres <<= CTL_LBP_EXPONENT;
13460                                 switch (page->descr[i].resource) {
13461                                 case 0x01:
13462                                         attr = "blocksavail";
13463                                         break;
13464                                 case 0x02:
13465                                         attr = "blocksused";
13466                                         break;
13467                                 case 0xf1:
13468                                         attr = "poolblocksavail";
13469                                         break;
13470                                 case 0xf2:
13471                                         attr = "poolblocksused";
13472                                         break;
13473                                 default:
13474                                         continue;
13475                                 }
13476                                 mtx_unlock(&softc->ctl_lock); // XXX
13477                                 val = lun->backend->lun_attr(
13478                                     lun->be_lun->be_lun, attr);
13479                                 mtx_lock(&softc->ctl_lock);
13480                                 if (val == UINT64_MAX)
13481                                         continue;
13482                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13483                                     == SLBPPD_ARMING_INC)
13484                                         e = (val >= thres);
13485                                 else
13486                                         e = (val <= thres);
13487                                 if (e)
13488                                         break;
13489                         }
13490                         mtx_lock(&lun->lun_lock);
13491                         if (e) {
13492                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13493                                     (uint8_t *)page, lun->ua_tpt_info);
13494                                 if (lun->lasttpt == 0 ||
13495                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13496                                         lun->lasttpt = time_uptime;
13497                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13498                                         set = 1;
13499                                 } else
13500                                         set = 0;
13501                         } else {
13502                                 lun->lasttpt = 0;
13503                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13504                                 set = -1;
13505                         }
13506                         mtx_unlock(&lun->lun_lock);
13507                         if (set != 0 &&
13508                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13509                                 /* Send msg to other side. */
13510                                 bzero(&msg.ua, sizeof(msg.ua));
13511                                 msg.hdr.msg_type = CTL_MSG_UA;
13512                                 msg.hdr.nexus.initid = -1;
13513                                 msg.hdr.nexus.targ_port = -1;
13514                                 msg.hdr.nexus.targ_lun = lun->lun;
13515                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13516                                 msg.ua.ua_all = 1;
13517                                 msg.ua.ua_set = (set > 0);
13518                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13519                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13520                                 mtx_unlock(&softc->ctl_lock); // XXX
13521                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13522                                     sizeof(msg.ua), M_WAITOK);
13523                                 mtx_lock(&softc->ctl_lock);
13524                         }
13525                 }
13526                 mtx_sleep(&softc->thresh_thread, &softc->ctl_lock,
13527                     PDROP | PRIBIO, "-", CTL_LBP_PERIOD * hz);
13528         }
13529         softc->thresh_thread = NULL;
13530         kthread_exit();
13531 }
13532
13533 static void
13534 ctl_enqueue_incoming(union ctl_io *io)
13535 {
13536         struct ctl_softc *softc = CTL_SOFTC(io);
13537         struct ctl_thread *thr;
13538         u_int idx;
13539
13540         idx = (io->io_hdr.nexus.targ_port * 127 +
13541                io->io_hdr.nexus.initid) % worker_threads;
13542         thr = &softc->threads[idx];
13543         mtx_lock(&thr->queue_lock);
13544         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13545         mtx_unlock(&thr->queue_lock);
13546         wakeup(thr);
13547 }
13548
13549 static void
13550 ctl_enqueue_rtr(union ctl_io *io)
13551 {
13552         struct ctl_softc *softc = CTL_SOFTC(io);
13553         struct ctl_thread *thr;
13554
13555         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13556         mtx_lock(&thr->queue_lock);
13557         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13558         mtx_unlock(&thr->queue_lock);
13559         wakeup(thr);
13560 }
13561
13562 static void
13563 ctl_enqueue_done(union ctl_io *io)
13564 {
13565         struct ctl_softc *softc = CTL_SOFTC(io);
13566         struct ctl_thread *thr;
13567
13568         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13569         mtx_lock(&thr->queue_lock);
13570         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13571         mtx_unlock(&thr->queue_lock);
13572         wakeup(thr);
13573 }
13574
13575 static void
13576 ctl_enqueue_isc(union ctl_io *io)
13577 {
13578         struct ctl_softc *softc = CTL_SOFTC(io);
13579         struct ctl_thread *thr;
13580
13581         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13582         mtx_lock(&thr->queue_lock);
13583         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13584         mtx_unlock(&thr->queue_lock);
13585         wakeup(thr);
13586 }
13587
13588 /*
13589  *  vim: ts=8
13590  */