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 #define _CTL_C
  44
  45 #include <sys/cdefs.h>
  46 __FBSDID("$FreeBSD$");
  47
  48 #include <sys/param.h>
  49 #include <sys/systm.h>
  50 #include <sys/ctype.h>
  51 #include <sys/kernel.h>
  52 #include <sys/types.h>
  53 #include <sys/kthread.h>
  54 #include <sys/bio.h>
  55 #include <sys/fcntl.h>
  56 #include <sys/lock.h>
  57 #include <sys/module.h>
  58 #include <sys/mutex.h>
  59 #include <sys/condvar.h>
  60 #include <sys/malloc.h>
  61 #include <sys/conf.h>
  62 #include <sys/ioccom.h>
  63 #include <sys/queue.h>
  64 #include <sys/sbuf.h>
  65 #include <sys/smp.h>
  66 #include <sys/endian.h>
  67 #include <sys/sysctl.h>
  68 #include <vm/uma.h>
  69
  70 #include <cam/cam.h>
  71 #include <cam/scsi/scsi_all.h>
  72 #include <cam/scsi/scsi_cd.h>
  73 #include <cam/scsi/scsi_da.h>
  74 #include <cam/ctl/ctl_io.h>
  75 #include <cam/ctl/ctl.h>
  76 #include <cam/ctl/ctl_frontend.h>
  77 #include <cam/ctl/ctl_util.h>
  78 #include <cam/ctl/ctl_backend.h>
  79 #include <cam/ctl/ctl_ioctl.h>
  80 #include <cam/ctl/ctl_ha.h>
  81 #include <cam/ctl/ctl_private.h>
  82 #include <cam/ctl/ctl_debug.h>
  83 #include <cam/ctl/ctl_scsi_all.h>
  84 #include <cam/ctl/ctl_error.h>
  85
  86 struct ctl_softc *control_softc = NULL;
  87
  88 /*
  89  * Template mode pages.
  90  */
  91
  92 /*
  93  * Note that these are default values only.  The actual values will be
  94  * filled in when the user does a mode sense.
  95  */
  96 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
  97         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
  98         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
  99         /*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
 100         /*read_retry_count*/0,
 101         /*correction_span*/0,
 102         /*head_offset_count*/0,
 103         /*data_strobe_offset_cnt*/0,
 104         /*byte8*/SMS_RWER_LBPERE,
 105         /*write_retry_count*/0,
 106         /*reserved2*/0,
 107         /*recovery_time_limit*/{0, 0},
 108 };
 109
 110 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
 111         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 112         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 113         /*byte3*/SMS_RWER_PER,
 114         /*read_retry_count*/0,
 115         /*correction_span*/0,
 116         /*head_offset_count*/0,
 117         /*data_strobe_offset_cnt*/0,
 118         /*byte8*/SMS_RWER_LBPERE,
 119         /*write_retry_count*/0,
 120         /*reserved2*/0,
 121         /*recovery_time_limit*/{0, 0},
 122 };
 123
 124 const static struct scsi_format_page format_page_default = {
 125         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 126         /*page_length*/sizeof(struct scsi_format_page) - 2,
 127         /*tracks_per_zone*/ {0, 0},
 128         /*alt_sectors_per_zone*/ {0, 0},
 129         /*alt_tracks_per_zone*/ {0, 0},
 130         /*alt_tracks_per_lun*/ {0, 0},
 131         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
 132                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
 133         /*bytes_per_sector*/ {0, 0},
 134         /*interleave*/ {0, 0},
 135         /*track_skew*/ {0, 0},
 136         /*cylinder_skew*/ {0, 0},
 137         /*flags*/ SFP_HSEC,
 138         /*reserved*/ {0, 0, 0}
 139 };
 140
 141 const static struct scsi_format_page format_page_changeable = {
 142         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 143         /*page_length*/sizeof(struct scsi_format_page) - 2,
 144         /*tracks_per_zone*/ {0, 0},
 145         /*alt_sectors_per_zone*/ {0, 0},
 146         /*alt_tracks_per_zone*/ {0, 0},
 147         /*alt_tracks_per_lun*/ {0, 0},
 148         /*sectors_per_track*/ {0, 0},
 149         /*bytes_per_sector*/ {0, 0},
 150         /*interleave*/ {0, 0},
 151         /*track_skew*/ {0, 0},
 152         /*cylinder_skew*/ {0, 0},
 153         /*flags*/ 0,
 154         /*reserved*/ {0, 0, 0}
 155 };
 156
 157 const static struct scsi_rigid_disk_page rigid_disk_page_default = {
 158         /*page_code*/SMS_RIGID_DISK_PAGE,
 159         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 160         /*cylinders*/ {0, 0, 0},
 161         /*heads*/ CTL_DEFAULT_HEADS,
 162         /*start_write_precomp*/ {0, 0, 0},
 163         /*start_reduced_current*/ {0, 0, 0},
 164         /*step_rate*/ {0, 0},
 165         /*landing_zone_cylinder*/ {0, 0, 0},
 166         /*rpl*/ SRDP_RPL_DISABLED,
 167         /*rotational_offset*/ 0,
 168         /*reserved1*/ 0,
 169         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
 170                            CTL_DEFAULT_ROTATION_RATE & 0xff},
 171         /*reserved2*/ {0, 0}
 172 };
 173
 174 const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
 175         /*page_code*/SMS_RIGID_DISK_PAGE,
 176         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 177         /*cylinders*/ {0, 0, 0},
 178         /*heads*/ 0,
 179         /*start_write_precomp*/ {0, 0, 0},
 180         /*start_reduced_current*/ {0, 0, 0},
 181         /*step_rate*/ {0, 0},
 182         /*landing_zone_cylinder*/ {0, 0, 0},
 183         /*rpl*/ 0,
 184         /*rotational_offset*/ 0,
 185         /*reserved1*/ 0,
 186         /*rotation_rate*/ {0, 0},
 187         /*reserved2*/ {0, 0}
 188 };
 189
 190 const static struct scsi_da_verify_recovery_page verify_er_page_default = {
 191         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 192         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 193         /*byte3*/0,
 194         /*read_retry_count*/0,
 195         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 196         /*recovery_time_limit*/{0, 0},
 197 };
 198
 199 const static struct scsi_da_verify_recovery_page verify_er_page_changeable = {
 200         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 201         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 202         /*byte3*/SMS_VER_PER,
 203         /*read_retry_count*/0,
 204         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 205         /*recovery_time_limit*/{0, 0},
 206 };
 207
 208 const static struct scsi_caching_page caching_page_default = {
 209         /*page_code*/SMS_CACHING_PAGE,
 210         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 211         /*flags1*/ SCP_DISC | SCP_WCE,
 212         /*ret_priority*/ 0,
 213         /*disable_pf_transfer_len*/ {0xff, 0xff},
 214         /*min_prefetch*/ {0, 0},
 215         /*max_prefetch*/ {0xff, 0xff},
 216         /*max_pf_ceiling*/ {0xff, 0xff},
 217         /*flags2*/ 0,
 218         /*cache_segments*/ 0,
 219         /*cache_seg_size*/ {0, 0},
 220         /*reserved*/ 0,
 221         /*non_cache_seg_size*/ {0, 0, 0}
 222 };
 223
 224 const static struct scsi_caching_page caching_page_changeable = {
 225         /*page_code*/SMS_CACHING_PAGE,
 226         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 227         /*flags1*/ SCP_WCE | SCP_RCD,
 228         /*ret_priority*/ 0,
 229         /*disable_pf_transfer_len*/ {0, 0},
 230         /*min_prefetch*/ {0, 0},
 231         /*max_prefetch*/ {0, 0},
 232         /*max_pf_ceiling*/ {0, 0},
 233         /*flags2*/ 0,
 234         /*cache_segments*/ 0,
 235         /*cache_seg_size*/ {0, 0},
 236         /*reserved*/ 0,
 237         /*non_cache_seg_size*/ {0, 0, 0}
 238 };
 239
 240 const static struct scsi_control_page control_page_default = {
 241         /*page_code*/SMS_CONTROL_MODE_PAGE,
 242         /*page_length*/sizeof(struct scsi_control_page) - 2,
 243         /*rlec*/0,
 244         /*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
 245         /*eca_and_aen*/0,
 246         /*flags4*/SCP_TAS,
 247         /*aen_holdoff_period*/{0, 0},
 248         /*busy_timeout_period*/{0, 0},
 249         /*extended_selftest_completion_time*/{0, 0}
 250 };
 251
 252 const static struct scsi_control_page control_page_changeable = {
 253         /*page_code*/SMS_CONTROL_MODE_PAGE,
 254         /*page_length*/sizeof(struct scsi_control_page) - 2,
 255         /*rlec*/SCP_DSENSE,
 256         /*queue_flags*/SCP_QUEUE_ALG_MASK | SCP_NUAR,
 257         /*eca_and_aen*/SCP_SWP,
 258         /*flags4*/0,
 259         /*aen_holdoff_period*/{0, 0},
 260         /*busy_timeout_period*/{0, 0},
 261         /*extended_selftest_completion_time*/{0, 0}
 262 };
 263
 264 #define CTL_CEM_LEN     (sizeof(struct scsi_control_ext_page) - 4)
 265
 266 const static struct scsi_control_ext_page control_ext_page_default = {
 267         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 268         /*subpage_code*/0x01,
 269         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 270         /*flags*/0,
 271         /*prio*/0,
 272         /*max_sense*/0
 273 };
 274
 275 const static struct scsi_control_ext_page control_ext_page_changeable = {
 276         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 277         /*subpage_code*/0x01,
 278         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 279         /*flags*/0,
 280         /*prio*/0,
 281         /*max_sense*/0xff
 282 };
 283
 284 const static struct scsi_info_exceptions_page ie_page_default = {
 285         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 286         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 287         /*info_flags*/SIEP_FLAGS_EWASC,
 288         /*mrie*/SIEP_MRIE_NO,
 289         /*interval_timer*/{0, 0, 0, 0},
 290         /*report_count*/{0, 0, 0, 1}
 291 };
 292
 293 const static struct scsi_info_exceptions_page ie_page_changeable = {
 294         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 295         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 296         /*info_flags*/SIEP_FLAGS_EWASC | SIEP_FLAGS_DEXCPT | SIEP_FLAGS_TEST |
 297             SIEP_FLAGS_LOGERR,
 298         /*mrie*/0x0f,
 299         /*interval_timer*/{0xff, 0xff, 0xff, 0xff},
 300         /*report_count*/{0xff, 0xff, 0xff, 0xff}
 301 };
 302
 303 #define CTL_LBPM_LEN    (sizeof(struct ctl_logical_block_provisioning_page) - 4)
 304
 305 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
 306         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 307         /*subpage_code*/0x02,
 308         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 309         /*flags*/0,
 310         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 311         /*descr*/{}},
 312         {{/*flags*/0,
 313           /*resource*/0x01,
 314           /*reserved*/{0, 0},
 315           /*count*/{0, 0, 0, 0}},
 316          {/*flags*/0,
 317           /*resource*/0x02,
 318           /*reserved*/{0, 0},
 319           /*count*/{0, 0, 0, 0}},
 320          {/*flags*/0,
 321           /*resource*/0xf1,
 322           /*reserved*/{0, 0},
 323           /*count*/{0, 0, 0, 0}},
 324          {/*flags*/0,
 325           /*resource*/0xf2,
 326           /*reserved*/{0, 0},
 327           /*count*/{0, 0, 0, 0}}
 328         }
 329 };
 330
 331 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
 332         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 333         /*subpage_code*/0x02,
 334         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 335         /*flags*/SLBPP_SITUA,
 336         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 337         /*descr*/{}},
 338         {{/*flags*/0,
 339           /*resource*/0,
 340           /*reserved*/{0, 0},
 341           /*count*/{0, 0, 0, 0}},
 342          {/*flags*/0,
 343           /*resource*/0,
 344           /*reserved*/{0, 0},
 345           /*count*/{0, 0, 0, 0}},
 346          {/*flags*/0,
 347           /*resource*/0,
 348           /*reserved*/{0, 0},
 349           /*count*/{0, 0, 0, 0}},
 350          {/*flags*/0,
 351           /*resource*/0,
 352           /*reserved*/{0, 0},
 353           /*count*/{0, 0, 0, 0}}
 354         }
 355 };
 356
 357 const static struct scsi_cddvd_capabilities_page cddvd_page_default = {
 358         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 359         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 360         /*caps1*/0x3f,
 361         /*caps2*/0x00,
 362         /*caps3*/0xf0,
 363         /*caps4*/0x00,
 364         /*caps5*/0x29,
 365         /*caps6*/0x00,
 366         /*obsolete*/{0, 0},
 367         /*nvol_levels*/{0, 0},
 368         /*buffer_size*/{8, 0},
 369         /*obsolete2*/{0, 0},
 370         /*reserved*/0,
 371         /*digital*/0,
 372         /*obsolete3*/0,
 373         /*copy_management*/0,
 374         /*reserved2*/0,
 375         /*rotation_control*/0,
 376         /*cur_write_speed*/0,
 377         /*num_speed_descr*/0,
 378 };
 379
 380 const static struct scsi_cddvd_capabilities_page cddvd_page_changeable = {
 381         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 382         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 383         /*caps1*/0,
 384         /*caps2*/0,
 385         /*caps3*/0,
 386         /*caps4*/0,
 387         /*caps5*/0,
 388         /*caps6*/0,
 389         /*obsolete*/{0, 0},
 390         /*nvol_levels*/{0, 0},
 391         /*buffer_size*/{0, 0},
 392         /*obsolete2*/{0, 0},
 393         /*reserved*/0,
 394         /*digital*/0,
 395         /*obsolete3*/0,
 396         /*copy_management*/0,
 397         /*reserved2*/0,
 398         /*rotation_control*/0,
 399         /*cur_write_speed*/0,
 400         /*num_speed_descr*/0,
 401 };
 402
 403 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
 404 static int worker_threads = -1;
 405 TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
 406 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 407     &worker_threads, 1, "Number of worker threads");
 408 static int ctl_debug = CTL_DEBUG_NONE;
 409 TUNABLE_INT("kern.cam.ctl.debug", &ctl_debug);
 410 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
 411     &ctl_debug, 0, "Enabled debug flags");
 412 static int ctl_lun_map_size = 1024;
 413 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, lun_map_size, CTLFLAG_RWTUN,
 414     &ctl_lun_map_size, 0, "Size of per-port LUN map (max LUN + 1)");
 415
 416 /*
 417  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 418  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 419  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
 420  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
 421  */
 422 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
 423
 424 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 425                                   int param);
 426 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 427 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
 428 static int ctl_init(void);
 429 void ctl_shutdown(void);
 430 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 431 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 432 static void ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 433 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 434                               struct ctl_ooa *ooa_hdr,
 435                               struct ctl_ooa_entry *kern_entries);
 436 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 437                      struct thread *td);
 438 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 439                          struct ctl_be_lun *be_lun);
 440 static int ctl_free_lun(struct ctl_lun *lun);
 441 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 442
 443 static int ctl_do_mode_select(union ctl_io *io);
 444 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 445                            uint64_t res_key, uint64_t sa_res_key,
 446                            uint8_t type, uint32_t residx,
 447                            struct ctl_scsiio *ctsio,
 448                            struct scsi_per_res_out *cdb,
 449                            struct scsi_per_res_out_parms* param);
 450 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 451                                   union ctl_ha_msg *msg);
 452 static void ctl_hndl_per_res_out_on_other_sc(union ctl_io *io);
 453 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 454 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 455 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 456 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 457 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 458 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 459                                          int alloc_len);
 460 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 461                                          int alloc_len);
 462 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 463 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 464 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 465 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 466 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 467 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
 468     bool seq);
 469 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
 470 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 471     union ctl_io *pending_io, union ctl_io *ooa_io);
 472 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 473                                 union ctl_io *starting_io);
 474 static int ctl_check_blocked(struct ctl_lun *lun);
 475 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
 476                                 const struct ctl_cmd_entry *entry,
 477                                 struct ctl_scsiio *ctsio);
 478 static void ctl_failover_lun(union ctl_io *io);
 479 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 480                                struct ctl_scsiio *ctsio);
 481 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 482
 483 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 484 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
 485                             ctl_ua_type ua_type);
 486 static int ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io,
 487                          ctl_ua_type ua_type);
 488 static int ctl_lun_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 489 static int ctl_abort_task(union ctl_io *io);
 490 static int ctl_abort_task_set(union ctl_io *io);
 491 static int ctl_query_task(union ctl_io *io, int task_set);
 492 static int ctl_i_t_nexus_reset(union ctl_io *io);
 493 static int ctl_query_async_event(union ctl_io *io);
 494 static void ctl_run_task(union ctl_io *io);
 495 #ifdef CTL_IO_DELAY
 496 static void ctl_datamove_timer_wakeup(void *arg);
 497 static void ctl_done_timer_wakeup(void *arg);
 498 #endif /* CTL_IO_DELAY */
 499
 500 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 501 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 502 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 503 static void ctl_datamove_remote_write(union ctl_io *io);
 504 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 505 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 506 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 507 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 508                                     ctl_ha_dt_cb callback);
 509 static void ctl_datamove_remote_read(union ctl_io *io);
 510 static void ctl_datamove_remote(union ctl_io *io);
 511 static void ctl_process_done(union ctl_io *io);
 512 static void ctl_lun_thread(void *arg);
 513 static void ctl_thresh_thread(void *arg);
 514 static void ctl_work_thread(void *arg);
 515 static void ctl_enqueue_incoming(union ctl_io *io);
 516 static void ctl_enqueue_rtr(union ctl_io *io);
 517 static void ctl_enqueue_done(union ctl_io *io);
 518 static void ctl_enqueue_isc(union ctl_io *io);
 519 static const struct ctl_cmd_entry *
 520     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 521 static const struct ctl_cmd_entry *
 522     ctl_validate_command(struct ctl_scsiio *ctsio);
 523 static int ctl_cmd_applicable(uint8_t lun_type,
 524     const struct ctl_cmd_entry *entry);
 525
 526 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
 527 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
 528 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
 529 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
 530
 531 /*
 532  * Load the serialization table.  This isn't very pretty, but is probably
 533  * the easiest way to do it.
 534  */
 535 #include "ctl_ser_table.c"
 536
 537 /*
 538  * We only need to define open, close and ioctl routines for this driver.
 539  */
 540 static struct cdevsw ctl_cdevsw = {
 541         .d_version =    D_VERSION,
 542         .d_flags =      0,
 543         .d_open =       ctl_open,
 544         .d_close =      ctl_close,
 545         .d_ioctl =      ctl_ioctl,
 546         .d_name =       "ctl",
 547 };
 548
 549
 550 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 551
 552 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 553
 554 static moduledata_t ctl_moduledata = {
 555         "ctl",
 556         ctl_module_event_handler,
 557         NULL
 558 };
 559
 560 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 561 MODULE_VERSION(ctl, 1);
 562
 563 static struct ctl_frontend ha_frontend =
 564 {
 565         .name = "ha",
 566 };
 567
 568 static void
 569 ctl_ha_datamove(union ctl_io *io)
 570 {
 571         struct ctl_lun *lun = CTL_LUN(io);
 572         struct ctl_sg_entry *sgl;
 573         union ctl_ha_msg msg;
 574         uint32_t sg_entries_sent;
 575         int do_sg_copy, i, j;
 576
 577         memset(&msg.dt, 0, sizeof(msg.dt));
 578         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
 579         msg.hdr.original_sc = io->io_hdr.original_sc;
 580         msg.hdr.serializing_sc = io;
 581         msg.hdr.nexus = io->io_hdr.nexus;
 582         msg.hdr.status = io->io_hdr.status;
 583         msg.dt.flags = io->io_hdr.flags;
 584
 585         /*
 586          * We convert everything into a S/G list here.  We can't
 587          * pass by reference, only by value between controllers.
 588          * So we can't pass a pointer to the S/G list, only as many
 589          * S/G entries as we can fit in here.  If it's possible for
 590          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
 591          * then we need to break this up into multiple transfers.
 592          */
 593         if (io->scsiio.kern_sg_entries == 0) {
 594                 msg.dt.kern_sg_entries = 1;
 595 #if 0
 596                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 597                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 598                 } else {
 599                         /* XXX KDM use busdma here! */
 600                         msg.dt.sg_list[0].addr =
 601                             (void *)vtophys(io->scsiio.kern_data_ptr);
 602                 }
 603 #else
 604                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
 605                     ("HA does not support BUS_ADDR"));
 606                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 607 #endif
 608                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
 609                 do_sg_copy = 0;
 610         } else {
 611                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
 612                 do_sg_copy = 1;
 613         }
 614
 615         msg.dt.kern_data_len = io->scsiio.kern_data_len;
 616         msg.dt.kern_total_len = io->scsiio.kern_total_len;
 617         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
 618         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
 619         msg.dt.sg_sequence = 0;
 620
 621         /*
 622          * Loop until we've sent all of the S/G entries.  On the
 623          * other end, we'll recompose these S/G entries into one
 624          * contiguous list before processing.
 625          */
 626         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
 627             msg.dt.sg_sequence++) {
 628                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
 629                     sizeof(msg.dt.sg_list[0])),
 630                     msg.dt.kern_sg_entries - sg_entries_sent);
 631                 if (do_sg_copy != 0) {
 632                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
 633                         for (i = sg_entries_sent, j = 0;
 634                              i < msg.dt.cur_sg_entries; i++, j++) {
 635 #if 0
 636                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 637                                         msg.dt.sg_list[j].addr = sgl[i].addr;
 638                                 } else {
 639                                         /* XXX KDM use busdma here! */
 640                                         msg.dt.sg_list[j].addr =
 641                                             (void *)vtophys(sgl[i].addr);
 642                                 }
 643 #else
 644                                 KASSERT((io->io_hdr.flags &
 645                                     CTL_FLAG_BUS_ADDR) == 0,
 646                                     ("HA does not support BUS_ADDR"));
 647                                 msg.dt.sg_list[j].addr = sgl[i].addr;
 648 #endif
 649                                 msg.dt.sg_list[j].len = sgl[i].len;
 650                         }
 651                 }
 652
 653                 sg_entries_sent += msg.dt.cur_sg_entries;
 654                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
 655                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 656                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
 657                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
 658                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
 659                         io->io_hdr.port_status = 31341;
 660                         io->scsiio.be_move_done(io);
 661                         return;
 662                 }
 663                 msg.dt.sent_sg_entries = sg_entries_sent;
 664         }
 665
 666         /*
 667          * Officially handover the request from us to peer.
 668          * If failover has just happened, then we must return error.
 669          * If failover happen just after, then it is not our problem.
 670          */
 671         if (lun)
 672                 mtx_lock(&lun->lun_lock);
 673         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
 674                 if (lun)
 675                         mtx_unlock(&lun->lun_lock);
 676                 io->io_hdr.port_status = 31342;
 677                 io->scsiio.be_move_done(io);
 678                 return;
 679         }
 680         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
 681         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
 682         if (lun)
 683                 mtx_unlock(&lun->lun_lock);
 684 }
 685
 686 static void
 687 ctl_ha_done(union ctl_io *io)
 688 {
 689         union ctl_ha_msg msg;
 690
 691         if (io->io_hdr.io_type == CTL_IO_SCSI) {
 692                 memset(&msg, 0, sizeof(msg));
 693                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
 694                 msg.hdr.original_sc = io->io_hdr.original_sc;
 695                 msg.hdr.nexus = io->io_hdr.nexus;
 696                 msg.hdr.status = io->io_hdr.status;
 697                 msg.scsi.scsi_status = io->scsiio.scsi_status;
 698                 msg.scsi.tag_num = io->scsiio.tag_num;
 699                 msg.scsi.tag_type = io->scsiio.tag_type;
 700                 msg.scsi.sense_len = io->scsiio.sense_len;
 701                 msg.scsi.sense_residual = io->scsiio.sense_residual;
 702                 msg.scsi.residual = io->scsiio.residual;
 703                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
 704                     io->scsiio.sense_len);
 705                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 706                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
 707                     msg.scsi.sense_len, M_WAITOK);
 708         }
 709         ctl_free_io(io);
 710 }
 711
 712 static void
 713 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 714                             union ctl_ha_msg *msg_info)
 715 {
 716         struct ctl_scsiio *ctsio;
 717
 718         if (msg_info->hdr.original_sc == NULL) {
 719                 printf("%s: original_sc == NULL!\n", __func__);
 720                 /* XXX KDM now what? */
 721                 return;
 722         }
 723
 724         ctsio = &msg_info->hdr.original_sc->scsiio;
 725         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 726         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 727         ctsio->io_hdr.status = msg_info->hdr.status;
 728         ctsio->scsi_status = msg_info->scsi.scsi_status;
 729         ctsio->sense_len = msg_info->scsi.sense_len;
 730         ctsio->sense_residual = msg_info->scsi.sense_residual;
 731         ctsio->residual = msg_info->scsi.residual;
 732         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 733                msg_info->scsi.sense_len);
 734         ctl_enqueue_isc((union ctl_io *)ctsio);
 735 }
 736
 737 static void
 738 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 739                                 union ctl_ha_msg *msg_info)
 740 {
 741         struct ctl_scsiio *ctsio;
 742
 743         if (msg_info->hdr.serializing_sc == NULL) {
 744                 printf("%s: serializing_sc == NULL!\n", __func__);
 745                 /* XXX KDM now what? */
 746                 return;
 747         }
 748
 749         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 750         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 751         ctl_enqueue_isc((union ctl_io *)ctsio);
 752 }
 753
 754 void
 755 ctl_isc_announce_lun(struct ctl_lun *lun)
 756 {
 757         struct ctl_softc *softc = lun->ctl_softc;
 758         union ctl_ha_msg *msg;
 759         struct ctl_ha_msg_lun_pr_key pr_key;
 760         int i, k;
 761
 762         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 763                 return;
 764         mtx_lock(&lun->lun_lock);
 765         i = sizeof(msg->lun);
 766         if (lun->lun_devid)
 767                 i += lun->lun_devid->len;
 768         i += sizeof(pr_key) * lun->pr_key_count;
 769 alloc:
 770         mtx_unlock(&lun->lun_lock);
 771         msg = malloc(i, M_CTL, M_WAITOK);
 772         mtx_lock(&lun->lun_lock);
 773         k = sizeof(msg->lun);
 774         if (lun->lun_devid)
 775                 k += lun->lun_devid->len;
 776         k += sizeof(pr_key) * lun->pr_key_count;
 777         if (i < k) {
 778                 free(msg, M_CTL);
 779                 i = k;
 780                 goto alloc;
 781         }
 782         bzero(&msg->lun, sizeof(msg->lun));
 783         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 784         msg->hdr.nexus.targ_lun = lun->lun;
 785         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 786         msg->lun.flags = lun->flags;
 787         msg->lun.pr_generation = lun->pr_generation;
 788         msg->lun.pr_res_idx = lun->pr_res_idx;
 789         msg->lun.pr_res_type = lun->pr_res_type;
 790         msg->lun.pr_key_count = lun->pr_key_count;
 791         i = 0;
 792         if (lun->lun_devid) {
 793                 msg->lun.lun_devid_len = lun->lun_devid->len;
 794                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 795                     msg->lun.lun_devid_len);
 796                 i += msg->lun.lun_devid_len;
 797         }
 798         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 799                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 800                         continue;
 801                 pr_key.pr_iid = k;
 802                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 803                 i += sizeof(pr_key);
 804         }
 805         mtx_unlock(&lun->lun_lock);
 806         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 807             M_WAITOK);
 808         free(msg, M_CTL);
 809
 810         if (lun->flags & CTL_LUN_PRIMARY_SC) {
 811                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 812                         ctl_isc_announce_mode(lun, -1,
 813                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
 814                             lun->mode_pages.index[i].subpage);
 815                 }
 816         }
 817 }
 818
 819 void
 820 ctl_isc_announce_port(struct ctl_port *port)
 821 {
 822         struct ctl_softc *softc = port->ctl_softc;
 823         union ctl_ha_msg *msg;
 824         int i;
 825
 826         if (port->targ_port < softc->port_min ||
 827             port->targ_port >= softc->port_max ||
 828             softc->ha_link != CTL_HA_LINK_ONLINE)
 829                 return;
 830         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 831         if (port->lun_map)
 832                 i += port->lun_map_size * sizeof(uint32_t);
 833         if (port->port_devid)
 834                 i += port->port_devid->len;
 835         if (port->target_devid)
 836                 i += port->target_devid->len;
 837         if (port->init_devid)
 838                 i += port->init_devid->len;
 839         msg = malloc(i, M_CTL, M_WAITOK);
 840         bzero(&msg->port, sizeof(msg->port));
 841         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 842         msg->hdr.nexus.targ_port = port->targ_port;
 843         msg->port.port_type = port->port_type;
 844         msg->port.physical_port = port->physical_port;
 845         msg->port.virtual_port = port->virtual_port;
 846         msg->port.status = port->status;
 847         i = 0;
 848         msg->port.name_len = sprintf(&msg->port.data[i],
 849             "%d:%s", softc->ha_id, port->port_name) + 1;
 850         i += msg->port.name_len;
 851         if (port->lun_map) {
 852                 msg->port.lun_map_len = port->lun_map_size * sizeof(uint32_t);
 853                 memcpy(&msg->port.data[i], port->lun_map,
 854                     msg->port.lun_map_len);
 855                 i += msg->port.lun_map_len;
 856         }
 857         if (port->port_devid) {
 858                 msg->port.port_devid_len = port->port_devid->len;
 859                 memcpy(&msg->port.data[i], port->port_devid->data,
 860                     msg->port.port_devid_len);
 861                 i += msg->port.port_devid_len;
 862         }
 863         if (port->target_devid) {
 864                 msg->port.target_devid_len = port->target_devid->len;
 865                 memcpy(&msg->port.data[i], port->target_devid->data,
 866                     msg->port.target_devid_len);
 867                 i += msg->port.target_devid_len;
 868         }
 869         if (port->init_devid) {
 870                 msg->port.init_devid_len = port->init_devid->len;
 871                 memcpy(&msg->port.data[i], port->init_devid->data,
 872                     msg->port.init_devid_len);
 873                 i += msg->port.init_devid_len;
 874         }
 875         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 876             M_WAITOK);
 877         free(msg, M_CTL);
 878 }
 879
 880 void
 881 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 882 {
 883         struct ctl_softc *softc = port->ctl_softc;
 884         union ctl_ha_msg *msg;
 885         int i, l;
 886
 887         if (port->targ_port < softc->port_min ||
 888             port->targ_port >= softc->port_max ||
 889             softc->ha_link != CTL_HA_LINK_ONLINE)
 890                 return;
 891         mtx_lock(&softc->ctl_lock);
 892         i = sizeof(msg->iid);
 893         l = 0;
 894         if (port->wwpn_iid[iid].name)
 895                 l = strlen(port->wwpn_iid[iid].name) + 1;
 896         i += l;
 897         msg = malloc(i, M_CTL, M_NOWAIT);
 898         if (msg == NULL) {
 899                 mtx_unlock(&softc->ctl_lock);
 900                 return;
 901         }
 902         bzero(&msg->iid, sizeof(msg->iid));
 903         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 904         msg->hdr.nexus.targ_port = port->targ_port;
 905         msg->hdr.nexus.initid = iid;
 906         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 907         msg->iid.name_len = l;
 908         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 909         if (port->wwpn_iid[iid].name)
 910                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 911         mtx_unlock(&softc->ctl_lock);
 912         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 913         free(msg, M_CTL);
 914 }
 915
 916 void
 917 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
 918     uint8_t page, uint8_t subpage)
 919 {
 920         struct ctl_softc *softc = lun->ctl_softc;
 921         union ctl_ha_msg msg;
 922         u_int i;
 923
 924         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 925                 return;
 926         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 927                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
 928                     page && lun->mode_pages.index[i].subpage == subpage)
 929                         break;
 930         }
 931         if (i == CTL_NUM_MODE_PAGES)
 932                 return;
 933
 934         /* Don't try to replicate pages not present on this device. */
 935         if (lun->mode_pages.index[i].page_data == NULL)
 936                 return;
 937
 938         bzero(&msg.mode, sizeof(msg.mode));
 939         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
 940         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
 941         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
 942         msg.hdr.nexus.targ_lun = lun->lun;
 943         msg.hdr.nexus.targ_mapped_lun = lun->lun;
 944         msg.mode.page_code = page;
 945         msg.mode.subpage = subpage;
 946         msg.mode.page_len = lun->mode_pages.index[i].page_len;
 947         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
 948             msg.mode.page_len);
 949         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
 950             M_WAITOK);
 951 }
 952
 953 static void
 954 ctl_isc_ha_link_up(struct ctl_softc *softc)
 955 {
 956         struct ctl_port *port;
 957         struct ctl_lun *lun;
 958         union ctl_ha_msg msg;
 959         int i;
 960
 961         /* Announce this node parameters to peer for validation. */
 962         msg.login.msg_type = CTL_MSG_LOGIN;
 963         msg.login.version = CTL_HA_VERSION;
 964         msg.login.ha_mode = softc->ha_mode;
 965         msg.login.ha_id = softc->ha_id;
 966         msg.login.max_luns = CTL_MAX_LUNS;
 967         msg.login.max_ports = CTL_MAX_PORTS;
 968         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
 969         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
 970             M_WAITOK);
 971
 972         STAILQ_FOREACH(port, &softc->port_list, links) {
 973                 ctl_isc_announce_port(port);
 974                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
 975                         if (port->wwpn_iid[i].in_use)
 976                                 ctl_isc_announce_iid(port, i);
 977                 }
 978         }
 979         STAILQ_FOREACH(lun, &softc->lun_list, links)
 980                 ctl_isc_announce_lun(lun);
 981 }
 982
 983 static void
 984 ctl_isc_ha_link_down(struct ctl_softc *softc)
 985 {
 986         struct ctl_port *port;
 987         struct ctl_lun *lun;
 988         union ctl_io *io;
 989         int i;
 990
 991         mtx_lock(&softc->ctl_lock);
 992         STAILQ_FOREACH(lun, &softc->lun_list, links) {
 993                 mtx_lock(&lun->lun_lock);
 994                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
 995                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
 996                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
 997                 }
 998                 mtx_unlock(&lun->lun_lock);
 999
1000                 mtx_unlock(&softc->ctl_lock);
1001                 io = ctl_alloc_io(softc->othersc_pool);
1002                 mtx_lock(&softc->ctl_lock);
1003                 ctl_zero_io(io);
1004                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1005                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1006                 ctl_enqueue_isc(io);
1007         }
1008
1009         STAILQ_FOREACH(port, &softc->port_list, links) {
1010                 if (port->targ_port >= softc->port_min &&
1011                     port->targ_port < softc->port_max)
1012                         continue;
1013                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1014                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1015                         port->wwpn_iid[i].in_use = 0;
1016                         free(port->wwpn_iid[i].name, M_CTL);
1017                         port->wwpn_iid[i].name = NULL;
1018                 }
1019         }
1020         mtx_unlock(&softc->ctl_lock);
1021 }
1022
1023 static void
1024 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1025 {
1026         struct ctl_lun *lun;
1027         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
1028
1029         mtx_lock(&softc->ctl_lock);
1030         if (msg->hdr.nexus.targ_mapped_lun >= CTL_MAX_LUNS ||
1031             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) == NULL) {
1032                 mtx_unlock(&softc->ctl_lock);
1033                 return;
1034         }
1035         mtx_lock(&lun->lun_lock);
1036         mtx_unlock(&softc->ctl_lock);
1037         if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES && msg->ua.ua_set)
1038                 memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1039         if (msg->ua.ua_all) {
1040                 if (msg->ua.ua_set)
1041                         ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1042                 else
1043                         ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1044         } else {
1045                 if (msg->ua.ua_set)
1046                         ctl_est_ua(lun, iid, msg->ua.ua_type);
1047                 else
1048                         ctl_clr_ua(lun, iid, msg->ua.ua_type);
1049         }
1050         mtx_unlock(&lun->lun_lock);
1051 }
1052
1053 static void
1054 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1055 {
1056         struct ctl_lun *lun;
1057         struct ctl_ha_msg_lun_pr_key pr_key;
1058         int i, k;
1059         ctl_lun_flags oflags;
1060         uint32_t targ_lun;
1061
1062         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1063         mtx_lock(&softc->ctl_lock);
1064         if (targ_lun >= CTL_MAX_LUNS ||
1065             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1066                 mtx_unlock(&softc->ctl_lock);
1067                 return;
1068         }
1069         mtx_lock(&lun->lun_lock);
1070         mtx_unlock(&softc->ctl_lock);
1071         if (lun->flags & CTL_LUN_DISABLED) {
1072                 mtx_unlock(&lun->lun_lock);
1073                 return;
1074         }
1075         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1076         if (msg->lun.lun_devid_len != i || (i > 0 &&
1077             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1078                 mtx_unlock(&lun->lun_lock);
1079                 printf("%s: Received conflicting HA LUN %d\n",
1080                     __func__, targ_lun);
1081                 return;
1082         } else {
1083                 /* Record whether peer is primary. */
1084                 oflags = lun->flags;
1085                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1086                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1087                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1088                 else
1089                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1090                 if (oflags != lun->flags)
1091                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1092
1093                 /* If peer is primary and we are not -- use data */
1094                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1095                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1096                         lun->pr_generation = msg->lun.pr_generation;
1097                         lun->pr_res_idx = msg->lun.pr_res_idx;
1098                         lun->pr_res_type = msg->lun.pr_res_type;
1099                         lun->pr_key_count = msg->lun.pr_key_count;
1100                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1101                                 ctl_clr_prkey(lun, k);
1102                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1103                                 memcpy(&pr_key, &msg->lun.data[i],
1104                                     sizeof(pr_key));
1105                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1106                                 ctl_set_prkey(lun, pr_key.pr_iid,
1107                                     pr_key.pr_key);
1108                                 i += sizeof(pr_key);
1109                         }
1110                 }
1111
1112                 mtx_unlock(&lun->lun_lock);
1113                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1114                     __func__, targ_lun,
1115                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1116                     "primary" : "secondary"));
1117
1118                 /* If we are primary but peer doesn't know -- notify */
1119                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1120                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1121                         ctl_isc_announce_lun(lun);
1122         }
1123 }
1124
1125 static void
1126 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1127 {
1128         struct ctl_port *port;
1129         struct ctl_lun *lun;
1130         int i, new;
1131
1132         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1133         if (port == NULL) {
1134                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1135                     msg->hdr.nexus.targ_port));
1136                 new = 1;
1137                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1138                 port->frontend = &ha_frontend;
1139                 port->targ_port = msg->hdr.nexus.targ_port;
1140                 port->fe_datamove = ctl_ha_datamove;
1141                 port->fe_done = ctl_ha_done;
1142         } else if (port->frontend == &ha_frontend) {
1143                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1144                     msg->hdr.nexus.targ_port));
1145                 new = 0;
1146         } else {
1147                 printf("%s: Received conflicting HA port %d\n",
1148                     __func__, msg->hdr.nexus.targ_port);
1149                 return;
1150         }
1151         port->port_type = msg->port.port_type;
1152         port->physical_port = msg->port.physical_port;
1153         port->virtual_port = msg->port.virtual_port;
1154         port->status = msg->port.status;
1155         i = 0;
1156         free(port->port_name, M_CTL);
1157         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1158             M_CTL);
1159         i += msg->port.name_len;
1160         if (msg->port.lun_map_len != 0) {
1161                 if (port->lun_map == NULL ||
1162                     port->lun_map_size * sizeof(uint32_t) <
1163                     msg->port.lun_map_len) {
1164                         port->lun_map_size = 0;
1165                         free(port->lun_map, M_CTL);
1166                         port->lun_map = malloc(msg->port.lun_map_len,
1167                             M_CTL, M_WAITOK);
1168                 }
1169                 memcpy(port->lun_map, &msg->port.data[i], msg->port.lun_map_len);
1170                 port->lun_map_size = msg->port.lun_map_len / sizeof(uint32_t);
1171                 i += msg->port.lun_map_len;
1172         } else {
1173                 port->lun_map_size = 0;
1174                 free(port->lun_map, M_CTL);
1175                 port->lun_map = NULL;
1176         }
1177         if (msg->port.port_devid_len != 0) {
1178                 if (port->port_devid == NULL ||
1179                     port->port_devid->len < msg->port.port_devid_len) {
1180                         free(port->port_devid, M_CTL);
1181                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1182                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1183                 }
1184                 memcpy(port->port_devid->data, &msg->port.data[i],
1185                     msg->port.port_devid_len);
1186                 port->port_devid->len = msg->port.port_devid_len;
1187                 i += msg->port.port_devid_len;
1188         } else {
1189                 free(port->port_devid, M_CTL);
1190                 port->port_devid = NULL;
1191         }
1192         if (msg->port.target_devid_len != 0) {
1193                 if (port->target_devid == NULL ||
1194                     port->target_devid->len < msg->port.target_devid_len) {
1195                         free(port->target_devid, M_CTL);
1196                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1197                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1198                 }
1199                 memcpy(port->target_devid->data, &msg->port.data[i],
1200                     msg->port.target_devid_len);
1201                 port->target_devid->len = msg->port.target_devid_len;
1202                 i += msg->port.target_devid_len;
1203         } else {
1204                 free(port->target_devid, M_CTL);
1205                 port->target_devid = NULL;
1206         }
1207         if (msg->port.init_devid_len != 0) {
1208                 if (port->init_devid == NULL ||
1209                     port->init_devid->len < msg->port.init_devid_len) {
1210                         free(port->init_devid, M_CTL);
1211                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1212                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1213                 }
1214                 memcpy(port->init_devid->data, &msg->port.data[i],
1215                     msg->port.init_devid_len);
1216                 port->init_devid->len = msg->port.init_devid_len;
1217                 i += msg->port.init_devid_len;
1218         } else {
1219                 free(port->init_devid, M_CTL);
1220                 port->init_devid = NULL;
1221         }
1222         if (new) {
1223                 if (ctl_port_register(port) != 0) {
1224                         printf("%s: ctl_port_register() failed with error\n",
1225                             __func__);
1226                 }
1227         }
1228         mtx_lock(&softc->ctl_lock);
1229         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1230                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
1231                         continue;
1232                 mtx_lock(&lun->lun_lock);
1233                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1234                 mtx_unlock(&lun->lun_lock);
1235         }
1236         mtx_unlock(&softc->ctl_lock);
1237 }
1238
1239 static void
1240 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1241 {
1242         struct ctl_port *port;
1243         int iid;
1244
1245         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1246         if (port == NULL) {
1247                 printf("%s: Received IID for unknown port %d\n",
1248                     __func__, msg->hdr.nexus.targ_port);
1249                 return;
1250         }
1251         iid = msg->hdr.nexus.initid;
1252         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1253         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1254         free(port->wwpn_iid[iid].name, M_CTL);
1255         if (msg->iid.name_len) {
1256                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1257                     msg->iid.name_len, M_CTL);
1258         } else
1259                 port->wwpn_iid[iid].name = NULL;
1260 }
1261
1262 static void
1263 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1264 {
1265
1266         if (msg->login.version != CTL_HA_VERSION) {
1267                 printf("CTL HA peers have different versions %d != %d\n",
1268                     msg->login.version, CTL_HA_VERSION);
1269                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1270                 return;
1271         }
1272         if (msg->login.ha_mode != softc->ha_mode) {
1273                 printf("CTL HA peers have different ha_mode %d != %d\n",
1274                     msg->login.ha_mode, softc->ha_mode);
1275                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1276                 return;
1277         }
1278         if (msg->login.ha_id == softc->ha_id) {
1279                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1280                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1281                 return;
1282         }
1283         if (msg->login.max_luns != CTL_MAX_LUNS ||
1284             msg->login.max_ports != CTL_MAX_PORTS ||
1285             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1286                 printf("CTL HA peers have different limits\n");
1287                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1288                 return;
1289         }
1290 }
1291
1292 static void
1293 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1294 {
1295         struct ctl_lun *lun;
1296         u_int i;
1297         uint32_t initidx, targ_lun;
1298
1299         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1300         mtx_lock(&softc->ctl_lock);
1301         if (targ_lun >= CTL_MAX_LUNS ||
1302             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1303                 mtx_unlock(&softc->ctl_lock);
1304                 return;
1305         }
1306         mtx_lock(&lun->lun_lock);
1307         mtx_unlock(&softc->ctl_lock);
1308         if (lun->flags & CTL_LUN_DISABLED) {
1309                 mtx_unlock(&lun->lun_lock);
1310                 return;
1311         }
1312         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1313                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1314                     msg->mode.page_code &&
1315                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1316                         break;
1317         }
1318         if (i == CTL_NUM_MODE_PAGES) {
1319                 mtx_unlock(&lun->lun_lock);
1320                 return;
1321         }
1322         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1323             lun->mode_pages.index[i].page_len);
1324         initidx = ctl_get_initindex(&msg->hdr.nexus);
1325         if (initidx != -1)
1326                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1327         mtx_unlock(&lun->lun_lock);
1328 }
1329
1330 /*
1331  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1332  * subsystem come in here.
1333  */
1334 static void
1335 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1336 {
1337         struct ctl_softc *softc = control_softc;
1338         union ctl_io *io;
1339         struct ctl_prio *presio;
1340         ctl_ha_status isc_status;
1341
1342         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1343         if (event == CTL_HA_EVT_MSG_RECV) {
1344                 union ctl_ha_msg *msg, msgbuf;
1345
1346                 if (param > sizeof(msgbuf))
1347                         msg = malloc(param, M_CTL, M_WAITOK);
1348                 else
1349                         msg = &msgbuf;
1350                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1351                     M_WAITOK);
1352                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1353                         printf("%s: Error receiving message: %d\n",
1354                             __func__, isc_status);
1355                         if (msg != &msgbuf)
1356                                 free(msg, M_CTL);
1357                         return;
1358                 }
1359
1360                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1361                 switch (msg->hdr.msg_type) {
1362                 case CTL_MSG_SERIALIZE:
1363                         io = ctl_alloc_io(softc->othersc_pool);
1364                         ctl_zero_io(io);
1365                         // populate ctsio from msg
1366                         io->io_hdr.io_type = CTL_IO_SCSI;
1367                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1368                         io->io_hdr.original_sc = msg->hdr.original_sc;
1369                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1370                                             CTL_FLAG_IO_ACTIVE;
1371                         /*
1372                          * If we're in serialization-only mode, we don't
1373                          * want to go through full done processing.  Thus
1374                          * the COPY flag.
1375                          *
1376                          * XXX KDM add another flag that is more specific.
1377                          */
1378                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1379                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1380                         io->io_hdr.nexus = msg->hdr.nexus;
1381 #if 0
1382                         printf("port %u, iid %u, lun %u\n",
1383                                io->io_hdr.nexus.targ_port,
1384                                io->io_hdr.nexus.initid,
1385                                io->io_hdr.nexus.targ_lun);
1386 #endif
1387                         io->scsiio.tag_num = msg->scsi.tag_num;
1388                         io->scsiio.tag_type = msg->scsi.tag_type;
1389 #ifdef CTL_TIME_IO
1390                         io->io_hdr.start_time = time_uptime;
1391                         getbinuptime(&io->io_hdr.start_bt);
1392 #endif /* CTL_TIME_IO */
1393                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1394                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1395                                CTL_MAX_CDBLEN);
1396                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1397                                 const struct ctl_cmd_entry *entry;
1398
1399                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1400                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1401                                 io->io_hdr.flags |=
1402                                         entry->flags & CTL_FLAG_DATA_MASK;
1403                         }
1404                         ctl_enqueue_isc(io);
1405                         break;
1406
1407                 /* Performed on the Originating SC, XFER mode only */
1408                 case CTL_MSG_DATAMOVE: {
1409                         struct ctl_sg_entry *sgl;
1410                         int i, j;
1411
1412                         io = msg->hdr.original_sc;
1413                         if (io == NULL) {
1414                                 printf("%s: original_sc == NULL!\n", __func__);
1415                                 /* XXX KDM do something here */
1416                                 break;
1417                         }
1418                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1419                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1420                         /*
1421                          * Keep track of this, we need to send it back over
1422                          * when the datamove is complete.
1423                          */
1424                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1425                         if (msg->hdr.status == CTL_SUCCESS)
1426                                 io->io_hdr.status = msg->hdr.status;
1427
1428                         if (msg->dt.sg_sequence == 0) {
1429 #ifdef CTL_TIME_IO
1430                                 getbinuptime(&io->io_hdr.dma_start_bt);
1431 #endif
1432                                 i = msg->dt.kern_sg_entries +
1433                                     msg->dt.kern_data_len /
1434                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1435                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1436                                     M_WAITOK | M_ZERO);
1437                                 io->io_hdr.remote_sglist = sgl;
1438                                 io->io_hdr.local_sglist =
1439                                     &sgl[msg->dt.kern_sg_entries];
1440
1441                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1442
1443                                 io->scsiio.kern_sg_entries =
1444                                         msg->dt.kern_sg_entries;
1445                                 io->scsiio.rem_sg_entries =
1446                                         msg->dt.kern_sg_entries;
1447                                 io->scsiio.kern_data_len =
1448                                         msg->dt.kern_data_len;
1449                                 io->scsiio.kern_total_len =
1450                                         msg->dt.kern_total_len;
1451                                 io->scsiio.kern_data_resid =
1452                                         msg->dt.kern_data_resid;
1453                                 io->scsiio.kern_rel_offset =
1454                                         msg->dt.kern_rel_offset;
1455                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1456                                 io->io_hdr.flags |= msg->dt.flags &
1457                                     CTL_FLAG_BUS_ADDR;
1458                         } else
1459                                 sgl = (struct ctl_sg_entry *)
1460                                         io->scsiio.kern_data_ptr;
1461
1462                         for (i = msg->dt.sent_sg_entries, j = 0;
1463                              i < (msg->dt.sent_sg_entries +
1464                              msg->dt.cur_sg_entries); i++, j++) {
1465                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1466                                 sgl[i].len = msg->dt.sg_list[j].len;
1467
1468 #if 0
1469                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1470                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1471 #endif
1472                         }
1473
1474                         /*
1475                          * If this is the last piece of the I/O, we've got
1476                          * the full S/G list.  Queue processing in the thread.
1477                          * Otherwise wait for the next piece.
1478                          */
1479                         if (msg->dt.sg_last != 0)
1480                                 ctl_enqueue_isc(io);
1481                         break;
1482                 }
1483                 /* Performed on the Serializing (primary) SC, XFER mode only */
1484                 case CTL_MSG_DATAMOVE_DONE: {
1485                         if (msg->hdr.serializing_sc == NULL) {
1486                                 printf("%s: serializing_sc == NULL!\n",
1487                                        __func__);
1488                                 /* XXX KDM now what? */
1489                                 break;
1490                         }
1491                         /*
1492                          * We grab the sense information here in case
1493                          * there was a failure, so we can return status
1494                          * back to the initiator.
1495                          */
1496                         io = msg->hdr.serializing_sc;
1497                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1498                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1499                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1500                         io->io_hdr.port_status = msg->scsi.fetd_status;
1501                         io->scsiio.residual = msg->scsi.residual;
1502                         if (msg->hdr.status != CTL_STATUS_NONE) {
1503                                 io->io_hdr.status = msg->hdr.status;
1504                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1505                                 io->scsiio.sense_len = msg->scsi.sense_len;
1506                                 io->scsiio.sense_residual =msg->scsi.sense_residual;
1507                                 memcpy(&io->scsiio.sense_data,
1508                                     &msg->scsi.sense_data,
1509                                     msg->scsi.sense_len);
1510                                 if (msg->hdr.status == CTL_SUCCESS)
1511                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1512                         }
1513                         ctl_enqueue_isc(io);
1514                         break;
1515                 }
1516
1517                 /* Preformed on Originating SC, SER_ONLY mode */
1518                 case CTL_MSG_R2R:
1519                         io = msg->hdr.original_sc;
1520                         if (io == NULL) {
1521                                 printf("%s: original_sc == NULL!\n",
1522                                     __func__);
1523                                 break;
1524                         }
1525                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1526                         io->io_hdr.msg_type = CTL_MSG_R2R;
1527                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1528                         ctl_enqueue_isc(io);
1529                         break;
1530
1531                 /*
1532                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1533                  * mode.
1534                  * Performed on the Originating (i.e. secondary) SC in XFER
1535                  * mode
1536                  */
1537                 case CTL_MSG_FINISH_IO:
1538                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1539                                 ctl_isc_handler_finish_xfer(softc, msg);
1540                         else
1541                                 ctl_isc_handler_finish_ser_only(softc, msg);
1542                         break;
1543
1544                 /* Preformed on Originating SC */
1545                 case CTL_MSG_BAD_JUJU:
1546                         io = msg->hdr.original_sc;
1547                         if (io == NULL) {
1548                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1549                                        __func__);
1550                                 break;
1551                         }
1552                         ctl_copy_sense_data(msg, io);
1553                         /*
1554                          * IO should have already been cleaned up on other
1555                          * SC so clear this flag so we won't send a message
1556                          * back to finish the IO there.
1557                          */
1558                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1559                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1560
1561                         /* io = msg->hdr.serializing_sc; */
1562                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1563                         ctl_enqueue_isc(io);
1564                         break;
1565
1566                 /* Handle resets sent from the other side */
1567                 case CTL_MSG_MANAGE_TASKS: {
1568                         struct ctl_taskio *taskio;
1569                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1570                             softc->othersc_pool);
1571                         ctl_zero_io((union ctl_io *)taskio);
1572                         taskio->io_hdr.io_type = CTL_IO_TASK;
1573                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1574                         taskio->io_hdr.nexus = msg->hdr.nexus;
1575                         taskio->task_action = msg->task.task_action;
1576                         taskio->tag_num = msg->task.tag_num;
1577                         taskio->tag_type = msg->task.tag_type;
1578 #ifdef CTL_TIME_IO
1579                         taskio->io_hdr.start_time = time_uptime;
1580                         getbinuptime(&taskio->io_hdr.start_bt);
1581 #endif /* CTL_TIME_IO */
1582                         ctl_run_task((union ctl_io *)taskio);
1583                         break;
1584                 }
1585                 /* Persistent Reserve action which needs attention */
1586                 case CTL_MSG_PERS_ACTION:
1587                         presio = (struct ctl_prio *)ctl_alloc_io(
1588                             softc->othersc_pool);
1589                         ctl_zero_io((union ctl_io *)presio);
1590                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1591                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1592                         presio->io_hdr.nexus = msg->hdr.nexus;
1593                         presio->pr_msg = msg->pr;
1594                         ctl_enqueue_isc((union ctl_io *)presio);
1595                         break;
1596                 case CTL_MSG_UA:
1597                         ctl_isc_ua(softc, msg, param);
1598                         break;
1599                 case CTL_MSG_PORT_SYNC:
1600                         ctl_isc_port_sync(softc, msg, param);
1601                         break;
1602                 case CTL_MSG_LUN_SYNC:
1603                         ctl_isc_lun_sync(softc, msg, param);
1604                         break;
1605                 case CTL_MSG_IID_SYNC:
1606                         ctl_isc_iid_sync(softc, msg, param);
1607                         break;
1608                 case CTL_MSG_LOGIN:
1609                         ctl_isc_login(softc, msg, param);
1610                         break;
1611                 case CTL_MSG_MODE_SYNC:
1612                         ctl_isc_mode_sync(softc, msg, param);
1613                         break;
1614                 default:
1615                         printf("Received HA message of unknown type %d\n",
1616                             msg->hdr.msg_type);
1617                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1618                         break;
1619                 }
1620                 if (msg != &msgbuf)
1621                         free(msg, M_CTL);
1622         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1623                 printf("CTL: HA link status changed from %d to %d\n",
1624                     softc->ha_link, param);
1625                 if (param == softc->ha_link)
1626                         return;
1627                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1628                         softc->ha_link = param;
1629                         ctl_isc_ha_link_down(softc);
1630                 } else {
1631                         softc->ha_link = param;
1632                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1633                                 ctl_isc_ha_link_up(softc);
1634                 }
1635                 return;
1636         } else {
1637                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1638                 return;
1639         }
1640 }
1641
1642 static void
1643 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1644 {
1645
1646         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1647             src->scsi.sense_len);
1648         dest->scsiio.scsi_status = src->scsi.scsi_status;
1649         dest->scsiio.sense_len = src->scsi.sense_len;
1650         dest->io_hdr.status = src->hdr.status;
1651 }
1652
1653 static void
1654 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1655 {
1656
1657         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1658             src->scsiio.sense_len);
1659         dest->scsi.scsi_status = src->scsiio.scsi_status;
1660         dest->scsi.sense_len = src->scsiio.sense_len;
1661         dest->hdr.status = src->io_hdr.status;
1662 }
1663
1664 void
1665 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1666 {
1667         struct ctl_softc *softc = lun->ctl_softc;
1668         ctl_ua_type *pu;
1669
1670         if (initidx < softc->init_min || initidx >= softc->init_max)
1671                 return;
1672         mtx_assert(&lun->lun_lock, MA_OWNED);
1673         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1674         if (pu == NULL)
1675                 return;
1676         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1677 }
1678
1679 void
1680 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1681 {
1682         int i;
1683
1684         mtx_assert(&lun->lun_lock, MA_OWNED);
1685         if (lun->pending_ua[port] == NULL)
1686                 return;
1687         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1688                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1689                         continue;
1690                 lun->pending_ua[port][i] |= ua;
1691         }
1692 }
1693
1694 void
1695 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1696 {
1697         struct ctl_softc *softc = lun->ctl_softc;
1698         int i;
1699
1700         mtx_assert(&lun->lun_lock, MA_OWNED);
1701         for (i = softc->port_min; i < softc->port_max; i++)
1702                 ctl_est_ua_port(lun, i, except, ua);
1703 }
1704
1705 void
1706 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1707 {
1708         struct ctl_softc *softc = lun->ctl_softc;
1709         ctl_ua_type *pu;
1710
1711         if (initidx < softc->init_min || initidx >= softc->init_max)
1712                 return;
1713         mtx_assert(&lun->lun_lock, MA_OWNED);
1714         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1715         if (pu == NULL)
1716                 return;
1717         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1718 }
1719
1720 void
1721 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1722 {
1723         struct ctl_softc *softc = lun->ctl_softc;
1724         int i, j;
1725
1726         mtx_assert(&lun->lun_lock, MA_OWNED);
1727         for (i = softc->port_min; i < softc->port_max; i++) {
1728                 if (lun->pending_ua[i] == NULL)
1729                         continue;
1730                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1731                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1732                                 continue;
1733                         lun->pending_ua[i][j] &= ~ua;
1734                 }
1735         }
1736 }
1737
1738 void
1739 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1740     ctl_ua_type ua_type)
1741 {
1742         struct ctl_lun *lun;
1743
1744         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1745         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1746                 mtx_lock(&lun->lun_lock);
1747                 ctl_clr_ua(lun, initidx, ua_type);
1748                 mtx_unlock(&lun->lun_lock);
1749         }
1750 }
1751
1752 static int
1753 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1754 {
1755         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1756         struct ctl_lun *lun;
1757         struct ctl_lun_req ireq;
1758         int error, value;
1759
1760         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1761         error = sysctl_handle_int(oidp, &value, 0, req);
1762         if ((error != 0) || (req->newptr == NULL))
1763                 return (error);
1764
1765         mtx_lock(&softc->ctl_lock);
1766         if (value == 0)
1767                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1768         else
1769                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1770         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1771                 mtx_unlock(&softc->ctl_lock);
1772                 bzero(&ireq, sizeof(ireq));
1773                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1774                 ireq.reqdata.modify.lun_id = lun->lun;
1775                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1776                     curthread);
1777                 if (ireq.status != CTL_LUN_OK) {
1778                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1779                             __func__, ireq.status, ireq.error_str);
1780                 }
1781                 mtx_lock(&softc->ctl_lock);
1782         }
1783         mtx_unlock(&softc->ctl_lock);
1784         return (0);
1785 }
1786
1787 static int
1788 ctl_init(void)
1789 {
1790         struct make_dev_args args;
1791         struct ctl_softc *softc;
1792         void *other_pool;
1793         int i, error;
1794
1795         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1796                                M_WAITOK | M_ZERO);
1797
1798         make_dev_args_init(&args);
1799         args.mda_devsw = &ctl_cdevsw;
1800         args.mda_uid = UID_ROOT;
1801         args.mda_gid = GID_OPERATOR;
1802         args.mda_mode = 0600;
1803         args.mda_si_drv1 = softc;
1804         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1805         if (error != 0) {
1806                 free(softc, M_DEVBUF);
1807                 control_softc = NULL;
1808                 return (error);
1809         }
1810
1811         sysctl_ctx_init(&softc->sysctl_ctx);
1812         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1813                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1814                 CTLFLAG_RD, 0, "CAM Target Layer");
1815
1816         if (softc->sysctl_tree == NULL) {
1817                 printf("%s: unable to allocate sysctl tree\n", __func__);
1818                 destroy_dev(softc->dev);
1819                 free(softc, M_DEVBUF);
1820                 control_softc = NULL;
1821                 return (ENOMEM);
1822         }
1823
1824         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1825         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1826             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1827         softc->flags = 0;
1828
1829         TUNABLE_INT_FETCH("kern.cam.ctl.ha_mode", (int *)&softc->ha_mode);
1830         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1831             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1832             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1833
1834         /*
1835          * In Copan's HA scheme, the "master" and "slave" roles are
1836          * figured out through the slot the controller is in.  Although it
1837          * is an active/active system, someone has to be in charge.
1838          */
1839         TUNABLE_INT_FETCH("kern.cam.ctl.ha_id", &softc->ha_id);
1840         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1841             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1842             "HA head ID (0 - no HA)");
1843         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1844                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1845                 softc->is_single = 1;
1846                 softc->port_cnt = CTL_MAX_PORTS;
1847                 softc->port_min = 0;
1848         } else {
1849                 softc->port_cnt = CTL_MAX_PORTS / NUM_HA_SHELVES;
1850                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1851         }
1852         softc->port_max = softc->port_min + softc->port_cnt;
1853         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1854         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1855
1856         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1857             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1858             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1859
1860         STAILQ_INIT(&softc->lun_list);
1861         STAILQ_INIT(&softc->pending_lun_queue);
1862         STAILQ_INIT(&softc->fe_list);
1863         STAILQ_INIT(&softc->port_list);
1864         STAILQ_INIT(&softc->be_list);
1865         ctl_tpc_init(softc);
1866
1867         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1868                             &other_pool) != 0)
1869         {
1870                 printf("ctl: can't allocate %d entry other SC pool, "
1871                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1872                 return (ENOMEM);
1873         }
1874         softc->othersc_pool = other_pool;
1875
1876         if (worker_threads <= 0)
1877                 worker_threads = max(1, mp_ncpus / 4);
1878         if (worker_threads > CTL_MAX_THREADS)
1879                 worker_threads = CTL_MAX_THREADS;
1880
1881         for (i = 0; i < worker_threads; i++) {
1882                 struct ctl_thread *thr = &softc->threads[i];
1883
1884                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1885                 thr->ctl_softc = softc;
1886                 STAILQ_INIT(&thr->incoming_queue);
1887                 STAILQ_INIT(&thr->rtr_queue);
1888                 STAILQ_INIT(&thr->done_queue);
1889                 STAILQ_INIT(&thr->isc_queue);
1890
1891                 error = kproc_kthread_add(ctl_work_thread, thr,
1892                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1893                 if (error != 0) {
1894                         printf("error creating CTL work thread!\n");
1895                         ctl_pool_free(other_pool);
1896                         return (error);
1897                 }
1898         }
1899         error = kproc_kthread_add(ctl_lun_thread, softc,
1900             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1901         if (error != 0) {
1902                 printf("error creating CTL lun thread!\n");
1903                 ctl_pool_free(other_pool);
1904                 return (error);
1905         }
1906         error = kproc_kthread_add(ctl_thresh_thread, softc,
1907             &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1908         if (error != 0) {
1909                 printf("error creating CTL threshold thread!\n");
1910                 ctl_pool_free(other_pool);
1911                 return (error);
1912         }
1913
1914         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1915             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1916             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1917
1918         if (softc->is_single == 0) {
1919                 ctl_frontend_register(&ha_frontend);
1920                 if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
1921                         printf("ctl_init: ctl_ha_msg_init failed.\n");
1922                         softc->is_single = 1;
1923                 } else
1924                 if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
1925                     != CTL_HA_STATUS_SUCCESS) {
1926                         printf("ctl_init: ctl_ha_msg_register failed.\n");
1927                         softc->is_single = 1;
1928                 }
1929         }
1930         return (0);
1931 }
1932
1933 void
1934 ctl_shutdown(void)
1935 {
1936         struct ctl_softc *softc = control_softc;
1937         struct ctl_lun *lun, *next_lun;
1938
1939         if (softc->is_single == 0) {
1940                 ctl_ha_msg_shutdown(softc);
1941                 if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL)
1942                     != CTL_HA_STATUS_SUCCESS)
1943                         printf("%s: ctl_ha_msg_deregister failed.\n", __func__);
1944                 if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
1945                         printf("%s: ctl_ha_msg_destroy failed.\n", __func__);
1946                 ctl_frontend_deregister(&ha_frontend);
1947         }
1948
1949         mtx_lock(&softc->ctl_lock);
1950
1951         STAILQ_FOREACH_SAFE(lun, &softc->lun_list, links, next_lun)
1952                 ctl_free_lun(lun);
1953
1954         mtx_unlock(&softc->ctl_lock);
1955
1956 #if 0
1957         ctl_shutdown_thread(softc->work_thread);
1958         mtx_destroy(&softc->queue_lock);
1959 #endif
1960
1961         ctl_tpc_shutdown(softc);
1962         uma_zdestroy(softc->io_zone);
1963         mtx_destroy(&softc->ctl_lock);
1964
1965         destroy_dev(softc->dev);
1966
1967         sysctl_ctx_free(&softc->sysctl_ctx);
1968
1969         free(softc, M_DEVBUF);
1970         control_softc = NULL;
1971 }
1972
1973 static int
1974 ctl_module_event_handler(module_t mod, int what, void *arg)
1975 {
1976
1977         switch (what) {
1978         case MOD_LOAD:
1979                 return (ctl_init());
1980         case MOD_UNLOAD:
1981                 return (EBUSY);
1982         default:
1983                 return (EOPNOTSUPP);
1984         }
1985 }
1986
1987 /*
1988  * XXX KDM should we do some access checks here?  Bump a reference count to
1989  * prevent a CTL module from being unloaded while someone has it open?
1990  */
1991 static int
1992 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1993 {
1994         return (0);
1995 }
1996
1997 static int
1998 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1999 {
2000         return (0);
2001 }
2002
2003 /*
2004  * Remove an initiator by port number and initiator ID.
2005  * Returns 0 for success, -1 for failure.
2006  */
2007 int
2008 ctl_remove_initiator(struct ctl_port *port, int iid)
2009 {
2010         struct ctl_softc *softc = port->ctl_softc;
2011
2012         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2013
2014         if (iid > CTL_MAX_INIT_PER_PORT) {
2015                 printf("%s: initiator ID %u > maximun %u!\n",
2016                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2017                 return (-1);
2018         }
2019
2020         mtx_lock(&softc->ctl_lock);
2021         port->wwpn_iid[iid].in_use--;
2022         port->wwpn_iid[iid].last_use = time_uptime;
2023         mtx_unlock(&softc->ctl_lock);
2024         ctl_isc_announce_iid(port, iid);
2025
2026         return (0);
2027 }
2028
2029 /*
2030  * Add an initiator to the initiator map.
2031  * Returns iid for success, < 0 for failure.
2032  */
2033 int
2034 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2035 {
2036         struct ctl_softc *softc = port->ctl_softc;
2037         time_t best_time;
2038         int i, best;
2039
2040         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2041
2042         if (iid >= CTL_MAX_INIT_PER_PORT) {
2043                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2044                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2045                 free(name, M_CTL);
2046                 return (-1);
2047         }
2048
2049         mtx_lock(&softc->ctl_lock);
2050
2051         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2052                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2053                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2054                                 iid = i;
2055                                 break;
2056                         }
2057                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2058                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2059                                 iid = i;
2060                                 break;
2061                         }
2062                 }
2063         }
2064
2065         if (iid < 0) {
2066                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2067                         if (port->wwpn_iid[i].in_use == 0 &&
2068                             port->wwpn_iid[i].wwpn == 0 &&
2069                             port->wwpn_iid[i].name == NULL) {
2070                                 iid = i;
2071                                 break;
2072                         }
2073                 }
2074         }
2075
2076         if (iid < 0) {
2077                 best = -1;
2078                 best_time = INT32_MAX;
2079                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2080                         if (port->wwpn_iid[i].in_use == 0) {
2081                                 if (port->wwpn_iid[i].last_use < best_time) {
2082                                         best = i;
2083                                         best_time = port->wwpn_iid[i].last_use;
2084                                 }
2085                         }
2086                 }
2087                 iid = best;
2088         }
2089
2090         if (iid < 0) {
2091                 mtx_unlock(&softc->ctl_lock);
2092                 free(name, M_CTL);
2093                 return (-2);
2094         }
2095
2096         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2097                 /*
2098                  * This is not an error yet.
2099                  */
2100                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2101 #if 0
2102                         printf("%s: port %d iid %u WWPN %#jx arrived"
2103                             " again\n", __func__, port->targ_port,
2104                             iid, (uintmax_t)wwpn);
2105 #endif
2106                         goto take;
2107                 }
2108                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2109                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2110 #if 0
2111                         printf("%s: port %d iid %u name '%s' arrived"
2112                             " again\n", __func__, port->targ_port,
2113                             iid, name);
2114 #endif
2115                         goto take;
2116                 }
2117
2118                 /*
2119                  * This is an error, but what do we do about it?  The
2120                  * driver is telling us we have a new WWPN for this
2121                  * initiator ID, so we pretty much need to use it.
2122                  */
2123                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2124                     " but WWPN %#jx '%s' is still at that address\n",
2125                     __func__, port->targ_port, iid, wwpn, name,
2126                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2127                     port->wwpn_iid[iid].name);
2128
2129                 /*
2130                  * XXX KDM clear have_ca and ua_pending on each LUN for
2131                  * this initiator.
2132                  */
2133         }
2134 take:
2135         free(port->wwpn_iid[iid].name, M_CTL);
2136         port->wwpn_iid[iid].name = name;
2137         port->wwpn_iid[iid].wwpn = wwpn;
2138         port->wwpn_iid[iid].in_use++;
2139         mtx_unlock(&softc->ctl_lock);
2140         ctl_isc_announce_iid(port, iid);
2141
2142         return (iid);
2143 }
2144
2145 static int
2146 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2147 {
2148         int len;
2149
2150         switch (port->port_type) {
2151         case CTL_PORT_FC:
2152         {
2153                 struct scsi_transportid_fcp *id =
2154                     (struct scsi_transportid_fcp *)buf;
2155                 if (port->wwpn_iid[iid].wwpn == 0)
2156                         return (0);
2157                 memset(id, 0, sizeof(*id));
2158                 id->format_protocol = SCSI_PROTO_FC;
2159                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2160                 return (sizeof(*id));
2161         }
2162         case CTL_PORT_ISCSI:
2163         {
2164                 struct scsi_transportid_iscsi_port *id =
2165                     (struct scsi_transportid_iscsi_port *)buf;
2166                 if (port->wwpn_iid[iid].name == NULL)
2167                         return (0);
2168                 memset(id, 0, 256);
2169                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2170                     SCSI_PROTO_ISCSI;
2171                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2172                 len = roundup2(min(len, 252), 4);
2173                 scsi_ulto2b(len, id->additional_length);
2174                 return (sizeof(*id) + len);
2175         }
2176         case CTL_PORT_SAS:
2177         {
2178                 struct scsi_transportid_sas *id =
2179                     (struct scsi_transportid_sas *)buf;
2180                 if (port->wwpn_iid[iid].wwpn == 0)
2181                         return (0);
2182                 memset(id, 0, sizeof(*id));
2183                 id->format_protocol = SCSI_PROTO_SAS;
2184                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2185                 return (sizeof(*id));
2186         }
2187         default:
2188         {
2189                 struct scsi_transportid_spi *id =
2190                     (struct scsi_transportid_spi *)buf;
2191                 memset(id, 0, sizeof(*id));
2192                 id->format_protocol = SCSI_PROTO_SPI;
2193                 scsi_ulto2b(iid, id->scsi_addr);
2194                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2195                 return (sizeof(*id));
2196         }
2197         }
2198 }
2199
2200 /*
2201  * Serialize a command that went down the "wrong" side, and so was sent to
2202  * this controller for execution.  The logic is a little different than the
2203  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2204  * sent back to the other side, but in the success case, we execute the
2205  * command on this side (XFER mode) or tell the other side to execute it
2206  * (SER_ONLY mode).
2207  */
2208 static void
2209 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2210 {
2211         struct ctl_softc *softc = CTL_SOFTC(ctsio);
2212         struct ctl_port *port = CTL_PORT(ctsio);
2213         union ctl_ha_msg msg_info;
2214         struct ctl_lun *lun;
2215         const struct ctl_cmd_entry *entry;
2216         uint32_t targ_lun;
2217
2218         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2219
2220         /* Make sure that we know about this port. */
2221         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2222                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2223                                          /*retry_count*/ 1);
2224                 goto badjuju;
2225         }
2226
2227         /* Make sure that we know about this LUN. */
2228         mtx_lock(&softc->ctl_lock);
2229         if (targ_lun >= CTL_MAX_LUNS ||
2230             (lun = softc->ctl_luns[targ_lun]) == NULL) {
2231                 mtx_unlock(&softc->ctl_lock);
2232
2233                 /*
2234                  * The other node would not send this request to us unless
2235                  * received announce that we are primary node for this LUN.
2236                  * If this LUN does not exist now, it is probably result of
2237                  * a race, so respond to initiator in the most opaque way.
2238                  */
2239                 ctl_set_busy(ctsio);
2240                 goto badjuju;
2241         }
2242         mtx_lock(&lun->lun_lock);
2243         mtx_unlock(&softc->ctl_lock);
2244
2245         /*
2246          * If the LUN is invalid, pretend that it doesn't exist.
2247          * It will go away as soon as all pending I/Os completed.
2248          */
2249         if (lun->flags & CTL_LUN_DISABLED) {
2250                 mtx_unlock(&lun->lun_lock);
2251                 ctl_set_busy(ctsio);
2252                 goto badjuju;
2253         }
2254
2255         entry = ctl_get_cmd_entry(ctsio, NULL);
2256         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2257                 mtx_unlock(&lun->lun_lock);
2258                 goto badjuju;
2259         }
2260
2261         CTL_LUN(ctsio) = lun;
2262         CTL_BACKEND_LUN(ctsio) = lun->be_lun;
2263
2264         /*
2265          * Every I/O goes into the OOA queue for a
2266          * particular LUN, and stays there until completion.
2267          */
2268 #ifdef CTL_TIME_IO
2269         if (TAILQ_EMPTY(&lun->ooa_queue))
2270                 lun->idle_time += getsbinuptime() - lun->last_busy;
2271 #endif
2272         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2273
2274         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2275                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2276                  ooa_links))) {
2277         case CTL_ACTION_BLOCK:
2278                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2279                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2280                                   blocked_links);
2281                 mtx_unlock(&lun->lun_lock);
2282                 break;
2283         case CTL_ACTION_PASS:
2284         case CTL_ACTION_SKIP:
2285                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2286                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2287                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2288                         mtx_unlock(&lun->lun_lock);
2289                 } else {
2290                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2291                         mtx_unlock(&lun->lun_lock);
2292
2293                         /* send msg back to other side */
2294                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2295                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2296                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2297                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2298                             sizeof(msg_info.hdr), M_WAITOK);
2299                 }
2300                 break;
2301         case CTL_ACTION_OVERLAP:
2302                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2303                 mtx_unlock(&lun->lun_lock);
2304                 ctl_set_overlapped_cmd(ctsio);
2305                 goto badjuju;
2306         case CTL_ACTION_OVERLAP_TAG:
2307                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2308                 mtx_unlock(&lun->lun_lock);
2309                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2310                 goto badjuju;
2311         case CTL_ACTION_ERROR:
2312         default:
2313                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2314                 mtx_unlock(&lun->lun_lock);
2315
2316                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2317                                          /*retry_count*/ 0);
2318 badjuju:
2319                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2320                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2321                 msg_info.hdr.serializing_sc = NULL;
2322                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2323                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2324                     sizeof(msg_info.scsi), M_WAITOK);
2325                 ctl_free_io((union ctl_io *)ctsio);
2326                 break;
2327         }
2328 }
2329
2330 /*
2331  * Returns 0 for success, errno for failure.
2332  */
2333 static void
2334 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2335                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2336 {
2337         union ctl_io *io;
2338
2339         mtx_lock(&lun->lun_lock);
2340         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2341              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2342              ooa_links)) {
2343                 struct ctl_ooa_entry *entry;
2344
2345                 /*
2346                  * If we've got more than we can fit, just count the
2347                  * remaining entries.
2348                  */
2349                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2350                         continue;
2351
2352                 entry = &kern_entries[*cur_fill_num];
2353
2354                 entry->tag_num = io->scsiio.tag_num;
2355                 entry->lun_num = lun->lun;
2356 #ifdef CTL_TIME_IO
2357                 entry->start_bt = io->io_hdr.start_bt;
2358 #endif
2359                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2360                 entry->cdb_len = io->scsiio.cdb_len;
2361                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2362                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2363
2364                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2365                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2366
2367                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2368                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2369
2370                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2371                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2372
2373                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2374                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2375         }
2376         mtx_unlock(&lun->lun_lock);
2377 }
2378
2379 static void *
2380 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2381                  size_t error_str_len)
2382 {
2383         void *kptr;
2384
2385         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2386
2387         if (copyin(user_addr, kptr, len) != 0) {
2388                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2389                          "from user address %p to kernel address %p", len,
2390                          user_addr, kptr);
2391                 free(kptr, M_CTL);
2392                 return (NULL);
2393         }
2394
2395         return (kptr);
2396 }
2397
2398 static void
2399 ctl_free_args(int num_args, struct ctl_be_arg *args)
2400 {
2401         int i;
2402
2403         if (args == NULL)
2404                 return;
2405
2406         for (i = 0; i < num_args; i++) {
2407                 free(args[i].kname, M_CTL);
2408                 free(args[i].kvalue, M_CTL);
2409         }
2410
2411         free(args, M_CTL);
2412 }
2413
2414 static struct ctl_be_arg *
2415 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2416                 char *error_str, size_t error_str_len)
2417 {
2418         struct ctl_be_arg *args;
2419         int i;
2420
2421         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2422                                 error_str, error_str_len);
2423
2424         if (args == NULL)
2425                 goto bailout;
2426
2427         for (i = 0; i < num_args; i++) {
2428                 args[i].kname = NULL;
2429                 args[i].kvalue = NULL;
2430         }
2431
2432         for (i = 0; i < num_args; i++) {
2433                 uint8_t *tmpptr;
2434
2435                 args[i].kname = ctl_copyin_alloc(args[i].name,
2436                         args[i].namelen, error_str, error_str_len);
2437                 if (args[i].kname == NULL)
2438                         goto bailout;
2439
2440                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2441                         snprintf(error_str, error_str_len, "Argument %d "
2442                                  "name is not NUL-terminated", i);
2443                         goto bailout;
2444                 }
2445
2446                 if (args[i].flags & CTL_BEARG_RD) {
2447                         tmpptr = ctl_copyin_alloc(args[i].value,
2448                                 args[i].vallen, error_str, error_str_len);
2449                         if (tmpptr == NULL)
2450                                 goto bailout;
2451                         if ((args[i].flags & CTL_BEARG_ASCII)
2452                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2453                                 snprintf(error_str, error_str_len, "Argument "
2454                                     "%d value is not NUL-terminated", i);
2455                                 free(tmpptr, M_CTL);
2456                                 goto bailout;
2457                         }
2458                         args[i].kvalue = tmpptr;
2459                 } else {
2460                         args[i].kvalue = malloc(args[i].vallen,
2461                             M_CTL, M_WAITOK | M_ZERO);
2462                 }
2463         }
2464
2465         return (args);
2466 bailout:
2467
2468         ctl_free_args(num_args, args);
2469
2470         return (NULL);
2471 }
2472
2473 static void
2474 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2475 {
2476         int i;
2477
2478         for (i = 0; i < num_args; i++) {
2479                 if (args[i].flags & CTL_BEARG_WR)
2480                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2481         }
2482 }
2483
2484 /*
2485  * Escape characters that are illegal or not recommended in XML.
2486  */
2487 int
2488 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2489 {
2490         char *end = str + size;
2491         int retval;
2492
2493         retval = 0;
2494
2495         for (; *str && str < end; str++) {
2496                 switch (*str) {
2497                 case '&':
2498                         retval = sbuf_printf(sb, "&amp;");
2499                         break;
2500                 case '>':
2501                         retval = sbuf_printf(sb, "&gt;");
2502                         break;
2503                 case '<':
2504                         retval = sbuf_printf(sb, "&lt;");
2505                         break;
2506                 default:
2507                         retval = sbuf_putc(sb, *str);
2508                         break;
2509                 }
2510
2511                 if (retval != 0)
2512                         break;
2513
2514         }
2515
2516         return (retval);
2517 }
2518
2519 static void
2520 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2521 {
2522         struct scsi_vpd_id_descriptor *desc;
2523         int i;
2524
2525         if (id == NULL || id->len < 4)
2526                 return;
2527         desc = (struct scsi_vpd_id_descriptor *)id->data;
2528         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2529         case SVPD_ID_TYPE_T10:
2530                 sbuf_printf(sb, "t10.");
2531                 break;
2532         case SVPD_ID_TYPE_EUI64:
2533                 sbuf_printf(sb, "eui.");
2534                 break;
2535         case SVPD_ID_TYPE_NAA:
2536                 sbuf_printf(sb, "naa.");
2537                 break;
2538         case SVPD_ID_TYPE_SCSI_NAME:
2539                 break;
2540         }
2541         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2542         case SVPD_ID_CODESET_BINARY:
2543                 for (i = 0; i < desc->length; i++)
2544                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2545                 break;
2546         case SVPD_ID_CODESET_ASCII:
2547                 sbuf_printf(sb, "%.*s", (int)desc->length,
2548                     (char *)desc->identifier);
2549                 break;
2550         case SVPD_ID_CODESET_UTF8:
2551                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2552                 break;
2553         }
2554 }
2555
2556 static int
2557 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2558           struct thread *td)
2559 {
2560         struct ctl_softc *softc = dev->si_drv1;
2561         struct ctl_port *port;
2562         struct ctl_lun *lun;
2563         int retval;
2564
2565         retval = 0;
2566
2567         switch (cmd) {
2568         case CTL_IO:
2569                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2570                 break;
2571         case CTL_ENABLE_PORT:
2572         case CTL_DISABLE_PORT:
2573         case CTL_SET_PORT_WWNS: {
2574                 struct ctl_port *port;
2575                 struct ctl_port_entry *entry;
2576
2577                 entry = (struct ctl_port_entry *)addr;
2578
2579                 mtx_lock(&softc->ctl_lock);
2580                 STAILQ_FOREACH(port, &softc->port_list, links) {
2581                         int action, done;
2582
2583                         if (port->targ_port < softc->port_min ||
2584                             port->targ_port >= softc->port_max)
2585                                 continue;
2586
2587                         action = 0;
2588                         done = 0;
2589                         if ((entry->port_type == CTL_PORT_NONE)
2590                          && (entry->targ_port == port->targ_port)) {
2591                                 /*
2592                                  * If the user only wants to enable or
2593                                  * disable or set WWNs on a specific port,
2594                                  * do the operation and we're done.
2595                                  */
2596                                 action = 1;
2597                                 done = 1;
2598                         } else if (entry->port_type & port->port_type) {
2599                                 /*
2600                                  * Compare the user's type mask with the
2601                                  * particular frontend type to see if we
2602                                  * have a match.
2603                                  */
2604                                 action = 1;
2605                                 done = 0;
2606
2607                                 /*
2608                                  * Make sure the user isn't trying to set
2609                                  * WWNs on multiple ports at the same time.
2610                                  */
2611                                 if (cmd == CTL_SET_PORT_WWNS) {
2612                                         printf("%s: Can't set WWNs on "
2613                                                "multiple ports\n", __func__);
2614                                         retval = EINVAL;
2615                                         break;
2616                                 }
2617                         }
2618                         if (action == 0)
2619                                 continue;
2620
2621                         /*
2622                          * XXX KDM we have to drop the lock here, because
2623                          * the online/offline operations can potentially
2624                          * block.  We need to reference count the frontends
2625                          * so they can't go away,
2626                          */
2627                         if (cmd == CTL_ENABLE_PORT) {
2628                                 mtx_unlock(&softc->ctl_lock);
2629                                 ctl_port_online(port);
2630                                 mtx_lock(&softc->ctl_lock);
2631                         } else if (cmd == CTL_DISABLE_PORT) {
2632                                 mtx_unlock(&softc->ctl_lock);
2633                                 ctl_port_offline(port);
2634                                 mtx_lock(&softc->ctl_lock);
2635                         } else if (cmd == CTL_SET_PORT_WWNS) {
2636                                 ctl_port_set_wwns(port,
2637                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2638                                     1 : 0, entry->wwnn,
2639                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2640                                     1 : 0, entry->wwpn);
2641                         }
2642                         if (done != 0)
2643                                 break;
2644                 }
2645                 mtx_unlock(&softc->ctl_lock);
2646                 break;
2647         }
2648         case CTL_GET_OOA: {
2649                 struct ctl_ooa *ooa_hdr;
2650                 struct ctl_ooa_entry *entries;
2651                 uint32_t cur_fill_num;
2652
2653                 ooa_hdr = (struct ctl_ooa *)addr;
2654
2655                 if ((ooa_hdr->alloc_len == 0)
2656                  || (ooa_hdr->alloc_num == 0)) {
2657                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2658                                "must be non-zero\n", __func__,
2659                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2660                         retval = EINVAL;
2661                         break;
2662                 }
2663
2664                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2665                     sizeof(struct ctl_ooa_entry))) {
2666                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2667                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2668                                __func__, ooa_hdr->alloc_len,
2669                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2670                         retval = EINVAL;
2671                         break;
2672                 }
2673
2674                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2675                 if (entries == NULL) {
2676                         printf("%s: could not allocate %d bytes for OOA "
2677                                "dump\n", __func__, ooa_hdr->alloc_len);
2678                         retval = ENOMEM;
2679                         break;
2680                 }
2681
2682                 mtx_lock(&softc->ctl_lock);
2683                 if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 &&
2684                     (ooa_hdr->lun_num >= CTL_MAX_LUNS ||
2685                      softc->ctl_luns[ooa_hdr->lun_num] == NULL)) {
2686                         mtx_unlock(&softc->ctl_lock);
2687                         free(entries, M_CTL);
2688                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2689                                __func__, (uintmax_t)ooa_hdr->lun_num);
2690                         retval = EINVAL;
2691                         break;
2692                 }
2693
2694                 cur_fill_num = 0;
2695
2696                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2697                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2698                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2699                                     ooa_hdr, entries);
2700                         }
2701                 } else {
2702                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2703                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2704                             entries);
2705                 }
2706                 mtx_unlock(&softc->ctl_lock);
2707
2708                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2709                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2710                         sizeof(struct ctl_ooa_entry);
2711                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2712                 if (retval != 0) {
2713                         printf("%s: error copying out %d bytes for OOA dump\n",
2714                                __func__, ooa_hdr->fill_len);
2715                 }
2716
2717                 getbinuptime(&ooa_hdr->cur_bt);
2718
2719                 if (cur_fill_num > ooa_hdr->alloc_num) {
2720                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2721                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2722                 } else {
2723                         ooa_hdr->dropped_num = 0;
2724                         ooa_hdr->status = CTL_OOA_OK;
2725                 }
2726
2727                 free(entries, M_CTL);
2728                 break;
2729         }
2730         case CTL_DELAY_IO: {
2731                 struct ctl_io_delay_info *delay_info;
2732
2733                 delay_info = (struct ctl_io_delay_info *)addr;
2734
2735 #ifdef CTL_IO_DELAY
2736                 mtx_lock(&softc->ctl_lock);
2737                 if (delay_info->lun_id >= CTL_MAX_LUNS ||
2738                     (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) {
2739                         mtx_unlock(&softc->ctl_lock);
2740                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2741                         break;
2742                 }
2743                 mtx_lock(&lun->lun_lock);
2744                 mtx_unlock(&softc->ctl_lock);
2745                 delay_info->status = CTL_DELAY_STATUS_OK;
2746                 switch (delay_info->delay_type) {
2747                 case CTL_DELAY_TYPE_CONT:
2748                 case CTL_DELAY_TYPE_ONESHOT:
2749                         break;
2750                 default:
2751                         delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE;
2752                         break;
2753                 }
2754                 switch (delay_info->delay_loc) {
2755                 case CTL_DELAY_LOC_DATAMOVE:
2756                         lun->delay_info.datamove_type = delay_info->delay_type;
2757                         lun->delay_info.datamove_delay = delay_info->delay_secs;
2758                         break;
2759                 case CTL_DELAY_LOC_DONE:
2760                         lun->delay_info.done_type = delay_info->delay_type;
2761                         lun->delay_info.done_delay = delay_info->delay_secs;
2762                         break;
2763                 default:
2764                         delay_info->status = CTL_DELAY_STATUS_INVALID_LOC;
2765                         break;
2766                 }
2767                 mtx_unlock(&lun->lun_lock);
2768 #else
2769                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2770 #endif /* CTL_IO_DELAY */
2771                 break;
2772         }
2773 #ifdef CTL_LEGACY_STATS
2774         case CTL_GETSTATS: {
2775                 struct ctl_stats *stats = (struct ctl_stats *)addr;
2776                 int i;
2777
2778                 /*
2779                  * XXX KDM no locking here.  If the LUN list changes,
2780                  * things can blow up.
2781                  */
2782                 i = 0;
2783                 stats->status = CTL_SS_OK;
2784                 stats->fill_len = 0;
2785                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2786                         if (stats->fill_len + sizeof(lun->legacy_stats) >
2787                             stats->alloc_len) {
2788                                 stats->status = CTL_SS_NEED_MORE_SPACE;
2789                                 break;
2790                         }
2791                         retval = copyout(&lun->legacy_stats, &stats->lun_stats[i++],
2792                                          sizeof(lun->legacy_stats));
2793                         if (retval != 0)
2794                                 break;
2795                         stats->fill_len += sizeof(lun->legacy_stats);
2796                 }
2797                 stats->num_luns = softc->num_luns;
2798                 stats->flags = CTL_STATS_FLAG_NONE;
2799 #ifdef CTL_TIME_IO
2800                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
2801 #endif
2802                 getnanouptime(&stats->timestamp);
2803                 break;
2804         }
2805 #endif /* CTL_LEGACY_STATS */
2806         case CTL_ERROR_INJECT: {
2807                 struct ctl_error_desc *err_desc, *new_err_desc;
2808
2809                 err_desc = (struct ctl_error_desc *)addr;
2810
2811                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2812                                       M_WAITOK | M_ZERO);
2813                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2814
2815                 mtx_lock(&softc->ctl_lock);
2816                 if (err_desc->lun_id >= CTL_MAX_LUNS ||
2817                     (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) {
2818                         mtx_unlock(&softc->ctl_lock);
2819                         free(new_err_desc, M_CTL);
2820                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2821                                __func__, (uintmax_t)err_desc->lun_id);
2822                         retval = EINVAL;
2823                         break;
2824                 }
2825                 mtx_lock(&lun->lun_lock);
2826                 mtx_unlock(&softc->ctl_lock);
2827
2828                 /*
2829                  * We could do some checking here to verify the validity
2830                  * of the request, but given the complexity of error
2831                  * injection requests, the checking logic would be fairly
2832                  * complex.
2833                  *
2834                  * For now, if the request is invalid, it just won't get
2835                  * executed and might get deleted.
2836                  */
2837                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2838
2839                 /*
2840                  * XXX KDM check to make sure the serial number is unique,
2841                  * in case we somehow manage to wrap.  That shouldn't
2842                  * happen for a very long time, but it's the right thing to
2843                  * do.
2844                  */
2845                 new_err_desc->serial = lun->error_serial;
2846                 err_desc->serial = lun->error_serial;
2847                 lun->error_serial++;
2848
2849                 mtx_unlock(&lun->lun_lock);
2850                 break;
2851         }
2852         case CTL_ERROR_INJECT_DELETE: {
2853                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2854                 int delete_done;
2855
2856                 delete_desc = (struct ctl_error_desc *)addr;
2857                 delete_done = 0;
2858
2859                 mtx_lock(&softc->ctl_lock);
2860                 if (delete_desc->lun_id >= CTL_MAX_LUNS ||
2861                     (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) {
2862                         mtx_unlock(&softc->ctl_lock);
2863                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2864                                __func__, (uintmax_t)delete_desc->lun_id);
2865                         retval = EINVAL;
2866                         break;
2867                 }
2868                 mtx_lock(&lun->lun_lock);
2869                 mtx_unlock(&softc->ctl_lock);
2870                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2871                         if (desc->serial != delete_desc->serial)
2872                                 continue;
2873
2874                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2875                                       links);
2876                         free(desc, M_CTL);
2877                         delete_done = 1;
2878                 }
2879                 mtx_unlock(&lun->lun_lock);
2880                 if (delete_done == 0) {
2881                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2882                                "error serial %ju on LUN %u\n", __func__,
2883                                delete_desc->serial, delete_desc->lun_id);
2884                         retval = EINVAL;
2885                         break;
2886                 }
2887                 break;
2888         }
2889         case CTL_DUMP_STRUCTS: {
2890                 int j, k;
2891                 struct ctl_port *port;
2892                 struct ctl_frontend *fe;
2893
2894                 mtx_lock(&softc->ctl_lock);
2895                 printf("CTL Persistent Reservation information start:\n");
2896                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2897                         mtx_lock(&lun->lun_lock);
2898                         if ((lun->flags & CTL_LUN_DISABLED) != 0) {
2899                                 mtx_unlock(&lun->lun_lock);
2900                                 continue;
2901                         }
2902
2903                         for (j = 0; j < CTL_MAX_PORTS; j++) {
2904                                 if (lun->pr_keys[j] == NULL)
2905                                         continue;
2906                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2907                                         if (lun->pr_keys[j][k] == 0)
2908                                                 continue;
2909                                         printf("  LUN %ju port %d iid %d key "
2910                                                "%#jx\n", lun->lun, j, k,
2911                                                (uintmax_t)lun->pr_keys[j][k]);
2912                                 }
2913                         }
2914                         mtx_unlock(&lun->lun_lock);
2915                 }
2916                 printf("CTL Persistent Reservation information end\n");
2917                 printf("CTL Ports:\n");
2918                 STAILQ_FOREACH(port, &softc->port_list, links) {
2919                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2920                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2921                                port->frontend->name, port->port_type,
2922                                port->physical_port, port->virtual_port,
2923                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2924                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2925                                 if (port->wwpn_iid[j].in_use == 0 &&
2926                                     port->wwpn_iid[j].wwpn == 0 &&
2927                                     port->wwpn_iid[j].name == NULL)
2928                                         continue;
2929
2930                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2931                                     j, port->wwpn_iid[j].in_use,
2932                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2933                                     port->wwpn_iid[j].name);
2934                         }
2935                 }
2936                 printf("CTL Port information end\n");
2937                 mtx_unlock(&softc->ctl_lock);
2938                 /*
2939                  * XXX KDM calling this without a lock.  We'd likely want
2940                  * to drop the lock before calling the frontend's dump
2941                  * routine anyway.
2942                  */
2943                 printf("CTL Frontends:\n");
2944                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2945                         printf("  Frontend '%s'\n", fe->name);
2946                         if (fe->fe_dump != NULL)
2947                                 fe->fe_dump();
2948                 }
2949                 printf("CTL Frontend information end\n");
2950                 break;
2951         }
2952         case CTL_LUN_REQ: {
2953                 struct ctl_lun_req *lun_req;
2954                 struct ctl_backend_driver *backend;
2955
2956                 lun_req = (struct ctl_lun_req *)addr;
2957
2958                 backend = ctl_backend_find(lun_req->backend);
2959                 if (backend == NULL) {
2960                         lun_req->status = CTL_LUN_ERROR;
2961                         snprintf(lun_req->error_str,
2962                                  sizeof(lun_req->error_str),
2963                                  "Backend \"%s\" not found.",
2964                                  lun_req->backend);
2965                         break;
2966                 }
2967                 if (lun_req->num_be_args > 0) {
2968                         lun_req->kern_be_args = ctl_copyin_args(
2969                                 lun_req->num_be_args,
2970                                 lun_req->be_args,
2971                                 lun_req->error_str,
2972                                 sizeof(lun_req->error_str));
2973                         if (lun_req->kern_be_args == NULL) {
2974                                 lun_req->status = CTL_LUN_ERROR;
2975                                 break;
2976                         }
2977                 }
2978
2979                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2980
2981                 if (lun_req->num_be_args > 0) {
2982                         ctl_copyout_args(lun_req->num_be_args,
2983                                       lun_req->kern_be_args);
2984                         ctl_free_args(lun_req->num_be_args,
2985                                       lun_req->kern_be_args);
2986                 }
2987                 break;
2988         }
2989         case CTL_LUN_LIST: {
2990                 struct sbuf *sb;
2991                 struct ctl_lun_list *list;
2992                 struct ctl_option *opt;
2993
2994                 list = (struct ctl_lun_list *)addr;
2995
2996                 /*
2997                  * Allocate a fixed length sbuf here, based on the length
2998                  * of the user's buffer.  We could allocate an auto-extending
2999                  * buffer, and then tell the user how much larger our
3000                  * amount of data is than his buffer, but that presents
3001                  * some problems:
3002                  *
3003                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3004                  *     we can't hold a lock while calling them with an
3005                  *     auto-extending buffer.
3006                  *
3007                  * 2.  There is not currently a LUN reference counting
3008                  *     mechanism, outside of outstanding transactions on
3009                  *     the LUN's OOA queue.  So a LUN could go away on us
3010                  *     while we're getting the LUN number, backend-specific
3011                  *     information, etc.  Thus, given the way things
3012                  *     currently work, we need to hold the CTL lock while
3013                  *     grabbing LUN information.
3014                  *
3015                  * So, from the user's standpoint, the best thing to do is
3016                  * allocate what he thinks is a reasonable buffer length,
3017                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3018                  * double the buffer length and try again.  (And repeat
3019                  * that until he succeeds.)
3020                  */
3021                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3022                 if (sb == NULL) {
3023                         list->status = CTL_LUN_LIST_ERROR;
3024                         snprintf(list->error_str, sizeof(list->error_str),
3025                                  "Unable to allocate %d bytes for LUN list",
3026                                  list->alloc_len);
3027                         break;
3028                 }
3029
3030                 sbuf_printf(sb, "<ctllunlist>\n");
3031
3032                 mtx_lock(&softc->ctl_lock);
3033                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3034                         mtx_lock(&lun->lun_lock);
3035                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3036                                              (uintmax_t)lun->lun);
3037
3038                         /*
3039                          * Bail out as soon as we see that we've overfilled
3040                          * the buffer.
3041                          */
3042                         if (retval != 0)
3043                                 break;
3044
3045                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3046                                              "</backend_type>\n",
3047                                              (lun->backend == NULL) ?  "none" :
3048                                              lun->backend->name);
3049
3050                         if (retval != 0)
3051                                 break;
3052
3053                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3054                                              lun->be_lun->lun_type);
3055
3056                         if (retval != 0)
3057                                 break;
3058
3059                         if (lun->backend == NULL) {
3060                                 retval = sbuf_printf(sb, "</lun>\n");
3061                                 if (retval != 0)
3062                                         break;
3063                                 continue;
3064                         }
3065
3066                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3067                                              (lun->be_lun->maxlba > 0) ?
3068                                              lun->be_lun->maxlba + 1 : 0);
3069
3070                         if (retval != 0)
3071                                 break;
3072
3073                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3074                                              lun->be_lun->blocksize);
3075
3076                         if (retval != 0)
3077                                 break;
3078
3079                         retval = sbuf_printf(sb, "\t<serial_number>");
3080
3081                         if (retval != 0)
3082                                 break;
3083
3084                         retval = ctl_sbuf_printf_esc(sb,
3085                             lun->be_lun->serial_num,
3086                             sizeof(lun->be_lun->serial_num));
3087
3088                         if (retval != 0)
3089                                 break;
3090
3091                         retval = sbuf_printf(sb, "</serial_number>\n");
3092
3093                         if (retval != 0)
3094                                 break;
3095
3096                         retval = sbuf_printf(sb, "\t<device_id>");
3097
3098                         if (retval != 0)
3099                                 break;
3100
3101                         retval = ctl_sbuf_printf_esc(sb,
3102                             lun->be_lun->device_id,
3103                             sizeof(lun->be_lun->device_id));
3104
3105                         if (retval != 0)
3106                                 break;
3107
3108                         retval = sbuf_printf(sb, "</device_id>\n");
3109
3110                         if (retval != 0)
3111                                 break;
3112
3113                         if (lun->backend->lun_info != NULL) {
3114                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3115                                 if (retval != 0)
3116                                         break;
3117                         }
3118                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3119                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3120                                     opt->name, opt->value, opt->name);
3121                                 if (retval != 0)
3122                                         break;
3123                         }
3124
3125                         retval = sbuf_printf(sb, "</lun>\n");
3126
3127                         if (retval != 0)
3128                                 break;
3129                         mtx_unlock(&lun->lun_lock);
3130                 }
3131                 if (lun != NULL)
3132                         mtx_unlock(&lun->lun_lock);
3133                 mtx_unlock(&softc->ctl_lock);
3134
3135                 if ((retval != 0)
3136                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3137                         retval = 0;
3138                         sbuf_delete(sb);
3139                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3140                         snprintf(list->error_str, sizeof(list->error_str),
3141                                  "Out of space, %d bytes is too small",
3142                                  list->alloc_len);
3143                         break;
3144                 }
3145
3146                 sbuf_finish(sb);
3147
3148                 retval = copyout(sbuf_data(sb), list->lun_xml,
3149                                  sbuf_len(sb) + 1);
3150
3151                 list->fill_len = sbuf_len(sb) + 1;
3152                 list->status = CTL_LUN_LIST_OK;
3153                 sbuf_delete(sb);
3154                 break;
3155         }
3156         case CTL_ISCSI: {
3157                 struct ctl_iscsi *ci;
3158                 struct ctl_frontend *fe;
3159
3160                 ci = (struct ctl_iscsi *)addr;
3161
3162                 fe = ctl_frontend_find("iscsi");
3163                 if (fe == NULL) {
3164                         ci->status = CTL_ISCSI_ERROR;
3165                         snprintf(ci->error_str, sizeof(ci->error_str),
3166                             "Frontend \"iscsi\" not found.");
3167                         break;
3168                 }
3169
3170                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3171                 break;
3172         }
3173         case CTL_PORT_REQ: {
3174                 struct ctl_req *req;
3175                 struct ctl_frontend *fe;
3176
3177                 req = (struct ctl_req *)addr;
3178
3179                 fe = ctl_frontend_find(req->driver);
3180                 if (fe == NULL) {
3181                         req->status = CTL_LUN_ERROR;
3182                         snprintf(req->error_str, sizeof(req->error_str),
3183                             "Frontend \"%s\" not found.", req->driver);
3184                         break;
3185                 }
3186                 if (req->num_args > 0) {
3187                         req->kern_args = ctl_copyin_args(req->num_args,
3188                             req->args, req->error_str, sizeof(req->error_str));
3189                         if (req->kern_args == NULL) {
3190                                 req->status = CTL_LUN_ERROR;
3191                                 break;
3192                         }
3193                 }
3194
3195                 if (fe->ioctl)
3196                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3197                 else
3198                         retval = ENODEV;
3199
3200                 if (req->num_args > 0) {
3201                         ctl_copyout_args(req->num_args, req->kern_args);
3202                         ctl_free_args(req->num_args, req->kern_args);
3203                 }
3204                 break;
3205         }
3206         case CTL_PORT_LIST: {
3207                 struct sbuf *sb;
3208                 struct ctl_port *port;
3209                 struct ctl_lun_list *list;
3210                 struct ctl_option *opt;
3211                 int j;
3212                 uint32_t plun;
3213
3214                 list = (struct ctl_lun_list *)addr;
3215
3216                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3217                 if (sb == NULL) {
3218                         list->status = CTL_LUN_LIST_ERROR;
3219                         snprintf(list->error_str, sizeof(list->error_str),
3220                                  "Unable to allocate %d bytes for LUN list",
3221                                  list->alloc_len);
3222                         break;
3223                 }
3224
3225                 sbuf_printf(sb, "<ctlportlist>\n");
3226
3227                 mtx_lock(&softc->ctl_lock);
3228                 STAILQ_FOREACH(port, &softc->port_list, links) {
3229                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3230                                              (uintmax_t)port->targ_port);
3231
3232                         /*
3233                          * Bail out as soon as we see that we've overfilled
3234                          * the buffer.
3235                          */
3236                         if (retval != 0)
3237                                 break;
3238
3239                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3240                             "</frontend_type>\n", port->frontend->name);
3241                         if (retval != 0)
3242                                 break;
3243
3244                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3245                                              port->port_type);
3246                         if (retval != 0)
3247                                 break;
3248
3249                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3250                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3251                         if (retval != 0)
3252                                 break;
3253
3254                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3255                             port->port_name);
3256                         if (retval != 0)
3257                                 break;
3258
3259                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3260                             port->physical_port);
3261                         if (retval != 0)
3262                                 break;
3263
3264                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3265                             port->virtual_port);
3266                         if (retval != 0)
3267                                 break;
3268
3269                         if (port->target_devid != NULL) {
3270                                 sbuf_printf(sb, "\t<target>");
3271                                 ctl_id_sbuf(port->target_devid, sb);
3272                                 sbuf_printf(sb, "</target>\n");
3273                         }
3274
3275                         if (port->port_devid != NULL) {
3276                                 sbuf_printf(sb, "\t<port>");
3277                                 ctl_id_sbuf(port->port_devid, sb);
3278                                 sbuf_printf(sb, "</port>\n");
3279                         }
3280
3281                         if (port->port_info != NULL) {
3282                                 retval = port->port_info(port->onoff_arg, sb);
3283                                 if (retval != 0)
3284                                         break;
3285                         }
3286                         STAILQ_FOREACH(opt, &port->options, links) {
3287                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3288                                     opt->name, opt->value, opt->name);
3289                                 if (retval != 0)
3290                                         break;
3291                         }
3292
3293                         if (port->lun_map != NULL) {
3294                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3295                                 for (j = 0; j < port->lun_map_size; j++) {
3296                                         plun = ctl_lun_map_from_port(port, j);
3297                                         if (plun == UINT32_MAX)
3298                                                 continue;
3299                                         sbuf_printf(sb,
3300                                             "\t<lun id=\"%u\">%u</lun>\n",
3301                                             j, plun);
3302                                 }
3303                         }
3304
3305                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3306                                 if (port->wwpn_iid[j].in_use == 0 ||
3307                                     (port->wwpn_iid[j].wwpn == 0 &&
3308                                      port->wwpn_iid[j].name == NULL))
3309                                         continue;
3310
3311                                 if (port->wwpn_iid[j].name != NULL)
3312                                         retval = sbuf_printf(sb,
3313                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3314                                             j, port->wwpn_iid[j].name);
3315                                 else
3316                                         retval = sbuf_printf(sb,
3317                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3318                                             j, port->wwpn_iid[j].wwpn);
3319                                 if (retval != 0)
3320                                         break;
3321                         }
3322                         if (retval != 0)
3323                                 break;
3324
3325                         retval = sbuf_printf(sb, "</targ_port>\n");
3326                         if (retval != 0)
3327                                 break;
3328                 }
3329                 mtx_unlock(&softc->ctl_lock);
3330
3331                 if ((retval != 0)
3332                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3333                         retval = 0;
3334                         sbuf_delete(sb);
3335                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3336                         snprintf(list->error_str, sizeof(list->error_str),
3337                                  "Out of space, %d bytes is too small",
3338                                  list->alloc_len);
3339                         break;
3340                 }
3341
3342                 sbuf_finish(sb);
3343
3344                 retval = copyout(sbuf_data(sb), list->lun_xml,
3345                                  sbuf_len(sb) + 1);
3346
3347                 list->fill_len = sbuf_len(sb) + 1;
3348                 list->status = CTL_LUN_LIST_OK;
3349                 sbuf_delete(sb);
3350                 break;
3351         }
3352         case CTL_LUN_MAP: {
3353                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3354                 struct ctl_port *port;
3355
3356                 mtx_lock(&softc->ctl_lock);
3357                 if (lm->port < softc->port_min ||
3358                     lm->port >= softc->port_max ||
3359                     (port = softc->ctl_ports[lm->port]) == NULL) {
3360                         mtx_unlock(&softc->ctl_lock);
3361                         return (ENXIO);
3362                 }
3363                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3364                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3365                                 if (ctl_lun_map_to_port(port, lun->lun) ==
3366                                     UINT32_MAX)
3367                                         continue;
3368                                 mtx_lock(&lun->lun_lock);
3369                                 ctl_est_ua_port(lun, lm->port, -1,
3370                                     CTL_UA_LUN_CHANGE);
3371                                 mtx_unlock(&lun->lun_lock);
3372                         }
3373                 }
3374                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3375                 if (lm->plun != UINT32_MAX) {
3376                         if (lm->lun == UINT32_MAX)
3377                                 retval = ctl_lun_map_unset(port, lm->plun);
3378                         else if (lm->lun < CTL_MAX_LUNS &&
3379                             softc->ctl_luns[lm->lun] != NULL)
3380                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3381                         else
3382                                 return (ENXIO);
3383                 } else {
3384                         if (lm->lun == UINT32_MAX)
3385                                 retval = ctl_lun_map_deinit(port);
3386                         else
3387                                 retval = ctl_lun_map_init(port);
3388                 }
3389                 if (port->status & CTL_PORT_STATUS_ONLINE)
3390                         ctl_isc_announce_port(port);
3391                 break;
3392         }
3393         case CTL_GET_LUN_STATS: {
3394                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3395                 int i;
3396
3397                 /*
3398                  * XXX KDM no locking here.  If the LUN list changes,
3399                  * things can blow up.
3400                  */
3401                 i = 0;
3402                 stats->status = CTL_SS_OK;
3403                 stats->fill_len = 0;
3404                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3405                         if (lun->lun < stats->first_item)
3406                                 continue;
3407                         if (stats->fill_len + sizeof(lun->stats) >
3408                             stats->alloc_len) {
3409                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3410                                 break;
3411                         }
3412                         retval = copyout(&lun->stats, &stats->stats[i++],
3413                                          sizeof(lun->stats));
3414                         if (retval != 0)
3415                                 break;
3416                         stats->fill_len += sizeof(lun->stats);
3417                 }
3418                 stats->num_items = softc->num_luns;
3419                 stats->flags = CTL_STATS_FLAG_NONE;
3420 #ifdef CTL_TIME_IO
3421                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3422 #endif
3423                 getnanouptime(&stats->timestamp);
3424                 break;
3425         }
3426         case CTL_GET_PORT_STATS: {
3427                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3428                 int i;
3429
3430                 /*
3431                  * XXX KDM no locking here.  If the LUN list changes,
3432                  * things can blow up.
3433                  */
3434                 i = 0;
3435                 stats->status = CTL_SS_OK;
3436                 stats->fill_len = 0;
3437                 STAILQ_FOREACH(port, &softc->port_list, links) {
3438                         if (port->targ_port < stats->first_item)
3439                                 continue;
3440                         if (stats->fill_len + sizeof(port->stats) >
3441                             stats->alloc_len) {
3442                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3443                                 break;
3444                         }
3445                         retval = copyout(&port->stats, &stats->stats[i++],
3446                                          sizeof(port->stats));
3447                         if (retval != 0)
3448                                 break;
3449                         stats->fill_len += sizeof(port->stats);
3450                 }
3451                 stats->num_items = softc->num_ports;
3452                 stats->flags = CTL_STATS_FLAG_NONE;
3453 #ifdef CTL_TIME_IO
3454                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3455 #endif
3456                 getnanouptime(&stats->timestamp);
3457                 break;
3458         }
3459         default: {
3460                 /* XXX KDM should we fix this? */
3461 #if 0
3462                 struct ctl_backend_driver *backend;
3463                 unsigned int type;
3464                 int found;
3465
3466                 found = 0;
3467
3468                 /*
3469                  * We encode the backend type as the ioctl type for backend
3470                  * ioctls.  So parse it out here, and then search for a
3471                  * backend of this type.
3472                  */
3473                 type = _IOC_TYPE(cmd);
3474
3475                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3476                         if (backend->type == type) {
3477                                 found = 1;
3478                                 break;
3479                         }
3480                 }
3481                 if (found == 0) {
3482                         printf("ctl: unknown ioctl command %#lx or backend "
3483                                "%d\n", cmd, type);
3484                         retval = EINVAL;
3485                         break;
3486                 }
3487                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3488 #endif
3489                 retval = ENOTTY;
3490                 break;
3491         }
3492         }
3493         return (retval);
3494 }
3495
3496 uint32_t
3497 ctl_get_initindex(struct ctl_nexus *nexus)
3498 {
3499         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3500 }
3501
3502 int
3503 ctl_lun_map_init(struct ctl_port *port)
3504 {
3505         struct ctl_softc *softc = port->ctl_softc;
3506         struct ctl_lun *lun;
3507         int size = ctl_lun_map_size;
3508         uint32_t i;
3509
3510         if (port->lun_map == NULL || port->lun_map_size < size) {
3511                 port->lun_map_size = 0;
3512                 free(port->lun_map, M_CTL);
3513                 port->lun_map = malloc(size * sizeof(uint32_t),
3514                     M_CTL, M_NOWAIT);
3515         }
3516         if (port->lun_map == NULL)
3517                 return (ENOMEM);
3518         for (i = 0; i < size; i++)
3519                 port->lun_map[i] = UINT32_MAX;
3520         port->lun_map_size = size;
3521         if (port->status & CTL_PORT_STATUS_ONLINE) {
3522                 if (port->lun_disable != NULL) {
3523                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3524                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3525                 }
3526                 ctl_isc_announce_port(port);
3527         }
3528         return (0);
3529 }
3530
3531 int
3532 ctl_lun_map_deinit(struct ctl_port *port)
3533 {
3534         struct ctl_softc *softc = port->ctl_softc;
3535         struct ctl_lun *lun;
3536
3537         if (port->lun_map == NULL)
3538                 return (0);
3539         port->lun_map_size = 0;
3540         free(port->lun_map, M_CTL);
3541         port->lun_map = NULL;
3542         if (port->status & CTL_PORT_STATUS_ONLINE) {
3543                 if (port->lun_enable != NULL) {
3544                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3545                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3546                 }
3547                 ctl_isc_announce_port(port);
3548         }
3549         return (0);
3550 }
3551
3552 int
3553 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3554 {
3555         int status;
3556         uint32_t old;
3557
3558         if (port->lun_map == NULL) {
3559                 status = ctl_lun_map_init(port);
3560                 if (status != 0)
3561                         return (status);
3562         }
3563         if (plun >= port->lun_map_size)
3564                 return (EINVAL);
3565         old = port->lun_map[plun];
3566         port->lun_map[plun] = glun;
3567         if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) {
3568                 if (port->lun_enable != NULL)
3569                         port->lun_enable(port->targ_lun_arg, plun);
3570                 ctl_isc_announce_port(port);
3571         }
3572         return (0);
3573 }
3574
3575 int
3576 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3577 {
3578         uint32_t old;
3579
3580         if (port->lun_map == NULL || plun >= port->lun_map_size)
3581                 return (0);
3582         old = port->lun_map[plun];
3583         port->lun_map[plun] = UINT32_MAX;
3584         if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) {
3585                 if (port->lun_disable != NULL)
3586                         port->lun_disable(port->targ_lun_arg, plun);
3587                 ctl_isc_announce_port(port);
3588         }
3589         return (0);
3590 }
3591
3592 uint32_t
3593 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3594 {
3595
3596         if (port == NULL)
3597                 return (UINT32_MAX);
3598         if (port->lun_map == NULL)
3599                 return (lun_id);
3600         if (lun_id > port->lun_map_size)
3601                 return (UINT32_MAX);
3602         return (port->lun_map[lun_id]);
3603 }
3604
3605 uint32_t
3606 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3607 {
3608         uint32_t i;
3609
3610         if (port == NULL)
3611                 return (UINT32_MAX);
3612         if (port->lun_map == NULL)
3613                 return (lun_id);
3614         for (i = 0; i < port->lun_map_size; i++) {
3615                 if (port->lun_map[i] == lun_id)
3616                         return (i);
3617         }
3618         return (UINT32_MAX);
3619 }
3620
3621 uint32_t
3622 ctl_decode_lun(uint64_t encoded)
3623 {
3624         uint8_t lun[8];
3625         uint32_t result = 0xffffffff;
3626
3627         be64enc(lun, encoded);
3628         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3629         case RPL_LUNDATA_ATYP_PERIPH:
3630                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3631                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3632                         result = lun[1];
3633                 break;
3634         case RPL_LUNDATA_ATYP_FLAT:
3635                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3636                     lun[6] == 0 && lun[7] == 0)
3637                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3638                 break;
3639         case RPL_LUNDATA_ATYP_EXTLUN:
3640                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3641                 case 0x02:
3642                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3643                         case 0x00:
3644                                 result = lun[1];
3645                                 break;
3646                         case 0x10:
3647                                 result = (lun[1] << 16) + (lun[2] << 8) +
3648                                     lun[3];
3649                                 break;
3650                         case 0x20:
3651                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3652                                         result = (lun[2] << 24) +
3653                                             (lun[3] << 16) + (lun[4] << 8) +
3654                                             lun[5];
3655                                 break;
3656                         }
3657                         break;
3658                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3659                         result = 0xffffffff;
3660                         break;
3661                 }
3662                 break;
3663         }
3664         return (result);
3665 }
3666
3667 uint64_t
3668 ctl_encode_lun(uint32_t decoded)
3669 {
3670         uint64_t l = decoded;
3671
3672         if (l <= 0xff)
3673                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3674         if (l <= 0x3fff)
3675                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3676         if (l <= 0xffffff)
3677                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3678                     (l << 32));
3679         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3680 }
3681
3682 int
3683 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3684 {
3685         int i;
3686
3687         for (i = first; i < last; i++) {
3688                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3689                         return (i);
3690         }
3691         return (-1);
3692 }
3693
3694 int
3695 ctl_set_mask(uint32_t *mask, uint32_t bit)
3696 {
3697         uint32_t chunk, piece;
3698
3699         chunk = bit >> 5;
3700         piece = bit % (sizeof(uint32_t) * 8);
3701
3702         if ((mask[chunk] & (1 << piece)) != 0)
3703                 return (-1);
3704         else
3705                 mask[chunk] |= (1 << piece);
3706
3707         return (0);
3708 }
3709
3710 int
3711 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3712 {
3713         uint32_t chunk, piece;
3714
3715         chunk = bit >> 5;
3716         piece = bit % (sizeof(uint32_t) * 8);
3717
3718         if ((mask[chunk] & (1 << piece)) == 0)
3719                 return (-1);
3720         else
3721                 mask[chunk] &= ~(1 << piece);
3722
3723         return (0);
3724 }
3725
3726 int
3727 ctl_is_set(uint32_t *mask, uint32_t bit)
3728 {
3729         uint32_t chunk, piece;
3730
3731         chunk = bit >> 5;
3732         piece = bit % (sizeof(uint32_t) * 8);
3733
3734         if ((mask[chunk] & (1 << piece)) == 0)
3735                 return (0);
3736         else
3737                 return (1);
3738 }
3739
3740 static uint64_t
3741 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3742 {
3743         uint64_t *t;
3744
3745         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3746         if (t == NULL)
3747                 return (0);
3748         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3749 }
3750
3751 static void
3752 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3753 {
3754         uint64_t *t;
3755
3756         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3757         if (t == NULL)
3758                 return;
3759         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3760 }
3761
3762 static void
3763 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3764 {
3765         uint64_t *p;
3766         u_int i;
3767
3768         i = residx/CTL_MAX_INIT_PER_PORT;
3769         if (lun->pr_keys[i] != NULL)
3770                 return;
3771         mtx_unlock(&lun->lun_lock);
3772         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3773             M_WAITOK | M_ZERO);
3774         mtx_lock(&lun->lun_lock);
3775         if (lun->pr_keys[i] == NULL)
3776                 lun->pr_keys[i] = p;
3777         else
3778                 free(p, M_CTL);
3779 }
3780
3781 static void
3782 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3783 {
3784         uint64_t *t;
3785
3786         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3787         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3788         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3789 }
3790
3791 /*
3792  * ctl_softc, pool_name, total_ctl_io are passed in.
3793  * npool is passed out.
3794  */
3795 int
3796 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3797                 uint32_t total_ctl_io, void **npool)
3798 {
3799         struct ctl_io_pool *pool;
3800
3801         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3802                                             M_NOWAIT | M_ZERO);
3803         if (pool == NULL)
3804                 return (ENOMEM);
3805
3806         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3807         pool->ctl_softc = ctl_softc;
3808 #ifdef IO_POOLS
3809         pool->zone = uma_zsecond_create(pool->name, NULL,
3810             NULL, NULL, NULL, ctl_softc->io_zone);
3811         /* uma_prealloc(pool->zone, total_ctl_io); */
3812 #else
3813         pool->zone = ctl_softc->io_zone;
3814 #endif
3815
3816         *npool = pool;
3817         return (0);
3818 }
3819
3820 void
3821 ctl_pool_free(struct ctl_io_pool *pool)
3822 {
3823
3824         if (pool == NULL)
3825                 return;
3826
3827 #ifdef IO_POOLS
3828         uma_zdestroy(pool->zone);
3829 #endif
3830         free(pool, M_CTL);
3831 }
3832
3833 union ctl_io *
3834 ctl_alloc_io(void *pool_ref)
3835 {
3836         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3837         union ctl_io *io;
3838
3839         io = uma_zalloc(pool->zone, M_WAITOK);
3840         if (io != NULL) {
3841                 io->io_hdr.pool = pool_ref;
3842                 CTL_SOFTC(io) = pool->ctl_softc;
3843         }
3844         return (io);
3845 }
3846
3847 union ctl_io *
3848 ctl_alloc_io_nowait(void *pool_ref)
3849 {
3850         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3851         union ctl_io *io;
3852
3853         io = uma_zalloc(pool->zone, M_NOWAIT);
3854         if (io != NULL) {
3855                 io->io_hdr.pool = pool_ref;
3856                 CTL_SOFTC(io) = pool->ctl_softc;
3857         }
3858         return (io);
3859 }
3860
3861 void
3862 ctl_free_io(union ctl_io *io)
3863 {
3864         struct ctl_io_pool *pool;
3865
3866         if (io == NULL)
3867                 return;
3868
3869         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3870         uma_zfree(pool->zone, io);
3871 }
3872
3873 void
3874 ctl_zero_io(union ctl_io *io)
3875 {
3876         struct ctl_io_pool *pool;
3877
3878         if (io == NULL)
3879                 return;
3880
3881         /*
3882          * May need to preserve linked list pointers at some point too.
3883          */
3884         pool = io->io_hdr.pool;
3885         memset(io, 0, sizeof(*io));
3886         io->io_hdr.pool = pool;
3887         CTL_SOFTC(io) = pool->ctl_softc;
3888 }
3889
3890 int
3891 ctl_expand_number(const char *buf, uint64_t *num)
3892 {
3893         char *endptr;
3894         uint64_t number;
3895         unsigned shift;
3896
3897         number = strtoq(buf, &endptr, 0);
3898
3899         switch (tolower((unsigned char)*endptr)) {
3900         case 'e':
3901                 shift = 60;
3902                 break;
3903         case 'p':
3904                 shift = 50;
3905                 break;
3906         case 't':
3907                 shift = 40;
3908                 break;
3909         case 'g':
3910                 shift = 30;
3911                 break;
3912         case 'm':
3913                 shift = 20;
3914                 break;
3915         case 'k':
3916                 shift = 10;
3917                 break;
3918         case 'b':
3919         case '\0': /* No unit. */
3920                 *num = number;
3921                 return (0);
3922         default:
3923                 /* Unrecognized unit. */
3924                 return (-1);
3925         }
3926
3927         if ((number << shift) >> shift != number) {
3928                 /* Overflow */
3929                 return (-1);
3930         }
3931         *num = number << shift;
3932         return (0);
3933 }
3934
3935
3936 /*
3937  * This routine could be used in the future to load default and/or saved
3938  * mode page parameters for a particuar lun.
3939  */
3940 static int
3941 ctl_init_page_index(struct ctl_lun *lun)
3942 {
3943         int i, page_code;
3944         struct ctl_page_index *page_index;
3945         const char *value;
3946         uint64_t ival;
3947
3948         memcpy(&lun->mode_pages.index, page_index_template,
3949                sizeof(page_index_template));
3950
3951         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3952
3953                 page_index = &lun->mode_pages.index[i];
3954                 if (lun->be_lun->lun_type == T_DIRECT &&
3955                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
3956                         continue;
3957                 if (lun->be_lun->lun_type == T_PROCESSOR &&
3958                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
3959                         continue;
3960                 if (lun->be_lun->lun_type == T_CDROM &&
3961                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
3962                         continue;
3963
3964                 page_code = page_index->page_code & SMPH_PC_MASK;
3965                 switch (page_code) {
3966                 case SMS_RW_ERROR_RECOVERY_PAGE: {
3967                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3968                             ("subpage %#x for page %#x is incorrect!",
3969                             page_index->subpage, page_code));
3970                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3971                                &rw_er_page_default,
3972                                sizeof(rw_er_page_default));
3973                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3974                                &rw_er_page_changeable,
3975                                sizeof(rw_er_page_changeable));
3976                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3977                                &rw_er_page_default,
3978                                sizeof(rw_er_page_default));
3979                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3980                                &rw_er_page_default,
3981                                sizeof(rw_er_page_default));
3982                         page_index->page_data =
3983                                 (uint8_t *)lun->mode_pages.rw_er_page;
3984                         break;
3985                 }
3986                 case SMS_FORMAT_DEVICE_PAGE: {
3987                         struct scsi_format_page *format_page;
3988
3989                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3990                             ("subpage %#x for page %#x is incorrect!",
3991                             page_index->subpage, page_code));
3992
3993                         /*
3994                          * Sectors per track are set above.  Bytes per
3995                          * sector need to be set here on a per-LUN basis.
3996                          */
3997                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3998                                &format_page_default,
3999                                sizeof(format_page_default));
4000                         memcpy(&lun->mode_pages.format_page[
4001                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4002                                sizeof(format_page_changeable));
4003                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4004                                &format_page_default,
4005                                sizeof(format_page_default));
4006                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4007                                &format_page_default,
4008                                sizeof(format_page_default));
4009
4010                         format_page = &lun->mode_pages.format_page[
4011                                 CTL_PAGE_CURRENT];
4012                         scsi_ulto2b(lun->be_lun->blocksize,
4013                                     format_page->bytes_per_sector);
4014
4015                         format_page = &lun->mode_pages.format_page[
4016                                 CTL_PAGE_DEFAULT];
4017                         scsi_ulto2b(lun->be_lun->blocksize,
4018                                     format_page->bytes_per_sector);
4019
4020                         format_page = &lun->mode_pages.format_page[
4021                                 CTL_PAGE_SAVED];
4022                         scsi_ulto2b(lun->be_lun->blocksize,
4023                                     format_page->bytes_per_sector);
4024
4025                         page_index->page_data =
4026                                 (uint8_t *)lun->mode_pages.format_page;
4027                         break;
4028                 }
4029                 case SMS_RIGID_DISK_PAGE: {
4030                         struct scsi_rigid_disk_page *rigid_disk_page;
4031                         uint32_t sectors_per_cylinder;
4032                         uint64_t cylinders;
4033 #ifndef __XSCALE__
4034                         int shift;
4035 #endif /* !__XSCALE__ */
4036
4037                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4038                             ("subpage %#x for page %#x is incorrect!",
4039                             page_index->subpage, page_code));
4040
4041                         /*
4042                          * Rotation rate and sectors per track are set
4043                          * above.  We calculate the cylinders here based on
4044                          * capacity.  Due to the number of heads and
4045                          * sectors per track we're using, smaller arrays
4046                          * may turn out to have 0 cylinders.  Linux and
4047                          * FreeBSD don't pay attention to these mode pages
4048                          * to figure out capacity, but Solaris does.  It
4049                          * seems to deal with 0 cylinders just fine, and
4050                          * works out a fake geometry based on the capacity.
4051                          */
4052                         memcpy(&lun->mode_pages.rigid_disk_page[
4053                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4054                                sizeof(rigid_disk_page_default));
4055                         memcpy(&lun->mode_pages.rigid_disk_page[
4056                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4057                                sizeof(rigid_disk_page_changeable));
4058
4059                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4060                                 CTL_DEFAULT_HEADS;
4061
4062                         /*
4063                          * The divide method here will be more accurate,
4064                          * probably, but results in floating point being
4065                          * used in the kernel on i386 (__udivdi3()).  On the
4066                          * XScale, though, __udivdi3() is implemented in
4067                          * software.
4068                          *
4069                          * The shift method for cylinder calculation is
4070                          * accurate if sectors_per_cylinder is a power of
4071                          * 2.  Otherwise it might be slightly off -- you
4072                          * might have a bit of a truncation problem.
4073                          */
4074 #ifdef  __XSCALE__
4075                         cylinders = (lun->be_lun->maxlba + 1) /
4076                                 sectors_per_cylinder;
4077 #else
4078                         for (shift = 31; shift > 0; shift--) {
4079                                 if (sectors_per_cylinder & (1 << shift))
4080                                         break;
4081                         }
4082                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4083 #endif
4084
4085                         /*
4086                          * We've basically got 3 bytes, or 24 bits for the
4087                          * cylinder size in the mode page.  If we're over,
4088                          * just round down to 2^24.
4089                          */
4090                         if (cylinders > 0xffffff)
4091                                 cylinders = 0xffffff;
4092
4093                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4094                                 CTL_PAGE_DEFAULT];
4095                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4096
4097                         if ((value = ctl_get_opt(&lun->be_lun->options,
4098                             "rpm")) != NULL) {
4099                                 scsi_ulto2b(strtol(value, NULL, 0),
4100                                      rigid_disk_page->rotation_rate);
4101                         }
4102
4103                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4104                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4105                                sizeof(rigid_disk_page_default));
4106                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4107                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4108                                sizeof(rigid_disk_page_default));
4109
4110                         page_index->page_data =
4111                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4112                         break;
4113                 }
4114                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4115                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4116                             ("subpage %#x for page %#x is incorrect!",
4117                             page_index->subpage, page_code));
4118                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4119                                &verify_er_page_default,
4120                                sizeof(verify_er_page_default));
4121                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4122                                &verify_er_page_changeable,
4123                                sizeof(verify_er_page_changeable));
4124                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4125                                &verify_er_page_default,
4126                                sizeof(verify_er_page_default));
4127                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4128                                &verify_er_page_default,
4129                                sizeof(verify_er_page_default));
4130                         page_index->page_data =
4131                                 (uint8_t *)lun->mode_pages.verify_er_page;
4132                         break;
4133                 }
4134                 case SMS_CACHING_PAGE: {
4135                         struct scsi_caching_page *caching_page;
4136
4137                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4138                             ("subpage %#x for page %#x is incorrect!",
4139                             page_index->subpage, page_code));
4140                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4141                                &caching_page_default,
4142                                sizeof(caching_page_default));
4143                         memcpy(&lun->mode_pages.caching_page[
4144                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4145                                sizeof(caching_page_changeable));
4146                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4147                                &caching_page_default,
4148                                sizeof(caching_page_default));
4149                         caching_page = &lun->mode_pages.caching_page[
4150                             CTL_PAGE_SAVED];
4151                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
4152                         if (value != NULL && strcmp(value, "off") == 0)
4153                                 caching_page->flags1 &= ~SCP_WCE;
4154                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4155                         if (value != NULL && strcmp(value, "off") == 0)
4156                                 caching_page->flags1 |= SCP_RCD;
4157                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4158                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4159                                sizeof(caching_page_default));
4160                         page_index->page_data =
4161                                 (uint8_t *)lun->mode_pages.caching_page;
4162                         break;
4163                 }
4164                 case SMS_CONTROL_MODE_PAGE: {
4165                         switch (page_index->subpage) {
4166                         case SMS_SUBPAGE_PAGE_0: {
4167                                 struct scsi_control_page *control_page;
4168
4169                                 memcpy(&lun->mode_pages.control_page[
4170                                     CTL_PAGE_DEFAULT],
4171                                        &control_page_default,
4172                                        sizeof(control_page_default));
4173                                 memcpy(&lun->mode_pages.control_page[
4174                                     CTL_PAGE_CHANGEABLE],
4175                                        &control_page_changeable,
4176                                        sizeof(control_page_changeable));
4177                                 memcpy(&lun->mode_pages.control_page[
4178                                     CTL_PAGE_SAVED],
4179                                        &control_page_default,
4180                                        sizeof(control_page_default));
4181                                 control_page = &lun->mode_pages.control_page[
4182                                     CTL_PAGE_SAVED];
4183                                 value = ctl_get_opt(&lun->be_lun->options,
4184                                     "reordering");
4185                                 if (value != NULL &&
4186                                     strcmp(value, "unrestricted") == 0) {
4187                                         control_page->queue_flags &=
4188                                             ~SCP_QUEUE_ALG_MASK;
4189                                         control_page->queue_flags |=
4190                                             SCP_QUEUE_ALG_UNRESTRICTED;
4191                                 }
4192                                 memcpy(&lun->mode_pages.control_page[
4193                                     CTL_PAGE_CURRENT],
4194                                        &lun->mode_pages.control_page[
4195                                     CTL_PAGE_SAVED],
4196                                        sizeof(control_page_default));
4197                                 page_index->page_data =
4198                                     (uint8_t *)lun->mode_pages.control_page;
4199                                 break;
4200                         }
4201                         case 0x01:
4202                                 memcpy(&lun->mode_pages.control_ext_page[
4203                                     CTL_PAGE_DEFAULT],
4204                                        &control_ext_page_default,
4205                                        sizeof(control_ext_page_default));
4206                                 memcpy(&lun->mode_pages.control_ext_page[
4207                                     CTL_PAGE_CHANGEABLE],
4208                                        &control_ext_page_changeable,
4209                                        sizeof(control_ext_page_changeable));
4210                                 memcpy(&lun->mode_pages.control_ext_page[
4211                                     CTL_PAGE_SAVED],
4212                                        &control_ext_page_default,
4213                                        sizeof(control_ext_page_default));
4214                                 memcpy(&lun->mode_pages.control_ext_page[
4215                                     CTL_PAGE_CURRENT],
4216                                        &lun->mode_pages.control_ext_page[
4217                                     CTL_PAGE_SAVED],
4218                                        sizeof(control_ext_page_default));
4219                                 page_index->page_data =
4220                                     (uint8_t *)lun->mode_pages.control_ext_page;
4221                                 break;
4222                         default:
4223                                 panic("subpage %#x for page %#x is incorrect!",
4224                                       page_index->subpage, page_code);
4225                         }
4226                         break;
4227                 }
4228                 case SMS_INFO_EXCEPTIONS_PAGE: {
4229                         switch (page_index->subpage) {
4230                         case SMS_SUBPAGE_PAGE_0:
4231                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4232                                        &ie_page_default,
4233                                        sizeof(ie_page_default));
4234                                 memcpy(&lun->mode_pages.ie_page[
4235                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4236                                        sizeof(ie_page_changeable));
4237                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4238                                        &ie_page_default,
4239                                        sizeof(ie_page_default));
4240                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4241                                        &ie_page_default,
4242                                        sizeof(ie_page_default));
4243                                 page_index->page_data =
4244                                         (uint8_t *)lun->mode_pages.ie_page;
4245                                 break;
4246                         case 0x02: {
4247                                 struct ctl_logical_block_provisioning_page *page;
4248
4249                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4250                                        &lbp_page_default,
4251                                        sizeof(lbp_page_default));
4252                                 memcpy(&lun->mode_pages.lbp_page[
4253                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4254                                        sizeof(lbp_page_changeable));
4255                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4256                                        &lbp_page_default,
4257                                        sizeof(lbp_page_default));
4258                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4259                                 value = ctl_get_opt(&lun->be_lun->options,
4260                                     "avail-threshold");
4261                                 if (value != NULL &&
4262                                     ctl_expand_number(value, &ival) == 0) {
4263                                         page->descr[0].flags |= SLBPPD_ENABLED |
4264                                             SLBPPD_ARMING_DEC;
4265                                         if (lun->be_lun->blocksize)
4266                                                 ival /= lun->be_lun->blocksize;
4267                                         else
4268                                                 ival /= 512;
4269                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4270                                             page->descr[0].count);
4271                                 }
4272                                 value = ctl_get_opt(&lun->be_lun->options,
4273                                     "used-threshold");
4274                                 if (value != NULL &&
4275                                     ctl_expand_number(value, &ival) == 0) {
4276                                         page->descr[1].flags |= SLBPPD_ENABLED |
4277                                             SLBPPD_ARMING_INC;
4278                                         if (lun->be_lun->blocksize)
4279                                                 ival /= lun->be_lun->blocksize;
4280                                         else
4281                                                 ival /= 512;
4282                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4283                                             page->descr[1].count);
4284                                 }
4285                                 value = ctl_get_opt(&lun->be_lun->options,
4286                                     "pool-avail-threshold");
4287                                 if (value != NULL &&
4288                                     ctl_expand_number(value, &ival) == 0) {
4289                                         page->descr[2].flags |= SLBPPD_ENABLED |
4290                                             SLBPPD_ARMING_DEC;
4291                                         if (lun->be_lun->blocksize)
4292                                                 ival /= lun->be_lun->blocksize;
4293                                         else
4294                                                 ival /= 512;
4295                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4296                                             page->descr[2].count);
4297                                 }
4298                                 value = ctl_get_opt(&lun->be_lun->options,
4299                                     "pool-used-threshold");
4300                                 if (value != NULL &&
4301                                     ctl_expand_number(value, &ival) == 0) {
4302                                         page->descr[3].flags |= SLBPPD_ENABLED |
4303                                             SLBPPD_ARMING_INC;
4304                                         if (lun->be_lun->blocksize)
4305                                                 ival /= lun->be_lun->blocksize;
4306                                         else
4307                                                 ival /= 512;
4308                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4309                                             page->descr[3].count);
4310                                 }
4311                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4312                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4313                                        sizeof(lbp_page_default));
4314                                 page_index->page_data =
4315                                         (uint8_t *)lun->mode_pages.lbp_page;
4316                                 break;
4317                         }
4318                         default:
4319                                 panic("subpage %#x for page %#x is incorrect!",
4320                                       page_index->subpage, page_code);
4321                         }
4322                         break;
4323                 }
4324                 case SMS_CDDVD_CAPS_PAGE:{
4325                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4326                             ("subpage %#x for page %#x is incorrect!",
4327                             page_index->subpage, page_code));
4328                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4329                                &cddvd_page_default,
4330                                sizeof(cddvd_page_default));
4331                         memcpy(&lun->mode_pages.cddvd_page[
4332                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4333                                sizeof(cddvd_page_changeable));
4334                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4335                                &cddvd_page_default,
4336                                sizeof(cddvd_page_default));
4337                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4338                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4339                                sizeof(cddvd_page_default));
4340                         page_index->page_data =
4341                                 (uint8_t *)lun->mode_pages.cddvd_page;
4342                         break;
4343                 }
4344                 default:
4345                         panic("invalid page code value %#x", page_code);
4346                 }
4347         }
4348
4349         return (CTL_RETVAL_COMPLETE);
4350 }
4351
4352 static int
4353 ctl_init_log_page_index(struct ctl_lun *lun)
4354 {
4355         struct ctl_page_index *page_index;
4356         int i, j, k, prev;
4357
4358         memcpy(&lun->log_pages.index, log_page_index_template,
4359                sizeof(log_page_index_template));
4360
4361         prev = -1;
4362         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4363
4364                 page_index = &lun->log_pages.index[i];
4365                 if (lun->be_lun->lun_type == T_DIRECT &&
4366                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4367                         continue;
4368                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4369                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4370                         continue;
4371                 if (lun->be_lun->lun_type == T_CDROM &&
4372                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4373                         continue;
4374
4375                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4376                     lun->backend->lun_attr == NULL)
4377                         continue;
4378
4379                 if (page_index->page_code != prev) {
4380                         lun->log_pages.pages_page[j] = page_index->page_code;
4381                         prev = page_index->page_code;
4382                         j++;
4383                 }
4384                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4385                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4386                 k++;
4387         }
4388         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4389         lun->log_pages.index[0].page_len = j;
4390         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4391         lun->log_pages.index[1].page_len = k * 2;
4392         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4393         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4394         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4395         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4396         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.ie_page;
4397         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.ie_page);
4398
4399         return (CTL_RETVAL_COMPLETE);
4400 }
4401
4402 static int
4403 hex2bin(const char *str, uint8_t *buf, int buf_size)
4404 {
4405         int i;
4406         u_char c;
4407
4408         memset(buf, 0, buf_size);
4409         while (isspace(str[0]))
4410                 str++;
4411         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4412                 str += 2;
4413         buf_size *= 2;
4414         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4415                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4416                         str++;
4417                 c = str[i];
4418                 if (isdigit(c))
4419                         c -= '0';
4420                 else if (isalpha(c))
4421                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4422                 else
4423                         break;
4424                 if (c >= 16)
4425                         break;
4426                 if ((i & 1) == 0)
4427                         buf[i / 2] |= (c << 4);
4428                 else
4429                         buf[i / 2] |= c;
4430         }
4431         return ((i + 1) / 2);
4432 }
4433
4434 /*
4435  * LUN allocation.
4436  *
4437  * Requirements:
4438  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4439  *   wants us to allocate the LUN and he can block.
4440  * - ctl_softc is always set
4441  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4442  *
4443  * Returns 0 for success, non-zero (errno) for failure.
4444  */
4445 static int
4446 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4447               struct ctl_be_lun *const be_lun)
4448 {
4449         struct ctl_lun *nlun, *lun;
4450         struct scsi_vpd_id_descriptor *desc;
4451         struct scsi_vpd_id_t10 *t10id;
4452         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4453         int lun_number, lun_malloced;
4454         int devidlen, idlen1, idlen2 = 0, len;
4455
4456         if (be_lun == NULL)
4457                 return (EINVAL);
4458
4459         /*
4460          * We currently only support Direct Access or Processor LUN types.
4461          */
4462         switch (be_lun->lun_type) {
4463         case T_DIRECT:
4464         case T_PROCESSOR:
4465         case T_CDROM:
4466                 break;
4467         case T_SEQUENTIAL:
4468         case T_CHANGER:
4469         default:
4470                 be_lun->lun_config_status(be_lun->be_lun,
4471                                           CTL_LUN_CONFIG_FAILURE);
4472                 break;
4473         }
4474         if (ctl_lun == NULL) {
4475                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4476                 lun_malloced = 1;
4477         } else {
4478                 lun_malloced = 0;
4479                 lun = ctl_lun;
4480         }
4481
4482         memset(lun, 0, sizeof(*lun));
4483         if (lun_malloced)
4484                 lun->flags = CTL_LUN_MALLOCED;
4485
4486         /* Generate LUN ID. */
4487         devidlen = max(CTL_DEVID_MIN_LEN,
4488             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4489         idlen1 = sizeof(*t10id) + devidlen;
4490         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4491         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4492         if (scsiname != NULL) {
4493                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4494                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4495         }
4496         eui = ctl_get_opt(&be_lun->options, "eui");
4497         if (eui != NULL) {
4498                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4499         }
4500         naa = ctl_get_opt(&be_lun->options, "naa");
4501         if (naa != NULL) {
4502                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4503         }
4504         uuid = ctl_get_opt(&be_lun->options, "uuid");
4505         if (uuid != NULL) {
4506                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4507         }
4508         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4509             M_CTL, M_WAITOK | M_ZERO);
4510         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4511         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4512         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4513         desc->length = idlen1;
4514         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4515         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4516         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4517                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4518         } else {
4519                 strncpy(t10id->vendor, vendor,
4520                     min(sizeof(t10id->vendor), strlen(vendor)));
4521         }
4522         strncpy((char *)t10id->vendor_spec_id,
4523             (char *)be_lun->device_id, devidlen);
4524         if (scsiname != NULL) {
4525                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4526                     desc->length);
4527                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4528                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4529                     SVPD_ID_TYPE_SCSI_NAME;
4530                 desc->length = idlen2;
4531                 strlcpy(desc->identifier, scsiname, idlen2);
4532         }
4533         if (eui != NULL) {
4534                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4535                     desc->length);
4536                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4537                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4538                     SVPD_ID_TYPE_EUI64;
4539                 desc->length = hex2bin(eui, desc->identifier, 16);
4540                 desc->length = desc->length > 12 ? 16 :
4541                     (desc->length > 8 ? 12 : 8);
4542                 len -= 16 - desc->length;
4543         }
4544         if (naa != NULL) {
4545                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4546                     desc->length);
4547                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4548                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4549                     SVPD_ID_TYPE_NAA;
4550                 desc->length = hex2bin(naa, desc->identifier, 16);
4551                 desc->length = desc->length > 8 ? 16 : 8;
4552                 len -= 16 - desc->length;
4553         }
4554         if (uuid != NULL) {
4555                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4556                     desc->length);
4557                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4558                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4559                     SVPD_ID_TYPE_UUID;
4560                 desc->identifier[0] = 0x10;
4561                 hex2bin(uuid, &desc->identifier[2], 16);
4562                 desc->length = 18;
4563         }
4564         lun->lun_devid->len = len;
4565
4566         mtx_lock(&ctl_softc->ctl_lock);
4567         /*
4568          * See if the caller requested a particular LUN number.  If so, see
4569          * if it is available.  Otherwise, allocate the first available LUN.
4570          */
4571         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4572                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4573                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4574                         mtx_unlock(&ctl_softc->ctl_lock);
4575                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4576                                 printf("ctl: requested LUN ID %d is higher "
4577                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4578                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4579                         } else {
4580                                 /*
4581                                  * XXX KDM return an error, or just assign
4582                                  * another LUN ID in this case??
4583                                  */
4584                                 printf("ctl: requested LUN ID %d is already "
4585                                        "in use\n", be_lun->req_lun_id);
4586                         }
4587 fail:
4588                         free(lun->lun_devid, M_CTL);
4589                         if (lun->flags & CTL_LUN_MALLOCED)
4590                                 free(lun, M_CTL);
4591                         be_lun->lun_config_status(be_lun->be_lun,
4592                                                   CTL_LUN_CONFIG_FAILURE);
4593                         return (ENOSPC);
4594                 }
4595                 lun_number = be_lun->req_lun_id;
4596         } else {
4597                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4598                 if (lun_number == -1) {
4599                         mtx_unlock(&ctl_softc->ctl_lock);
4600                         printf("ctl: can't allocate LUN, out of LUNs\n");
4601                         goto fail;
4602                 }
4603         }
4604         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4605         mtx_unlock(&ctl_softc->ctl_lock);
4606
4607         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4608         lun->lun = lun_number;
4609         lun->be_lun = be_lun;
4610         /*
4611          * The processor LUN is always enabled.  Disk LUNs come on line
4612          * disabled, and must be enabled by the backend.
4613          */
4614         lun->flags |= CTL_LUN_DISABLED;
4615         lun->backend = be_lun->be;
4616         be_lun->ctl_lun = lun;
4617         be_lun->lun_id = lun_number;
4618         atomic_add_int(&be_lun->be->num_luns, 1);
4619         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4620                 lun->flags |= CTL_LUN_EJECTED;
4621         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4622                 lun->flags |= CTL_LUN_NO_MEDIA;
4623         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4624                 lun->flags |= CTL_LUN_STOPPED;
4625
4626         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4627                 lun->flags |= CTL_LUN_PRIMARY_SC;
4628
4629         value = ctl_get_opt(&be_lun->options, "removable");
4630         if (value != NULL) {
4631                 if (strcmp(value, "on") == 0)
4632                         lun->flags |= CTL_LUN_REMOVABLE;
4633         } else if (be_lun->lun_type == T_CDROM)
4634                 lun->flags |= CTL_LUN_REMOVABLE;
4635
4636         lun->ctl_softc = ctl_softc;
4637 #ifdef CTL_TIME_IO
4638         lun->last_busy = getsbinuptime();
4639 #endif
4640         TAILQ_INIT(&lun->ooa_queue);
4641         TAILQ_INIT(&lun->blocked_queue);
4642         STAILQ_INIT(&lun->error_list);
4643         lun->ie_reported = 1;
4644         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4645         ctl_tpc_lun_init(lun);
4646         if (lun->flags & CTL_LUN_REMOVABLE) {
4647                 lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4,
4648                     M_CTL, M_WAITOK);
4649         }
4650
4651         /*
4652          * Initialize the mode and log page index.
4653          */
4654         ctl_init_page_index(lun);
4655         ctl_init_log_page_index(lun);
4656
4657         /* Setup statistics gathering */
4658 #ifdef CTL_LEGACY_STATS
4659         lun->legacy_stats.device_type = be_lun->lun_type;
4660         lun->legacy_stats.lun_number = lun_number;
4661         lun->legacy_stats.blocksize = be_lun->blocksize;
4662         if (be_lun->blocksize == 0)
4663                 lun->legacy_stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4664         for (len = 0; len < CTL_MAX_PORTS; len++)
4665                 lun->legacy_stats.ports[len].targ_port = len;
4666 #endif /* CTL_LEGACY_STATS */
4667         lun->stats.item = lun_number;
4668
4669         /*
4670          * Now, before we insert this lun on the lun list, set the lun
4671          * inventory changed UA for all other luns.
4672          */
4673         mtx_lock(&ctl_softc->ctl_lock);
4674         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4675                 mtx_lock(&nlun->lun_lock);
4676                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4677                 mtx_unlock(&nlun->lun_lock);
4678         }
4679         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4680         ctl_softc->ctl_luns[lun_number] = lun;
4681         ctl_softc->num_luns++;
4682         mtx_unlock(&ctl_softc->ctl_lock);
4683
4684         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4685         return (0);
4686 }
4687
4688 /*
4689  * Delete a LUN.
4690  * Assumptions:
4691  * - LUN has already been marked invalid and any pending I/O has been taken
4692  *   care of.
4693  */
4694 static int
4695 ctl_free_lun(struct ctl_lun *lun)
4696 {
4697         struct ctl_softc *softc = lun->ctl_softc;
4698         struct ctl_lun *nlun;
4699         int i;
4700
4701         mtx_assert(&softc->ctl_lock, MA_OWNED);
4702
4703         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4704
4705         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4706
4707         softc->ctl_luns[lun->lun] = NULL;
4708
4709         if (!TAILQ_EMPTY(&lun->ooa_queue))
4710                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4711
4712         softc->num_luns--;
4713
4714         /*
4715          * Tell the backend to free resources, if this LUN has a backend.
4716          */
4717         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4718         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4719
4720         lun->ie_reportcnt = UINT32_MAX;
4721         callout_drain(&lun->ie_callout);
4722
4723         ctl_tpc_lun_shutdown(lun);
4724         mtx_destroy(&lun->lun_lock);
4725         free(lun->lun_devid, M_CTL);
4726         for (i = 0; i < CTL_MAX_PORTS; i++)
4727                 free(lun->pending_ua[i], M_CTL);
4728         for (i = 0; i < CTL_MAX_PORTS; i++)
4729                 free(lun->pr_keys[i], M_CTL);
4730         free(lun->write_buffer, M_CTL);
4731         free(lun->prevent, M_CTL);
4732         if (lun->flags & CTL_LUN_MALLOCED)
4733                 free(lun, M_CTL);
4734
4735         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4736                 mtx_lock(&nlun->lun_lock);
4737                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4738                 mtx_unlock(&nlun->lun_lock);
4739         }
4740
4741         return (0);
4742 }
4743
4744 static void
4745 ctl_create_lun(struct ctl_be_lun *be_lun)
4746 {
4747
4748         /*
4749          * ctl_alloc_lun() should handle all potential failure cases.
4750          */
4751         ctl_alloc_lun(control_softc, NULL, be_lun);
4752 }
4753
4754 int
4755 ctl_add_lun(struct ctl_be_lun *be_lun)
4756 {
4757         struct ctl_softc *softc = control_softc;
4758
4759         mtx_lock(&softc->ctl_lock);
4760         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4761         mtx_unlock(&softc->ctl_lock);
4762         wakeup(&softc->pending_lun_queue);
4763
4764         return (0);
4765 }
4766
4767 int
4768 ctl_enable_lun(struct ctl_be_lun *be_lun)
4769 {
4770         struct ctl_softc *softc;
4771         struct ctl_port *port, *nport;
4772         struct ctl_lun *lun;
4773         int retval;
4774
4775         lun = (struct ctl_lun *)be_lun->ctl_lun;
4776         softc = lun->ctl_softc;
4777
4778         mtx_lock(&softc->ctl_lock);
4779         mtx_lock(&lun->lun_lock);
4780         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4781                 /*
4782                  * eh?  Why did we get called if the LUN is already
4783                  * enabled?
4784                  */
4785                 mtx_unlock(&lun->lun_lock);
4786                 mtx_unlock(&softc->ctl_lock);
4787                 return (0);
4788         }
4789         lun->flags &= ~CTL_LUN_DISABLED;
4790         mtx_unlock(&lun->lun_lock);
4791
4792         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4793                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4794                     port->lun_map != NULL || port->lun_enable == NULL)
4795                         continue;
4796
4797                 /*
4798                  * Drop the lock while we call the FETD's enable routine.
4799                  * This can lead to a callback into CTL (at least in the
4800                  * case of the internal initiator frontend.
4801                  */
4802                 mtx_unlock(&softc->ctl_lock);
4803                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4804                 mtx_lock(&softc->ctl_lock);
4805                 if (retval != 0) {
4806                         printf("%s: FETD %s port %d returned error "
4807                                "%d for lun_enable on lun %jd\n",
4808                                __func__, port->port_name, port->targ_port,
4809                                retval, (intmax_t)lun->lun);
4810                 }
4811         }
4812
4813         mtx_unlock(&softc->ctl_lock);
4814         ctl_isc_announce_lun(lun);
4815
4816         return (0);
4817 }
4818
4819 int
4820 ctl_disable_lun(struct ctl_be_lun *be_lun)
4821 {
4822         struct ctl_softc *softc;
4823         struct ctl_port *port;
4824         struct ctl_lun *lun;
4825         int retval;
4826
4827         lun = (struct ctl_lun *)be_lun->ctl_lun;
4828         softc = lun->ctl_softc;
4829
4830         mtx_lock(&softc->ctl_lock);
4831         mtx_lock(&lun->lun_lock);
4832         if (lun->flags & CTL_LUN_DISABLED) {
4833                 mtx_unlock(&lun->lun_lock);
4834                 mtx_unlock(&softc->ctl_lock);
4835                 return (0);
4836         }
4837         lun->flags |= CTL_LUN_DISABLED;
4838         mtx_unlock(&lun->lun_lock);
4839
4840         STAILQ_FOREACH(port, &softc->port_list, links) {
4841                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4842                     port->lun_map != NULL || port->lun_disable == NULL)
4843                         continue;
4844
4845                 /*
4846                  * Drop the lock before we call the frontend's disable
4847                  * routine, to avoid lock order reversals.
4848                  *
4849                  * XXX KDM what happens if the frontend list changes while
4850                  * we're traversing it?  It's unlikely, but should be handled.
4851                  */
4852                 mtx_unlock(&softc->ctl_lock);
4853                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4854                 mtx_lock(&softc->ctl_lock);
4855                 if (retval != 0) {
4856                         printf("%s: FETD %s port %d returned error "
4857                                "%d for lun_disable on lun %jd\n",
4858                                __func__, port->port_name, port->targ_port,
4859                                retval, (intmax_t)lun->lun);
4860                 }
4861         }
4862
4863         mtx_unlock(&softc->ctl_lock);
4864         ctl_isc_announce_lun(lun);
4865
4866         return (0);
4867 }
4868
4869 int
4870 ctl_start_lun(struct ctl_be_lun *be_lun)
4871 {
4872         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4873
4874         mtx_lock(&lun->lun_lock);
4875         lun->flags &= ~CTL_LUN_STOPPED;
4876         mtx_unlock(&lun->lun_lock);
4877         return (0);
4878 }
4879
4880 int
4881 ctl_stop_lun(struct ctl_be_lun *be_lun)
4882 {
4883         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4884
4885         mtx_lock(&lun->lun_lock);
4886         lun->flags |= CTL_LUN_STOPPED;
4887         mtx_unlock(&lun->lun_lock);
4888         return (0);
4889 }
4890
4891 int
4892 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4893 {
4894         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4895
4896         mtx_lock(&lun->lun_lock);
4897         lun->flags |= CTL_LUN_NO_MEDIA;
4898         mtx_unlock(&lun->lun_lock);
4899         return (0);
4900 }
4901
4902 int
4903 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4904 {
4905         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4906         union ctl_ha_msg msg;
4907
4908         mtx_lock(&lun->lun_lock);
4909         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4910         if (lun->flags & CTL_LUN_REMOVABLE)
4911                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4912         mtx_unlock(&lun->lun_lock);
4913         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4914             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4915                 bzero(&msg.ua, sizeof(msg.ua));
4916                 msg.hdr.msg_type = CTL_MSG_UA;
4917                 msg.hdr.nexus.initid = -1;
4918                 msg.hdr.nexus.targ_port = -1;
4919                 msg.hdr.nexus.targ_lun = lun->lun;
4920                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4921                 msg.ua.ua_all = 1;
4922                 msg.ua.ua_set = 1;
4923                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4924                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4925                     M_WAITOK);
4926         }
4927         return (0);
4928 }
4929
4930 int
4931 ctl_lun_ejected(struct ctl_be_lun *be_lun)
4932 {
4933         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4934
4935         mtx_lock(&lun->lun_lock);
4936         lun->flags |= CTL_LUN_EJECTED;
4937         mtx_unlock(&lun->lun_lock);
4938         return (0);
4939 }
4940
4941 int
4942 ctl_lun_primary(struct ctl_be_lun *be_lun)
4943 {
4944         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4945
4946         mtx_lock(&lun->lun_lock);
4947         lun->flags |= CTL_LUN_PRIMARY_SC;
4948         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4949         mtx_unlock(&lun->lun_lock);
4950         ctl_isc_announce_lun(lun);
4951         return (0);
4952 }
4953
4954 int
4955 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4956 {
4957         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4958
4959         mtx_lock(&lun->lun_lock);
4960         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4961         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4962         mtx_unlock(&lun->lun_lock);
4963         ctl_isc_announce_lun(lun);
4964         return (0);
4965 }
4966
4967 int
4968 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4969 {
4970         struct ctl_softc *softc;
4971         struct ctl_lun *lun;
4972
4973         lun = (struct ctl_lun *)be_lun->ctl_lun;
4974         softc = lun->ctl_softc;
4975
4976         mtx_lock(&lun->lun_lock);
4977
4978         /*
4979          * The LUN needs to be disabled before it can be marked invalid.
4980          */
4981         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4982                 mtx_unlock(&lun->lun_lock);
4983                 return (-1);
4984         }
4985         /*
4986          * Mark the LUN invalid.
4987          */
4988         lun->flags |= CTL_LUN_INVALID;
4989
4990         /*
4991          * If there is nothing in the OOA queue, go ahead and free the LUN.
4992          * If we have something in the OOA queue, we'll free it when the
4993          * last I/O completes.
4994          */
4995         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4996                 mtx_unlock(&lun->lun_lock);
4997                 mtx_lock(&softc->ctl_lock);
4998                 ctl_free_lun(lun);
4999                 mtx_unlock(&softc->ctl_lock);
5000         } else
5001                 mtx_unlock(&lun->lun_lock);
5002
5003         return (0);
5004 }
5005
5006 void
5007 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5008 {
5009         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5010         union ctl_ha_msg msg;
5011
5012         mtx_lock(&lun->lun_lock);
5013         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
5014         mtx_unlock(&lun->lun_lock);
5015         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5016                 /* Send msg to other side. */
5017                 bzero(&msg.ua, sizeof(msg.ua));
5018                 msg.hdr.msg_type = CTL_MSG_UA;
5019                 msg.hdr.nexus.initid = -1;
5020                 msg.hdr.nexus.targ_port = -1;
5021                 msg.hdr.nexus.targ_lun = lun->lun;
5022                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5023                 msg.ua.ua_all = 1;
5024                 msg.ua.ua_set = 1;
5025                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
5026                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5027                     M_WAITOK);
5028         }
5029 }
5030
5031 /*
5032  * Backend "memory move is complete" callback for requests that never
5033  * make it down to say RAIDCore's configuration code.
5034  */
5035 int
5036 ctl_config_move_done(union ctl_io *io)
5037 {
5038         int retval;
5039
5040         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5041         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5042             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5043
5044         if ((io->io_hdr.port_status != 0) &&
5045             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5046              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5047                 /*
5048                  * For hardware error sense keys, the sense key
5049                  * specific value is defined to be a retry count,
5050                  * but we use it to pass back an internal FETD
5051                  * error code.  XXX KDM  Hopefully the FETD is only
5052                  * using 16 bits for an error code, since that's
5053                  * all the space we have in the sks field.
5054                  */
5055                 ctl_set_internal_failure(&io->scsiio,
5056                                          /*sks_valid*/ 1,
5057                                          /*retry_count*/
5058                                          io->io_hdr.port_status);
5059         }
5060
5061         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5062                 ctl_data_print(io);
5063         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5064             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5065              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5066             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5067                 /*
5068                  * XXX KDM just assuming a single pointer here, and not a
5069                  * S/G list.  If we start using S/G lists for config data,
5070                  * we'll need to know how to clean them up here as well.
5071                  */
5072                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5073                         free(io->scsiio.kern_data_ptr, M_CTL);
5074                 ctl_done(io);
5075                 retval = CTL_RETVAL_COMPLETE;
5076         } else {
5077                 /*
5078                  * XXX KDM now we need to continue data movement.  Some
5079                  * options:
5080                  * - call ctl_scsiio() again?  We don't do this for data
5081                  *   writes, because for those at least we know ahead of
5082                  *   time where the write will go and how long it is.  For
5083                  *   config writes, though, that information is largely
5084                  *   contained within the write itself, thus we need to
5085                  *   parse out the data again.
5086                  *
5087                  * - Call some other function once the data is in?
5088                  */
5089
5090                 /*
5091                  * XXX KDM call ctl_scsiio() again for now, and check flag
5092                  * bits to see whether we're allocated or not.
5093                  */
5094                 retval = ctl_scsiio(&io->scsiio);
5095         }
5096         return (retval);
5097 }
5098
5099 /*
5100  * This gets called by a backend driver when it is done with a
5101  * data_submit method.
5102  */
5103 void
5104 ctl_data_submit_done(union ctl_io *io)
5105 {
5106         /*
5107          * If the IO_CONT flag is set, we need to call the supplied
5108          * function to continue processing the I/O, instead of completing
5109          * the I/O just yet.
5110          *
5111          * If there is an error, though, we don't want to keep processing.
5112          * Instead, just send status back to the initiator.
5113          */
5114         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5115             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5116             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5117              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5118                 io->scsiio.io_cont(io);
5119                 return;
5120         }
5121         ctl_done(io);
5122 }
5123
5124 /*
5125  * This gets called by a backend driver when it is done with a
5126  * configuration write.
5127  */
5128 void
5129 ctl_config_write_done(union ctl_io *io)
5130 {
5131         uint8_t *buf;
5132
5133         /*
5134          * If the IO_CONT flag is set, we need to call the supplied
5135          * function to continue processing the I/O, instead of completing
5136          * the I/O just yet.
5137          *
5138          * If there is an error, though, we don't want to keep processing.
5139          * Instead, just send status back to the initiator.
5140          */
5141         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5142             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5143             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5144              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5145                 io->scsiio.io_cont(io);
5146                 return;
5147         }
5148         /*
5149          * Since a configuration write can be done for commands that actually
5150          * have data allocated, like write buffer, and commands that have
5151          * no data, like start/stop unit, we need to check here.
5152          */
5153         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5154                 buf = io->scsiio.kern_data_ptr;
5155         else
5156                 buf = NULL;
5157         ctl_done(io);
5158         if (buf)
5159                 free(buf, M_CTL);
5160 }
5161
5162 void
5163 ctl_config_read_done(union ctl_io *io)
5164 {
5165         uint8_t *buf;
5166
5167         /*
5168          * If there is some error -- we are done, skip data transfer.
5169          */
5170         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5171             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5172              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5173                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5174                         buf = io->scsiio.kern_data_ptr;
5175                 else
5176                         buf = NULL;
5177                 ctl_done(io);
5178                 if (buf)
5179                         free(buf, M_CTL);
5180                 return;
5181         }
5182
5183         /*
5184          * If the IO_CONT flag is set, we need to call the supplied
5185          * function to continue processing the I/O, instead of completing
5186          * the I/O just yet.
5187          */
5188         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5189                 io->scsiio.io_cont(io);
5190                 return;
5191         }
5192
5193         ctl_datamove(io);
5194 }
5195
5196 /*
5197  * SCSI release command.
5198  */
5199 int
5200 ctl_scsi_release(struct ctl_scsiio *ctsio)
5201 {
5202         struct ctl_lun *lun = CTL_LUN(ctsio);
5203         uint32_t residx;
5204
5205         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5206
5207         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5208
5209         /*
5210          * XXX KDM right now, we only support LUN reservation.  We don't
5211          * support 3rd party reservations, or extent reservations, which
5212          * might actually need the parameter list.  If we've gotten this
5213          * far, we've got a LUN reservation.  Anything else got kicked out
5214          * above.  So, according to SPC, ignore the length.
5215          */
5216
5217         mtx_lock(&lun->lun_lock);
5218
5219         /*
5220          * According to SPC, it is not an error for an intiator to attempt
5221          * to release a reservation on a LUN that isn't reserved, or that
5222          * is reserved by another initiator.  The reservation can only be
5223          * released, though, by the initiator who made it or by one of
5224          * several reset type events.
5225          */
5226         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5227                         lun->flags &= ~CTL_LUN_RESERVED;
5228
5229         mtx_unlock(&lun->lun_lock);
5230
5231         ctl_set_success(ctsio);
5232         ctl_done((union ctl_io *)ctsio);
5233         return (CTL_RETVAL_COMPLETE);
5234 }
5235
5236 int
5237 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5238 {
5239         struct ctl_lun *lun = CTL_LUN(ctsio);
5240         uint32_t residx;
5241
5242         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5243
5244         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5245
5246         /*
5247          * XXX KDM right now, we only support LUN reservation.  We don't
5248          * support 3rd party reservations, or extent reservations, which
5249          * might actually need the parameter list.  If we've gotten this
5250          * far, we've got a LUN reservation.  Anything else got kicked out
5251          * above.  So, according to SPC, ignore the length.
5252          */
5253
5254         mtx_lock(&lun->lun_lock);
5255         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5256                 ctl_set_reservation_conflict(ctsio);
5257                 goto bailout;
5258         }
5259
5260         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5261         if (lun->flags & CTL_LUN_PR_RESERVED) {
5262                 ctl_set_success(ctsio);
5263                 goto bailout;
5264         }
5265
5266         lun->flags |= CTL_LUN_RESERVED;
5267         lun->res_idx = residx;
5268         ctl_set_success(ctsio);
5269
5270 bailout:
5271         mtx_unlock(&lun->lun_lock);
5272         ctl_done((union ctl_io *)ctsio);
5273         return (CTL_RETVAL_COMPLETE);
5274 }
5275
5276 int
5277 ctl_start_stop(struct ctl_scsiio *ctsio)
5278 {
5279         struct ctl_lun *lun = CTL_LUN(ctsio);
5280         struct scsi_start_stop_unit *cdb;
5281         int retval;
5282
5283         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5284
5285         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5286
5287         if ((cdb->how & SSS_PC_MASK) == 0) {
5288                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5289                     (cdb->how & SSS_START) == 0) {
5290                         uint32_t residx;
5291
5292                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5293                         if (ctl_get_prkey(lun, residx) == 0 ||
5294                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5295
5296                                 ctl_set_reservation_conflict(ctsio);
5297                                 ctl_done((union ctl_io *)ctsio);
5298                                 return (CTL_RETVAL_COMPLETE);
5299                         }
5300                 }
5301
5302                 if ((cdb->how & SSS_LOEJ) &&
5303                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5304                         ctl_set_invalid_field(ctsio,
5305                                               /*sks_valid*/ 1,
5306                                               /*command*/ 1,
5307                                               /*field*/ 4,
5308                                               /*bit_valid*/ 1,
5309                                               /*bit*/ 1);
5310                         ctl_done((union ctl_io *)ctsio);
5311                         return (CTL_RETVAL_COMPLETE);
5312                 }
5313
5314                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5315                     lun->prevent_count > 0) {
5316                         /* "Medium removal prevented" */
5317                         ctl_set_sense(ctsio, /*current_error*/ 1,
5318                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5319                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5320                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5321                         ctl_done((union ctl_io *)ctsio);
5322                         return (CTL_RETVAL_COMPLETE);
5323                 }
5324         }
5325
5326         retval = lun->backend->config_write((union ctl_io *)ctsio);
5327         return (retval);
5328 }
5329
5330 int
5331 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5332 {
5333         struct ctl_lun *lun = CTL_LUN(ctsio);
5334         struct scsi_prevent *cdb;
5335         int retval;
5336         uint32_t initidx;
5337
5338         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5339
5340         cdb = (struct scsi_prevent *)ctsio->cdb;
5341
5342         if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) {
5343                 ctl_set_invalid_opcode(ctsio);
5344                 ctl_done((union ctl_io *)ctsio);
5345                 return (CTL_RETVAL_COMPLETE);
5346         }
5347
5348         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5349         mtx_lock(&lun->lun_lock);
5350         if ((cdb->how & PR_PREVENT) &&
5351             ctl_is_set(lun->prevent, initidx) == 0) {
5352                 ctl_set_mask(lun->prevent, initidx);
5353                 lun->prevent_count++;
5354         } else if ((cdb->how & PR_PREVENT) == 0 &&
5355             ctl_is_set(lun->prevent, initidx)) {
5356                 ctl_clear_mask(lun->prevent, initidx);
5357                 lun->prevent_count--;
5358         }
5359         mtx_unlock(&lun->lun_lock);
5360         retval = lun->backend->config_write((union ctl_io *)ctsio);
5361         return (retval);
5362 }
5363
5364 /*
5365  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5366  * we don't really do anything with the LBA and length fields if the user
5367  * passes them in.  Instead we'll just flush out the cache for the entire
5368  * LUN.
5369  */
5370 int
5371 ctl_sync_cache(struct ctl_scsiio *ctsio)
5372 {
5373         struct ctl_lun *lun = CTL_LUN(ctsio);
5374         struct ctl_lba_len_flags *lbalen;
5375         uint64_t starting_lba;
5376         uint32_t block_count;
5377         int retval;
5378         uint8_t byte2;
5379
5380         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5381
5382         retval = 0;
5383
5384         switch (ctsio->cdb[0]) {
5385         case SYNCHRONIZE_CACHE: {
5386                 struct scsi_sync_cache *cdb;
5387                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5388
5389                 starting_lba = scsi_4btoul(cdb->begin_lba);
5390                 block_count = scsi_2btoul(cdb->lb_count);
5391                 byte2 = cdb->byte2;
5392                 break;
5393         }
5394         case SYNCHRONIZE_CACHE_16: {
5395                 struct scsi_sync_cache_16 *cdb;
5396                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5397
5398                 starting_lba = scsi_8btou64(cdb->begin_lba);
5399                 block_count = scsi_4btoul(cdb->lb_count);
5400                 byte2 = cdb->byte2;
5401                 break;
5402         }
5403         default:
5404                 ctl_set_invalid_opcode(ctsio);
5405                 ctl_done((union ctl_io *)ctsio);
5406                 goto bailout;
5407                 break; /* NOTREACHED */
5408         }
5409
5410         /*
5411          * We check the LBA and length, but don't do anything with them.
5412          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5413          * get flushed.  This check will just help satisfy anyone who wants
5414          * to see an error for an out of range LBA.
5415          */
5416         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5417                 ctl_set_lba_out_of_range(ctsio,
5418                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5419                 ctl_done((union ctl_io *)ctsio);
5420                 goto bailout;
5421         }
5422
5423         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5424         lbalen->lba = starting_lba;
5425         lbalen->len = block_count;
5426         lbalen->flags = byte2;
5427         retval = lun->backend->config_write((union ctl_io *)ctsio);
5428
5429 bailout:
5430         return (retval);
5431 }
5432
5433 int
5434 ctl_format(struct ctl_scsiio *ctsio)
5435 {
5436         struct scsi_format *cdb;
5437         int length, defect_list_len;
5438
5439         CTL_DEBUG_PRINT(("ctl_format\n"));
5440
5441         cdb = (struct scsi_format *)ctsio->cdb;
5442
5443         length = 0;
5444         if (cdb->byte2 & SF_FMTDATA) {
5445                 if (cdb->byte2 & SF_LONGLIST)
5446                         length = sizeof(struct scsi_format_header_long);
5447                 else
5448                         length = sizeof(struct scsi_format_header_short);
5449         }
5450
5451         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5452          && (length > 0)) {
5453                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5454                 ctsio->kern_data_len = length;
5455                 ctsio->kern_total_len = length;
5456                 ctsio->kern_data_resid = 0;
5457                 ctsio->kern_rel_offset = 0;
5458                 ctsio->kern_sg_entries = 0;
5459                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5460                 ctsio->be_move_done = ctl_config_move_done;
5461                 ctl_datamove((union ctl_io *)ctsio);
5462
5463                 return (CTL_RETVAL_COMPLETE);
5464         }
5465
5466         defect_list_len = 0;
5467
5468         if (cdb->byte2 & SF_FMTDATA) {
5469                 if (cdb->byte2 & SF_LONGLIST) {
5470                         struct scsi_format_header_long *header;
5471
5472                         header = (struct scsi_format_header_long *)
5473                                 ctsio->kern_data_ptr;
5474
5475                         defect_list_len = scsi_4btoul(header->defect_list_len);
5476                         if (defect_list_len != 0) {
5477                                 ctl_set_invalid_field(ctsio,
5478                                                       /*sks_valid*/ 1,
5479                                                       /*command*/ 0,
5480                                                       /*field*/ 2,
5481                                                       /*bit_valid*/ 0,
5482                                                       /*bit*/ 0);
5483                                 goto bailout;
5484                         }
5485                 } else {
5486                         struct scsi_format_header_short *header;
5487
5488                         header = (struct scsi_format_header_short *)
5489                                 ctsio->kern_data_ptr;
5490
5491                         defect_list_len = scsi_2btoul(header->defect_list_len);
5492                         if (defect_list_len != 0) {
5493                                 ctl_set_invalid_field(ctsio,
5494                                                       /*sks_valid*/ 1,
5495                                                       /*command*/ 0,
5496                                                       /*field*/ 2,
5497                                                       /*bit_valid*/ 0,
5498                                                       /*bit*/ 0);
5499                                 goto bailout;
5500                         }
5501                 }
5502         }
5503
5504         ctl_set_success(ctsio);
5505 bailout:
5506
5507         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5508                 free(ctsio->kern_data_ptr, M_CTL);
5509                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5510         }
5511
5512         ctl_done((union ctl_io *)ctsio);
5513         return (CTL_RETVAL_COMPLETE);
5514 }
5515
5516 int
5517 ctl_read_buffer(struct ctl_scsiio *ctsio)
5518 {
5519         struct ctl_lun *lun = CTL_LUN(ctsio);
5520         uint64_t buffer_offset;
5521         uint32_t len;
5522         uint8_t byte2;
5523         static uint8_t descr[4];
5524         static uint8_t echo_descr[4] = { 0 };
5525
5526         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5527
5528         switch (ctsio->cdb[0]) {
5529         case READ_BUFFER: {
5530                 struct scsi_read_buffer *cdb;
5531
5532                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5533                 buffer_offset = scsi_3btoul(cdb->offset);
5534                 len = scsi_3btoul(cdb->length);
5535                 byte2 = cdb->byte2;
5536                 break;
5537         }
5538         case READ_BUFFER_16: {
5539                 struct scsi_read_buffer_16 *cdb;
5540
5541                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5542                 buffer_offset = scsi_8btou64(cdb->offset);
5543                 len = scsi_4btoul(cdb->length);
5544                 byte2 = cdb->byte2;
5545                 break;
5546         }
5547         default: /* This shouldn't happen. */
5548                 ctl_set_invalid_opcode(ctsio);
5549                 ctl_done((union ctl_io *)ctsio);
5550                 return (CTL_RETVAL_COMPLETE);
5551         }
5552
5553         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5554             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5555                 ctl_set_invalid_field(ctsio,
5556                                       /*sks_valid*/ 1,
5557                                       /*command*/ 1,
5558                                       /*field*/ 6,
5559                                       /*bit_valid*/ 0,
5560                                       /*bit*/ 0);
5561                 ctl_done((union ctl_io *)ctsio);
5562                 return (CTL_RETVAL_COMPLETE);
5563         }
5564
5565         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5566                 descr[0] = 0;
5567                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5568                 ctsio->kern_data_ptr = descr;
5569                 len = min(len, sizeof(descr));
5570         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5571                 ctsio->kern_data_ptr = echo_descr;
5572                 len = min(len, sizeof(echo_descr));
5573         } else {
5574                 if (lun->write_buffer == NULL) {
5575                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5576                             M_CTL, M_WAITOK);
5577                 }
5578                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5579         }
5580         ctsio->kern_data_len = len;
5581         ctsio->kern_total_len = len;
5582         ctsio->kern_data_resid = 0;
5583         ctsio->kern_rel_offset = 0;
5584         ctsio->kern_sg_entries = 0;
5585         ctl_set_success(ctsio);
5586         ctsio->be_move_done = ctl_config_move_done;
5587         ctl_datamove((union ctl_io *)ctsio);
5588         return (CTL_RETVAL_COMPLETE);
5589 }
5590
5591 int
5592 ctl_write_buffer(struct ctl_scsiio *ctsio)
5593 {
5594         struct ctl_lun *lun = CTL_LUN(ctsio);
5595         struct scsi_write_buffer *cdb;
5596         int buffer_offset, len;
5597
5598         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5599
5600         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5601
5602         len = scsi_3btoul(cdb->length);
5603         buffer_offset = scsi_3btoul(cdb->offset);
5604
5605         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5606                 ctl_set_invalid_field(ctsio,
5607                                       /*sks_valid*/ 1,
5608                                       /*command*/ 1,
5609                                       /*field*/ 6,
5610                                       /*bit_valid*/ 0,
5611                                       /*bit*/ 0);
5612                 ctl_done((union ctl_io *)ctsio);
5613                 return (CTL_RETVAL_COMPLETE);
5614         }
5615
5616         /*
5617          * If we've got a kernel request that hasn't been malloced yet,
5618          * malloc it and tell the caller the data buffer is here.
5619          */
5620         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5621                 if (lun->write_buffer == NULL) {
5622                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5623                             M_CTL, M_WAITOK);
5624                 }
5625                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5626                 ctsio->kern_data_len = len;
5627                 ctsio->kern_total_len = len;
5628                 ctsio->kern_data_resid = 0;
5629                 ctsio->kern_rel_offset = 0;
5630                 ctsio->kern_sg_entries = 0;
5631                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5632                 ctsio->be_move_done = ctl_config_move_done;
5633                 ctl_datamove((union ctl_io *)ctsio);
5634
5635                 return (CTL_RETVAL_COMPLETE);
5636         }
5637
5638         ctl_set_success(ctsio);
5639         ctl_done((union ctl_io *)ctsio);
5640         return (CTL_RETVAL_COMPLETE);
5641 }
5642
5643 int
5644 ctl_write_same(struct ctl_scsiio *ctsio)
5645 {
5646         struct ctl_lun *lun = CTL_LUN(ctsio);
5647         struct ctl_lba_len_flags *lbalen;
5648         uint64_t lba;
5649         uint32_t num_blocks;
5650         int len, retval;
5651         uint8_t byte2;
5652
5653         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5654
5655         switch (ctsio->cdb[0]) {
5656         case WRITE_SAME_10: {
5657                 struct scsi_write_same_10 *cdb;
5658
5659                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5660
5661                 lba = scsi_4btoul(cdb->addr);
5662                 num_blocks = scsi_2btoul(cdb->length);
5663                 byte2 = cdb->byte2;
5664                 break;
5665         }
5666         case WRITE_SAME_16: {
5667                 struct scsi_write_same_16 *cdb;
5668
5669                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5670
5671                 lba = scsi_8btou64(cdb->addr);
5672                 num_blocks = scsi_4btoul(cdb->length);
5673                 byte2 = cdb->byte2;
5674                 break;
5675         }
5676         default:
5677                 /*
5678                  * We got a command we don't support.  This shouldn't
5679                  * happen, commands should be filtered out above us.
5680                  */
5681                 ctl_set_invalid_opcode(ctsio);
5682                 ctl_done((union ctl_io *)ctsio);
5683
5684                 return (CTL_RETVAL_COMPLETE);
5685                 break; /* NOTREACHED */
5686         }
5687
5688         /* ANCHOR flag can be used only together with UNMAP */
5689         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5690                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5691                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5692                 ctl_done((union ctl_io *)ctsio);
5693                 return (CTL_RETVAL_COMPLETE);
5694         }
5695
5696         /*
5697          * The first check is to make sure we're in bounds, the second
5698          * check is to catch wrap-around problems.  If the lba + num blocks
5699          * is less than the lba, then we've wrapped around and the block
5700          * range is invalid anyway.
5701          */
5702         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5703          || ((lba + num_blocks) < lba)) {
5704                 ctl_set_lba_out_of_range(ctsio,
5705                     MAX(lba, lun->be_lun->maxlba + 1));
5706                 ctl_done((union ctl_io *)ctsio);
5707                 return (CTL_RETVAL_COMPLETE);
5708         }
5709
5710         /* Zero number of blocks means "to the last logical block" */
5711         if (num_blocks == 0) {
5712                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5713                         ctl_set_invalid_field(ctsio,
5714                                               /*sks_valid*/ 0,
5715                                               /*command*/ 1,
5716                                               /*field*/ 0,
5717                                               /*bit_valid*/ 0,
5718                                               /*bit*/ 0);
5719                         ctl_done((union ctl_io *)ctsio);
5720                         return (CTL_RETVAL_COMPLETE);
5721                 }
5722                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5723         }
5724
5725         len = lun->be_lun->blocksize;
5726
5727         /*
5728          * If we've got a kernel request that hasn't been malloced yet,
5729          * malloc it and tell the caller the data buffer is here.
5730          */
5731         if ((byte2 & SWS_NDOB) == 0 &&
5732             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5733                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5734                 ctsio->kern_data_len = len;
5735                 ctsio->kern_total_len = len;
5736                 ctsio->kern_data_resid = 0;
5737                 ctsio->kern_rel_offset = 0;
5738                 ctsio->kern_sg_entries = 0;
5739                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5740                 ctsio->be_move_done = ctl_config_move_done;
5741                 ctl_datamove((union ctl_io *)ctsio);
5742
5743                 return (CTL_RETVAL_COMPLETE);
5744         }
5745
5746         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5747         lbalen->lba = lba;
5748         lbalen->len = num_blocks;
5749         lbalen->flags = byte2;
5750         retval = lun->backend->config_write((union ctl_io *)ctsio);
5751
5752         return (retval);
5753 }
5754
5755 int
5756 ctl_unmap(struct ctl_scsiio *ctsio)
5757 {
5758         struct ctl_lun *lun = CTL_LUN(ctsio);
5759         struct scsi_unmap *cdb;
5760         struct ctl_ptr_len_flags *ptrlen;
5761         struct scsi_unmap_header *hdr;
5762         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5763         uint64_t lba;
5764         uint32_t num_blocks;
5765         int len, retval;
5766         uint8_t byte2;
5767
5768         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5769
5770         cdb = (struct scsi_unmap *)ctsio->cdb;
5771         len = scsi_2btoul(cdb->length);
5772         byte2 = cdb->byte2;
5773
5774         /*
5775          * If we've got a kernel request that hasn't been malloced yet,
5776          * malloc it and tell the caller the data buffer is here.
5777          */
5778         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5779                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5780                 ctsio->kern_data_len = len;
5781                 ctsio->kern_total_len = len;
5782                 ctsio->kern_data_resid = 0;
5783                 ctsio->kern_rel_offset = 0;
5784                 ctsio->kern_sg_entries = 0;
5785                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5786                 ctsio->be_move_done = ctl_config_move_done;
5787                 ctl_datamove((union ctl_io *)ctsio);
5788
5789                 return (CTL_RETVAL_COMPLETE);
5790         }
5791
5792         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5793         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5794         if (len < sizeof (*hdr) ||
5795             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5796             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5797             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5798                 ctl_set_invalid_field(ctsio,
5799                                       /*sks_valid*/ 0,
5800                                       /*command*/ 0,
5801                                       /*field*/ 0,
5802                                       /*bit_valid*/ 0,
5803                                       /*bit*/ 0);
5804                 goto done;
5805         }
5806         len = scsi_2btoul(hdr->desc_length);
5807         buf = (struct scsi_unmap_desc *)(hdr + 1);
5808         end = buf + len / sizeof(*buf);
5809
5810         endnz = buf;
5811         for (range = buf; range < end; range++) {
5812                 lba = scsi_8btou64(range->lba);
5813                 num_blocks = scsi_4btoul(range->length);
5814                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5815                  || ((lba + num_blocks) < lba)) {
5816                         ctl_set_lba_out_of_range(ctsio,
5817                             MAX(lba, lun->be_lun->maxlba + 1));
5818                         ctl_done((union ctl_io *)ctsio);
5819                         return (CTL_RETVAL_COMPLETE);
5820                 }
5821                 if (num_blocks != 0)
5822                         endnz = range + 1;
5823         }
5824
5825         /*
5826          * Block backend can not handle zero last range.
5827          * Filter it out and return if there is nothing left.
5828          */
5829         len = (uint8_t *)endnz - (uint8_t *)buf;
5830         if (len == 0) {
5831                 ctl_set_success(ctsio);
5832                 goto done;
5833         }
5834
5835         mtx_lock(&lun->lun_lock);
5836         ptrlen = (struct ctl_ptr_len_flags *)
5837             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5838         ptrlen->ptr = (void *)buf;
5839         ptrlen->len = len;
5840         ptrlen->flags = byte2;
5841         ctl_check_blocked(lun);
5842         mtx_unlock(&lun->lun_lock);
5843
5844         retval = lun->backend->config_write((union ctl_io *)ctsio);
5845         return (retval);
5846
5847 done:
5848         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5849                 free(ctsio->kern_data_ptr, M_CTL);
5850                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5851         }
5852         ctl_done((union ctl_io *)ctsio);
5853         return (CTL_RETVAL_COMPLETE);
5854 }
5855
5856 int
5857 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5858                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5859 {
5860         struct ctl_lun *lun = CTL_LUN(ctsio);
5861         uint8_t *current_cp;
5862         int set_ua;
5863         uint32_t initidx;
5864
5865         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5866         set_ua = 0;
5867
5868         current_cp = (page_index->page_data + (page_index->page_len *
5869             CTL_PAGE_CURRENT));
5870
5871         mtx_lock(&lun->lun_lock);
5872         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5873                 memcpy(current_cp, page_ptr, page_index->page_len);
5874                 set_ua = 1;
5875         }
5876         if (set_ua != 0)
5877                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5878         mtx_unlock(&lun->lun_lock);
5879         if (set_ua) {
5880                 ctl_isc_announce_mode(lun,
5881                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5882                     page_index->page_code, page_index->subpage);
5883         }
5884         return (CTL_RETVAL_COMPLETE);
5885 }
5886
5887 static void
5888 ctl_ie_timer(void *arg)
5889 {
5890         struct ctl_lun *lun = arg;
5891         uint64_t t;
5892
5893         if (lun->ie_asc == 0)
5894                 return;
5895
5896         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5897                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5898         else
5899                 lun->ie_reported = 0;
5900
5901         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5902                 lun->ie_reportcnt++;
5903                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5904                 if (t == 0 || t == UINT32_MAX)
5905                         t = 3000;  /* 5 min */
5906                 callout_schedule(&lun->ie_callout, t * hz / 10);
5907         }
5908 }
5909
5910 int
5911 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5912                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5913 {
5914         struct ctl_lun *lun = CTL_LUN(ctsio);
5915         struct scsi_info_exceptions_page *pg;
5916         uint64_t t;
5917
5918         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5919
5920         pg = (struct scsi_info_exceptions_page *)page_ptr;
5921         mtx_lock(&lun->lun_lock);
5922         if (pg->info_flags & SIEP_FLAGS_TEST) {
5923                 lun->ie_asc = 0x5d;
5924                 lun->ie_ascq = 0xff;
5925                 if (pg->mrie == SIEP_MRIE_UA) {
5926                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
5927                         lun->ie_reported = 1;
5928                 } else {
5929                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5930                         lun->ie_reported = -1;
5931                 }
5932                 lun->ie_reportcnt = 1;
5933                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
5934                         lun->ie_reportcnt++;
5935                         t = scsi_4btoul(pg->interval_timer);
5936                         if (t == 0 || t == UINT32_MAX)
5937                                 t = 3000;  /* 5 min */
5938                         callout_reset(&lun->ie_callout, t * hz / 10,
5939                             ctl_ie_timer, lun);
5940                 }
5941         } else {
5942                 lun->ie_asc = 0;
5943                 lun->ie_ascq = 0;
5944                 lun->ie_reported = 1;
5945                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5946                 lun->ie_reportcnt = UINT32_MAX;
5947                 callout_stop(&lun->ie_callout);
5948         }
5949         mtx_unlock(&lun->lun_lock);
5950         return (CTL_RETVAL_COMPLETE);
5951 }
5952
5953 static int
5954 ctl_do_mode_select(union ctl_io *io)
5955 {
5956         struct ctl_lun *lun = CTL_LUN(io);
5957         struct scsi_mode_page_header *page_header;
5958         struct ctl_page_index *page_index;
5959         struct ctl_scsiio *ctsio;
5960         int page_len, page_len_offset, page_len_size;
5961         union ctl_modepage_info *modepage_info;
5962         uint16_t *len_left, *len_used;
5963         int retval, i;
5964
5965         ctsio = &io->scsiio;
5966         page_index = NULL;
5967         page_len = 0;
5968
5969         modepage_info = (union ctl_modepage_info *)
5970                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5971         len_left = &modepage_info->header.len_left;
5972         len_used = &modepage_info->header.len_used;
5973
5974 do_next_page:
5975
5976         page_header = (struct scsi_mode_page_header *)
5977                 (ctsio->kern_data_ptr + *len_used);
5978
5979         if (*len_left == 0) {
5980                 free(ctsio->kern_data_ptr, M_CTL);
5981                 ctl_set_success(ctsio);
5982                 ctl_done((union ctl_io *)ctsio);
5983                 return (CTL_RETVAL_COMPLETE);
5984         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
5985
5986                 free(ctsio->kern_data_ptr, M_CTL);
5987                 ctl_set_param_len_error(ctsio);
5988                 ctl_done((union ctl_io *)ctsio);
5989                 return (CTL_RETVAL_COMPLETE);
5990
5991         } else if ((page_header->page_code & SMPH_SPF)
5992                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
5993
5994                 free(ctsio->kern_data_ptr, M_CTL);
5995                 ctl_set_param_len_error(ctsio);
5996                 ctl_done((union ctl_io *)ctsio);
5997                 return (CTL_RETVAL_COMPLETE);
5998         }
5999
6000
6001         /*
6002          * XXX KDM should we do something with the block descriptor?
6003          */
6004         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6005                 page_index = &lun->mode_pages.index[i];
6006                 if (lun->be_lun->lun_type == T_DIRECT &&
6007                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6008                         continue;
6009                 if (lun->be_lun->lun_type == T_PROCESSOR &&
6010                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6011                         continue;
6012                 if (lun->be_lun->lun_type == T_CDROM &&
6013                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6014                         continue;
6015
6016                 if ((page_index->page_code & SMPH_PC_MASK) !=
6017                     (page_header->page_code & SMPH_PC_MASK))
6018                         continue;
6019
6020                 /*
6021                  * If neither page has a subpage code, then we've got a
6022                  * match.
6023                  */
6024                 if (((page_index->page_code & SMPH_SPF) == 0)
6025                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6026                         page_len = page_header->page_length;
6027                         break;
6028                 }
6029
6030                 /*
6031                  * If both pages have subpages, then the subpage numbers
6032                  * have to match.
6033                  */
6034                 if ((page_index->page_code & SMPH_SPF)
6035                   && (page_header->page_code & SMPH_SPF)) {
6036                         struct scsi_mode_page_header_sp *sph;
6037
6038                         sph = (struct scsi_mode_page_header_sp *)page_header;
6039                         if (page_index->subpage == sph->subpage) {
6040                                 page_len = scsi_2btoul(sph->page_length);
6041                                 break;
6042                         }
6043                 }
6044         }
6045
6046         /*
6047          * If we couldn't find the page, or if we don't have a mode select
6048          * handler for it, send back an error to the user.
6049          */
6050         if ((i >= CTL_NUM_MODE_PAGES)
6051          || (page_index->select_handler == NULL)) {
6052                 ctl_set_invalid_field(ctsio,
6053                                       /*sks_valid*/ 1,
6054                                       /*command*/ 0,
6055                                       /*field*/ *len_used,
6056                                       /*bit_valid*/ 0,
6057                                       /*bit*/ 0);
6058                 free(ctsio->kern_data_ptr, M_CTL);
6059                 ctl_done((union ctl_io *)ctsio);
6060                 return (CTL_RETVAL_COMPLETE);
6061         }
6062
6063         if (page_index->page_code & SMPH_SPF) {
6064                 page_len_offset = 2;
6065                 page_len_size = 2;
6066         } else {
6067                 page_len_size = 1;
6068                 page_len_offset = 1;
6069         }
6070
6071         /*
6072          * If the length the initiator gives us isn't the one we specify in
6073          * the mode page header, or if they didn't specify enough data in
6074          * the CDB to avoid truncating this page, kick out the request.
6075          */
6076         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6077                 ctl_set_invalid_field(ctsio,
6078                                       /*sks_valid*/ 1,
6079                                       /*command*/ 0,
6080                                       /*field*/ *len_used + page_len_offset,
6081                                       /*bit_valid*/ 0,
6082                                       /*bit*/ 0);
6083                 free(ctsio->kern_data_ptr, M_CTL);
6084                 ctl_done((union ctl_io *)ctsio);
6085                 return (CTL_RETVAL_COMPLETE);
6086         }
6087         if (*len_left < page_index->page_len) {
6088                 free(ctsio->kern_data_ptr, M_CTL);
6089                 ctl_set_param_len_error(ctsio);
6090                 ctl_done((union ctl_io *)ctsio);
6091                 return (CTL_RETVAL_COMPLETE);
6092         }
6093
6094         /*
6095          * Run through the mode page, checking to make sure that the bits
6096          * the user changed are actually legal for him to change.
6097          */
6098         for (i = 0; i < page_index->page_len; i++) {
6099                 uint8_t *user_byte, *change_mask, *current_byte;
6100                 int bad_bit;
6101                 int j;
6102
6103                 user_byte = (uint8_t *)page_header + i;
6104                 change_mask = page_index->page_data +
6105                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6106                 current_byte = page_index->page_data +
6107                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6108
6109                 /*
6110                  * Check to see whether the user set any bits in this byte
6111                  * that he is not allowed to set.
6112                  */
6113                 if ((*user_byte & ~(*change_mask)) ==
6114                     (*current_byte & ~(*change_mask)))
6115                         continue;
6116
6117                 /*
6118                  * Go through bit by bit to determine which one is illegal.
6119                  */
6120                 bad_bit = 0;
6121                 for (j = 7; j >= 0; j--) {
6122                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6123                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6124                                 bad_bit = i;
6125                                 break;
6126                         }
6127                 }
6128                 ctl_set_invalid_field(ctsio,
6129                                       /*sks_valid*/ 1,
6130                                       /*command*/ 0,
6131                                       /*field*/ *len_used + i,
6132                                       /*bit_valid*/ 1,
6133                                       /*bit*/ bad_bit);
6134                 free(ctsio->kern_data_ptr, M_CTL);
6135                 ctl_done((union ctl_io *)ctsio);
6136                 return (CTL_RETVAL_COMPLETE);
6137         }
6138
6139         /*
6140          * Decrement these before we call the page handler, since we may
6141          * end up getting called back one way or another before the handler
6142          * returns to this context.
6143          */
6144         *len_left -= page_index->page_len;
6145         *len_used += page_index->page_len;
6146
6147         retval = page_index->select_handler(ctsio, page_index,
6148                                             (uint8_t *)page_header);
6149
6150         /*
6151          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6152          * wait until this queued command completes to finish processing
6153          * the mode page.  If it returns anything other than
6154          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6155          * already set the sense information, freed the data pointer, and
6156          * completed the io for us.
6157          */
6158         if (retval != CTL_RETVAL_COMPLETE)
6159                 goto bailout_no_done;
6160
6161         /*
6162          * If the initiator sent us more than one page, parse the next one.
6163          */
6164         if (*len_left > 0)
6165                 goto do_next_page;
6166
6167         ctl_set_success(ctsio);
6168         free(ctsio->kern_data_ptr, M_CTL);
6169         ctl_done((union ctl_io *)ctsio);
6170
6171 bailout_no_done:
6172
6173         return (CTL_RETVAL_COMPLETE);
6174
6175 }
6176
6177 int
6178 ctl_mode_select(struct ctl_scsiio *ctsio)
6179 {
6180         struct ctl_lun *lun = CTL_LUN(ctsio);
6181         union ctl_modepage_info *modepage_info;
6182         int bd_len, i, header_size, param_len, pf, rtd, sp;
6183         uint32_t initidx;
6184
6185         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6186         switch (ctsio->cdb[0]) {
6187         case MODE_SELECT_6: {
6188                 struct scsi_mode_select_6 *cdb;
6189
6190                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6191
6192                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6193                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6194                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6195                 param_len = cdb->length;
6196                 header_size = sizeof(struct scsi_mode_header_6);
6197                 break;
6198         }
6199         case MODE_SELECT_10: {
6200                 struct scsi_mode_select_10 *cdb;
6201
6202                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6203
6204                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6205                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6206                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6207                 param_len = scsi_2btoul(cdb->length);
6208                 header_size = sizeof(struct scsi_mode_header_10);
6209                 break;
6210         }
6211         default:
6212                 ctl_set_invalid_opcode(ctsio);
6213                 ctl_done((union ctl_io *)ctsio);
6214                 return (CTL_RETVAL_COMPLETE);
6215         }
6216
6217         if (rtd) {
6218                 if (param_len != 0) {
6219                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 0,
6220                             /*command*/ 1, /*field*/ 0,
6221                             /*bit_valid*/ 0, /*bit*/ 0);
6222                         ctl_done((union ctl_io *)ctsio);
6223                         return (CTL_RETVAL_COMPLETE);
6224                 }
6225
6226                 /* Revert to defaults. */
6227                 ctl_init_page_index(lun);
6228                 mtx_lock(&lun->lun_lock);
6229                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6230                 mtx_unlock(&lun->lun_lock);
6231                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6232                         ctl_isc_announce_mode(lun, -1,
6233                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
6234                             lun->mode_pages.index[i].subpage);
6235                 }
6236                 ctl_set_success(ctsio);
6237                 ctl_done((union ctl_io *)ctsio);
6238                 return (CTL_RETVAL_COMPLETE);
6239         }
6240
6241         /*
6242          * From SPC-3:
6243          * "A parameter list length of zero indicates that the Data-Out Buffer
6244          * shall be empty. This condition shall not be considered as an error."
6245          */
6246         if (param_len == 0) {
6247                 ctl_set_success(ctsio);
6248                 ctl_done((union ctl_io *)ctsio);
6249                 return (CTL_RETVAL_COMPLETE);
6250         }
6251
6252         /*
6253          * Since we'll hit this the first time through, prior to
6254          * allocation, we don't need to free a data buffer here.
6255          */
6256         if (param_len < header_size) {
6257                 ctl_set_param_len_error(ctsio);
6258                 ctl_done((union ctl_io *)ctsio);
6259                 return (CTL_RETVAL_COMPLETE);
6260         }
6261
6262         /*
6263          * Allocate the data buffer and grab the user's data.  In theory,
6264          * we shouldn't have to sanity check the parameter list length here
6265          * because the maximum size is 64K.  We should be able to malloc
6266          * that much without too many problems.
6267          */
6268         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6269                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6270                 ctsio->kern_data_len = param_len;
6271                 ctsio->kern_total_len = param_len;
6272                 ctsio->kern_data_resid = 0;
6273                 ctsio->kern_rel_offset = 0;
6274                 ctsio->kern_sg_entries = 0;
6275                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6276                 ctsio->be_move_done = ctl_config_move_done;
6277                 ctl_datamove((union ctl_io *)ctsio);
6278
6279                 return (CTL_RETVAL_COMPLETE);
6280         }
6281
6282         switch (ctsio->cdb[0]) {
6283         case MODE_SELECT_6: {
6284                 struct scsi_mode_header_6 *mh6;
6285
6286                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6287                 bd_len = mh6->blk_desc_len;
6288                 break;
6289         }
6290         case MODE_SELECT_10: {
6291                 struct scsi_mode_header_10 *mh10;
6292
6293                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6294                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6295                 break;
6296         }
6297         default:
6298                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6299         }
6300
6301         if (param_len < (header_size + bd_len)) {
6302                 free(ctsio->kern_data_ptr, M_CTL);
6303                 ctl_set_param_len_error(ctsio);
6304                 ctl_done((union ctl_io *)ctsio);
6305                 return (CTL_RETVAL_COMPLETE);
6306         }
6307
6308         /*
6309          * Set the IO_CONT flag, so that if this I/O gets passed to
6310          * ctl_config_write_done(), it'll get passed back to
6311          * ctl_do_mode_select() for further processing, or completion if
6312          * we're all done.
6313          */
6314         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6315         ctsio->io_cont = ctl_do_mode_select;
6316
6317         modepage_info = (union ctl_modepage_info *)
6318                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6319         memset(modepage_info, 0, sizeof(*modepage_info));
6320         modepage_info->header.len_left = param_len - header_size - bd_len;
6321         modepage_info->header.len_used = header_size + bd_len;
6322
6323         return (ctl_do_mode_select((union ctl_io *)ctsio));
6324 }
6325
6326 int
6327 ctl_mode_sense(struct ctl_scsiio *ctsio)
6328 {
6329         struct ctl_lun *lun = CTL_LUN(ctsio);
6330         int pc, page_code, dbd, llba, subpage;
6331         int alloc_len, page_len, header_len, total_len;
6332         struct scsi_mode_block_descr *block_desc;
6333         struct ctl_page_index *page_index;
6334
6335         dbd = 0;
6336         llba = 0;
6337         block_desc = NULL;
6338
6339         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6340
6341         switch (ctsio->cdb[0]) {
6342         case MODE_SENSE_6: {
6343                 struct scsi_mode_sense_6 *cdb;
6344
6345                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6346
6347                 header_len = sizeof(struct scsi_mode_hdr_6);
6348                 if (cdb->byte2 & SMS_DBD)
6349                         dbd = 1;
6350                 else
6351                         header_len += sizeof(struct scsi_mode_block_descr);
6352
6353                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6354                 page_code = cdb->page & SMS_PAGE_CODE;
6355                 subpage = cdb->subpage;
6356                 alloc_len = cdb->length;
6357                 break;
6358         }
6359         case MODE_SENSE_10: {
6360                 struct scsi_mode_sense_10 *cdb;
6361
6362                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6363
6364                 header_len = sizeof(struct scsi_mode_hdr_10);
6365
6366                 if (cdb->byte2 & SMS_DBD)
6367                         dbd = 1;
6368                 else
6369                         header_len += sizeof(struct scsi_mode_block_descr);
6370                 if (cdb->byte2 & SMS10_LLBAA)
6371                         llba = 1;
6372                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6373                 page_code = cdb->page & SMS_PAGE_CODE;
6374                 subpage = cdb->subpage;
6375                 alloc_len = scsi_2btoul(cdb->length);
6376                 break;
6377         }
6378         default:
6379                 ctl_set_invalid_opcode(ctsio);
6380                 ctl_done((union ctl_io *)ctsio);
6381                 return (CTL_RETVAL_COMPLETE);
6382                 break; /* NOTREACHED */
6383         }
6384
6385         /*
6386          * We have to make a first pass through to calculate the size of
6387          * the pages that match the user's query.  Then we allocate enough
6388          * memory to hold it, and actually copy the data into the buffer.
6389          */
6390         switch (page_code) {
6391         case SMS_ALL_PAGES_PAGE: {
6392                 u_int i;
6393
6394                 page_len = 0;
6395
6396                 /*
6397                  * At the moment, values other than 0 and 0xff here are
6398                  * reserved according to SPC-3.
6399                  */
6400                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6401                  && (subpage != SMS_SUBPAGE_ALL)) {
6402                         ctl_set_invalid_field(ctsio,
6403                                               /*sks_valid*/ 1,
6404                                               /*command*/ 1,
6405                                               /*field*/ 3,
6406                                               /*bit_valid*/ 0,
6407                                               /*bit*/ 0);
6408                         ctl_done((union ctl_io *)ctsio);
6409                         return (CTL_RETVAL_COMPLETE);
6410                 }
6411
6412                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6413                         page_index = &lun->mode_pages.index[i];
6414
6415                         /* Make sure the page is supported for this dev type */
6416                         if (lun->be_lun->lun_type == T_DIRECT &&
6417                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6418                                 continue;
6419                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6420                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6421                                 continue;
6422                         if (lun->be_lun->lun_type == T_CDROM &&
6423                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6424                                 continue;
6425
6426                         /*
6427                          * We don't use this subpage if the user didn't
6428                          * request all subpages.
6429                          */
6430                         if ((page_index->subpage != 0)
6431                          && (subpage == SMS_SUBPAGE_PAGE_0))
6432                                 continue;
6433
6434 #if 0
6435                         printf("found page %#x len %d\n",
6436                                page_index->page_code & SMPH_PC_MASK,
6437                                page_index->page_len);
6438 #endif
6439                         page_len += page_index->page_len;
6440                 }
6441                 break;
6442         }
6443         default: {
6444                 u_int i;
6445
6446                 page_len = 0;
6447
6448                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6449                         page_index = &lun->mode_pages.index[i];
6450
6451                         /* Make sure the page is supported for this dev type */
6452                         if (lun->be_lun->lun_type == T_DIRECT &&
6453                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6454                                 continue;
6455                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6456                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6457                                 continue;
6458                         if (lun->be_lun->lun_type == T_CDROM &&
6459                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6460                                 continue;
6461
6462                         /* Look for the right page code */
6463                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6464                                 continue;
6465
6466                         /* Look for the right subpage or the subpage wildcard*/
6467                         if ((page_index->subpage != subpage)
6468                          && (subpage != SMS_SUBPAGE_ALL))
6469                                 continue;
6470
6471 #if 0
6472                         printf("found page %#x len %d\n",
6473                                page_index->page_code & SMPH_PC_MASK,
6474                                page_index->page_len);
6475 #endif
6476
6477                         page_len += page_index->page_len;
6478                 }
6479
6480                 if (page_len == 0) {
6481                         ctl_set_invalid_field(ctsio,
6482                                               /*sks_valid*/ 1,
6483                                               /*command*/ 1,
6484                                               /*field*/ 2,
6485                                               /*bit_valid*/ 1,
6486                                               /*bit*/ 5);
6487                         ctl_done((union ctl_io *)ctsio);
6488                         return (CTL_RETVAL_COMPLETE);
6489                 }
6490                 break;
6491         }
6492         }
6493
6494         total_len = header_len + page_len;
6495 #if 0
6496         printf("header_len = %d, page_len = %d, total_len = %d\n",
6497                header_len, page_len, total_len);
6498 #endif
6499
6500         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6501         ctsio->kern_sg_entries = 0;
6502         ctsio->kern_data_resid = 0;
6503         ctsio->kern_rel_offset = 0;
6504         if (total_len < alloc_len) {
6505                 ctsio->residual = alloc_len - total_len;
6506                 ctsio->kern_data_len = total_len;
6507                 ctsio->kern_total_len = total_len;
6508         } else {
6509                 ctsio->residual = 0;
6510                 ctsio->kern_data_len = alloc_len;
6511                 ctsio->kern_total_len = alloc_len;
6512         }
6513
6514         switch (ctsio->cdb[0]) {
6515         case MODE_SENSE_6: {
6516                 struct scsi_mode_hdr_6 *header;
6517
6518                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6519
6520                 header->datalen = MIN(total_len - 1, 254);
6521                 if (lun->be_lun->lun_type == T_DIRECT) {
6522                         header->dev_specific = 0x10; /* DPOFUA */
6523                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6524                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6525                                 header->dev_specific |= 0x80; /* WP */
6526                 }
6527                 if (dbd)
6528                         header->block_descr_len = 0;
6529                 else
6530                         header->block_descr_len =
6531                                 sizeof(struct scsi_mode_block_descr);
6532                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6533                 break;
6534         }
6535         case MODE_SENSE_10: {
6536                 struct scsi_mode_hdr_10 *header;
6537                 int datalen;
6538
6539                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6540
6541                 datalen = MIN(total_len - 2, 65533);
6542                 scsi_ulto2b(datalen, header->datalen);
6543                 if (lun->be_lun->lun_type == T_DIRECT) {
6544                         header->dev_specific = 0x10; /* DPOFUA */
6545                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6546                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6547                                 header->dev_specific |= 0x80; /* WP */
6548                 }
6549                 if (dbd)
6550                         scsi_ulto2b(0, header->block_descr_len);
6551                 else
6552                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6553                                     header->block_descr_len);
6554                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6555                 break;
6556         }
6557         default:
6558                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6559         }
6560
6561         /*
6562          * If we've got a disk, use its blocksize in the block
6563          * descriptor.  Otherwise, just set it to 0.
6564          */
6565         if (dbd == 0) {
6566                 if (lun->be_lun->lun_type == T_DIRECT)
6567                         scsi_ulto3b(lun->be_lun->blocksize,
6568                                     block_desc->block_len);
6569                 else
6570                         scsi_ulto3b(0, block_desc->block_len);
6571         }
6572
6573         switch (page_code) {
6574         case SMS_ALL_PAGES_PAGE: {
6575                 int i, data_used;
6576
6577                 data_used = header_len;
6578                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6579                         struct ctl_page_index *page_index;
6580
6581                         page_index = &lun->mode_pages.index[i];
6582                         if (lun->be_lun->lun_type == T_DIRECT &&
6583                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6584                                 continue;
6585                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6586                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6587                                 continue;
6588                         if (lun->be_lun->lun_type == T_CDROM &&
6589                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6590                                 continue;
6591
6592                         /*
6593                          * We don't use this subpage if the user didn't
6594                          * request all subpages.  We already checked (above)
6595                          * to make sure the user only specified a subpage
6596                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6597                          */
6598                         if ((page_index->subpage != 0)
6599                          && (subpage == SMS_SUBPAGE_PAGE_0))
6600                                 continue;
6601
6602                         /*
6603                          * Call the handler, if it exists, to update the
6604                          * page to the latest values.
6605                          */
6606                         if (page_index->sense_handler != NULL)
6607                                 page_index->sense_handler(ctsio, page_index,pc);
6608
6609                         memcpy(ctsio->kern_data_ptr + data_used,
6610                                page_index->page_data +
6611                                (page_index->page_len * pc),
6612                                page_index->page_len);
6613                         data_used += page_index->page_len;
6614                 }
6615                 break;
6616         }
6617         default: {
6618                 int i, data_used;
6619
6620                 data_used = header_len;
6621
6622                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6623                         struct ctl_page_index *page_index;
6624
6625                         page_index = &lun->mode_pages.index[i];
6626
6627                         /* Look for the right page code */
6628                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6629                                 continue;
6630
6631                         /* Look for the right subpage or the subpage wildcard*/
6632                         if ((page_index->subpage != subpage)
6633                          && (subpage != SMS_SUBPAGE_ALL))
6634                                 continue;
6635
6636                         /* Make sure the page is supported for this dev type */
6637                         if (lun->be_lun->lun_type == T_DIRECT &&
6638                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6639                                 continue;
6640                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6641                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6642                                 continue;
6643                         if (lun->be_lun->lun_type == T_CDROM &&
6644                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6645                                 continue;
6646
6647                         /*
6648                          * Call the handler, if it exists, to update the
6649                          * page to the latest values.
6650                          */
6651                         if (page_index->sense_handler != NULL)
6652                                 page_index->sense_handler(ctsio, page_index,pc);
6653
6654                         memcpy(ctsio->kern_data_ptr + data_used,
6655                                page_index->page_data +
6656                                (page_index->page_len * pc),
6657                                page_index->page_len);
6658                         data_used += page_index->page_len;
6659                 }
6660                 break;
6661         }
6662         }
6663
6664         ctl_set_success(ctsio);
6665         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6666         ctsio->be_move_done = ctl_config_move_done;
6667         ctl_datamove((union ctl_io *)ctsio);
6668         return (CTL_RETVAL_COMPLETE);
6669 }
6670
6671 int
6672 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6673                                struct ctl_page_index *page_index,
6674                                int pc)
6675 {
6676         struct ctl_lun *lun = CTL_LUN(ctsio);
6677         struct scsi_log_param_header *phdr;
6678         uint8_t *data;
6679         uint64_t val;
6680
6681         data = page_index->page_data;
6682
6683         if (lun->backend->lun_attr != NULL &&
6684             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6685              != UINT64_MAX) {
6686                 phdr = (struct scsi_log_param_header *)data;
6687                 scsi_ulto2b(0x0001, phdr->param_code);
6688                 phdr->param_control = SLP_LBIN | SLP_LP;
6689                 phdr->param_len = 8;
6690                 data = (uint8_t *)(phdr + 1);
6691                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6692                 data[4] = 0x02; /* per-pool */
6693                 data += phdr->param_len;
6694         }
6695
6696         if (lun->backend->lun_attr != NULL &&
6697             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6698              != UINT64_MAX) {
6699                 phdr = (struct scsi_log_param_header *)data;
6700                 scsi_ulto2b(0x0002, phdr->param_code);
6701                 phdr->param_control = SLP_LBIN | SLP_LP;
6702                 phdr->param_len = 8;
6703                 data = (uint8_t *)(phdr + 1);
6704                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6705                 data[4] = 0x01; /* per-LUN */
6706                 data += phdr->param_len;
6707         }
6708
6709         if (lun->backend->lun_attr != NULL &&
6710             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6711              != UINT64_MAX) {
6712                 phdr = (struct scsi_log_param_header *)data;
6713                 scsi_ulto2b(0x00f1, phdr->param_code);
6714                 phdr->param_control = SLP_LBIN | SLP_LP;
6715                 phdr->param_len = 8;
6716                 data = (uint8_t *)(phdr + 1);
6717                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6718                 data[4] = 0x02; /* per-pool */
6719                 data += phdr->param_len;
6720         }
6721
6722         if (lun->backend->lun_attr != NULL &&
6723             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6724              != UINT64_MAX) {
6725                 phdr = (struct scsi_log_param_header *)data;
6726                 scsi_ulto2b(0x00f2, phdr->param_code);
6727                 phdr->param_control = SLP_LBIN | SLP_LP;
6728                 phdr->param_len = 8;
6729                 data = (uint8_t *)(phdr + 1);
6730                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6731                 data[4] = 0x02; /* per-pool */
6732                 data += phdr->param_len;
6733         }
6734
6735         page_index->page_len = data - page_index->page_data;
6736         return (0);
6737 }
6738
6739 int
6740 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6741                                struct ctl_page_index *page_index,
6742                                int pc)
6743 {
6744         struct ctl_lun *lun = CTL_LUN(ctsio);
6745         struct stat_page *data;
6746         struct bintime *t;
6747
6748         data = (struct stat_page *)page_index->page_data;
6749
6750         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6751         data->sap.hdr.param_control = SLP_LBIN;
6752         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6753             sizeof(struct scsi_log_param_header);
6754         scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
6755             data->sap.read_num);
6756         scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
6757             data->sap.write_num);
6758         if (lun->be_lun->blocksize > 0) {
6759                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
6760                     lun->be_lun->blocksize, data->sap.recvieved_lba);
6761                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
6762                     lun->be_lun->blocksize, data->sap.transmitted_lba);
6763         }
6764         t = &lun->stats.time[CTL_STATS_READ];
6765         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6766             data->sap.read_int);
6767         t = &lun->stats.time[CTL_STATS_WRITE];
6768         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6769             data->sap.write_int);
6770         scsi_u64to8b(0, data->sap.weighted_num);
6771         scsi_u64to8b(0, data->sap.weighted_int);
6772         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6773         data->it.hdr.param_control = SLP_LBIN;
6774         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6775             sizeof(struct scsi_log_param_header);
6776 #ifdef CTL_TIME_IO
6777         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6778 #endif
6779         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6780         data->it.hdr.param_control = SLP_LBIN;
6781         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6782             sizeof(struct scsi_log_param_header);
6783         scsi_ulto4b(3, data->ti.exponent);
6784         scsi_ulto4b(1, data->ti.integer);
6785         return (0);
6786 }
6787
6788 int
6789 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6790                                struct ctl_page_index *page_index,
6791                                int pc)
6792 {
6793         struct ctl_lun *lun = CTL_LUN(ctsio);
6794         struct scsi_log_informational_exceptions *data;
6795
6796         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6797
6798         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6799         data->hdr.param_control = SLP_LBIN;
6800         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6801             sizeof(struct scsi_log_param_header);
6802         data->ie_asc = lun->ie_asc;
6803         data->ie_ascq = lun->ie_ascq;
6804         data->temperature = 0xff;
6805         return (0);
6806 }
6807
6808 int
6809 ctl_log_sense(struct ctl_scsiio *ctsio)
6810 {
6811         struct ctl_lun *lun = CTL_LUN(ctsio);
6812         int i, pc, page_code, subpage;
6813         int alloc_len, total_len;
6814         struct ctl_page_index *page_index;
6815         struct scsi_log_sense *cdb;
6816         struct scsi_log_header *header;
6817
6818         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6819
6820         cdb = (struct scsi_log_sense *)ctsio->cdb;
6821         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6822         page_code = cdb->page & SLS_PAGE_CODE;
6823         subpage = cdb->subpage;
6824         alloc_len = scsi_2btoul(cdb->length);
6825
6826         page_index = NULL;
6827         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6828                 page_index = &lun->log_pages.index[i];
6829
6830                 /* Look for the right page code */
6831                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6832                         continue;
6833
6834                 /* Look for the right subpage or the subpage wildcard*/
6835                 if (page_index->subpage != subpage)
6836                         continue;
6837
6838                 break;
6839         }
6840         if (i >= CTL_NUM_LOG_PAGES) {
6841                 ctl_set_invalid_field(ctsio,
6842                                       /*sks_valid*/ 1,
6843                                       /*command*/ 1,
6844                                       /*field*/ 2,
6845                                       /*bit_valid*/ 0,
6846                                       /*bit*/ 0);
6847                 ctl_done((union ctl_io *)ctsio);
6848                 return (CTL_RETVAL_COMPLETE);
6849         }
6850
6851         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6852
6853         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6854         ctsio->kern_sg_entries = 0;
6855         ctsio->kern_data_resid = 0;
6856         ctsio->kern_rel_offset = 0;
6857         if (total_len < alloc_len) {
6858                 ctsio->residual = alloc_len - total_len;
6859                 ctsio->kern_data_len = total_len;
6860                 ctsio->kern_total_len = total_len;
6861         } else {
6862                 ctsio->residual = 0;
6863                 ctsio->kern_data_len = alloc_len;
6864                 ctsio->kern_total_len = alloc_len;
6865         }
6866
6867         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6868         header->page = page_index->page_code;
6869         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6870                 header->page |= SL_DS;
6871         if (page_index->subpage) {
6872                 header->page |= SL_SPF;
6873                 header->subpage = page_index->subpage;
6874         }
6875         scsi_ulto2b(page_index->page_len, header->datalen);
6876
6877         /*
6878          * Call the handler, if it exists, to update the
6879          * page to the latest values.
6880          */
6881         if (page_index->sense_handler != NULL)
6882                 page_index->sense_handler(ctsio, page_index, pc);
6883
6884         memcpy(header + 1, page_index->page_data, page_index->page_len);
6885
6886         ctl_set_success(ctsio);
6887         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6888         ctsio->be_move_done = ctl_config_move_done;
6889         ctl_datamove((union ctl_io *)ctsio);
6890         return (CTL_RETVAL_COMPLETE);
6891 }
6892
6893 int
6894 ctl_read_capacity(struct ctl_scsiio *ctsio)
6895 {
6896         struct ctl_lun *lun = CTL_LUN(ctsio);
6897         struct scsi_read_capacity *cdb;
6898         struct scsi_read_capacity_data *data;
6899         uint32_t lba;
6900
6901         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6902
6903         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6904
6905         lba = scsi_4btoul(cdb->addr);
6906         if (((cdb->pmi & SRC_PMI) == 0)
6907          && (lba != 0)) {
6908                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6909                                       /*sks_valid*/ 1,
6910                                       /*command*/ 1,
6911                                       /*field*/ 2,
6912                                       /*bit_valid*/ 0,
6913                                       /*bit*/ 0);
6914                 ctl_done((union ctl_io *)ctsio);
6915                 return (CTL_RETVAL_COMPLETE);
6916         }
6917
6918         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6919         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6920         ctsio->residual = 0;
6921         ctsio->kern_data_len = sizeof(*data);
6922         ctsio->kern_total_len = sizeof(*data);
6923         ctsio->kern_data_resid = 0;
6924         ctsio->kern_rel_offset = 0;
6925         ctsio->kern_sg_entries = 0;
6926
6927         /*
6928          * If the maximum LBA is greater than 0xfffffffe, the user must
6929          * issue a SERVICE ACTION IN (16) command, with the read capacity
6930          * serivce action set.
6931          */
6932         if (lun->be_lun->maxlba > 0xfffffffe)
6933                 scsi_ulto4b(0xffffffff, data->addr);
6934         else
6935                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6936
6937         /*
6938          * XXX KDM this may not be 512 bytes...
6939          */
6940         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6941
6942         ctl_set_success(ctsio);
6943         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6944         ctsio->be_move_done = ctl_config_move_done;
6945         ctl_datamove((union ctl_io *)ctsio);
6946         return (CTL_RETVAL_COMPLETE);
6947 }
6948
6949 int
6950 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6951 {
6952         struct ctl_lun *lun = CTL_LUN(ctsio);
6953         struct scsi_read_capacity_16 *cdb;
6954         struct scsi_read_capacity_data_long *data;
6955         uint64_t lba;
6956         uint32_t alloc_len;
6957
6958         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6959
6960         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6961
6962         alloc_len = scsi_4btoul(cdb->alloc_len);
6963         lba = scsi_8btou64(cdb->addr);
6964
6965         if ((cdb->reladr & SRC16_PMI)
6966          && (lba != 0)) {
6967                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6968                                       /*sks_valid*/ 1,
6969                                       /*command*/ 1,
6970                                       /*field*/ 2,
6971                                       /*bit_valid*/ 0,
6972                                       /*bit*/ 0);
6973                 ctl_done((union ctl_io *)ctsio);
6974                 return (CTL_RETVAL_COMPLETE);
6975         }
6976
6977         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6978         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6979
6980         if (sizeof(*data) < alloc_len) {
6981                 ctsio->residual = alloc_len - sizeof(*data);
6982                 ctsio->kern_data_len = sizeof(*data);
6983                 ctsio->kern_total_len = sizeof(*data);
6984         } else {
6985                 ctsio->residual = 0;
6986                 ctsio->kern_data_len = alloc_len;
6987                 ctsio->kern_total_len = alloc_len;
6988         }
6989         ctsio->kern_data_resid = 0;
6990         ctsio->kern_rel_offset = 0;
6991         ctsio->kern_sg_entries = 0;
6992
6993         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6994         /* XXX KDM this may not be 512 bytes... */
6995         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6996         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6997         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6998         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6999                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
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_get_lba_status(struct ctl_scsiio *ctsio)
7010 {
7011         struct ctl_lun *lun = CTL_LUN(ctsio);
7012         struct scsi_get_lba_status *cdb;
7013         struct scsi_get_lba_status_data *data;
7014         struct ctl_lba_len_flags *lbalen;
7015         uint64_t lba;
7016         uint32_t alloc_len, total_len;
7017         int retval;
7018
7019         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7020
7021         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7022         lba = scsi_8btou64(cdb->addr);
7023         alloc_len = scsi_4btoul(cdb->alloc_len);
7024
7025         if (lba > lun->be_lun->maxlba) {
7026                 ctl_set_lba_out_of_range(ctsio, lba);
7027                 ctl_done((union ctl_io *)ctsio);
7028                 return (CTL_RETVAL_COMPLETE);
7029         }
7030
7031         total_len = sizeof(*data) + sizeof(data->descr[0]);
7032         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7033         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7034
7035         if (total_len < alloc_len) {
7036                 ctsio->residual = alloc_len - total_len;
7037                 ctsio->kern_data_len = total_len;
7038                 ctsio->kern_total_len = total_len;
7039         } else {
7040                 ctsio->residual = 0;
7041                 ctsio->kern_data_len = alloc_len;
7042                 ctsio->kern_total_len = alloc_len;
7043         }
7044         ctsio->kern_data_resid = 0;
7045         ctsio->kern_rel_offset = 0;
7046         ctsio->kern_sg_entries = 0;
7047
7048         /* Fill dummy data in case backend can't tell anything. */
7049         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7050         scsi_u64to8b(lba, data->descr[0].addr);
7051         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7052             data->descr[0].length);
7053         data->descr[0].status = 0; /* Mapped or unknown. */
7054
7055         ctl_set_success(ctsio);
7056         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7057         ctsio->be_move_done = ctl_config_move_done;
7058
7059         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7060         lbalen->lba = lba;
7061         lbalen->len = total_len;
7062         lbalen->flags = 0;
7063         retval = lun->backend->config_read((union ctl_io *)ctsio);
7064         return (CTL_RETVAL_COMPLETE);
7065 }
7066
7067 int
7068 ctl_read_defect(struct ctl_scsiio *ctsio)
7069 {
7070         struct scsi_read_defect_data_10 *ccb10;
7071         struct scsi_read_defect_data_12 *ccb12;
7072         struct scsi_read_defect_data_hdr_10 *data10;
7073         struct scsi_read_defect_data_hdr_12 *data12;
7074         uint32_t alloc_len, data_len;
7075         uint8_t format;
7076
7077         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7078
7079         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7080                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7081                 format = ccb10->format;
7082                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7083                 data_len = sizeof(*data10);
7084         } else {
7085                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7086                 format = ccb12->format;
7087                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7088                 data_len = sizeof(*data12);
7089         }
7090         if (alloc_len == 0) {
7091                 ctl_set_success(ctsio);
7092                 ctl_done((union ctl_io *)ctsio);
7093                 return (CTL_RETVAL_COMPLETE);
7094         }
7095
7096         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7097         if (data_len < alloc_len) {
7098                 ctsio->residual = alloc_len - data_len;
7099                 ctsio->kern_data_len = data_len;
7100                 ctsio->kern_total_len = data_len;
7101         } else {
7102                 ctsio->residual = 0;
7103                 ctsio->kern_data_len = alloc_len;
7104                 ctsio->kern_total_len = alloc_len;
7105         }
7106         ctsio->kern_data_resid = 0;
7107         ctsio->kern_rel_offset = 0;
7108         ctsio->kern_sg_entries = 0;
7109
7110         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7111                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7112                     ctsio->kern_data_ptr;
7113                 data10->format = format;
7114                 scsi_ulto2b(0, data10->length);
7115         } else {
7116                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7117                     ctsio->kern_data_ptr;
7118                 data12->format = format;
7119                 scsi_ulto2b(0, data12->generation);
7120                 scsi_ulto4b(0, data12->length);
7121         }
7122
7123         ctl_set_success(ctsio);
7124         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7125         ctsio->be_move_done = ctl_config_move_done;
7126         ctl_datamove((union ctl_io *)ctsio);
7127         return (CTL_RETVAL_COMPLETE);
7128 }
7129
7130 int
7131 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7132 {
7133         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7134         struct ctl_lun *lun = CTL_LUN(ctsio);
7135         struct scsi_maintenance_in *cdb;
7136         int retval;
7137         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7138         int num_ha_groups, num_target_ports, shared_group;
7139         struct ctl_port *port;
7140         struct scsi_target_group_data *rtg_ptr;
7141         struct scsi_target_group_data_extended *rtg_ext_ptr;
7142         struct scsi_target_port_group_descriptor *tpg_desc;
7143
7144         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7145
7146         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7147         retval = CTL_RETVAL_COMPLETE;
7148
7149         switch (cdb->byte2 & STG_PDF_MASK) {
7150         case STG_PDF_LENGTH:
7151                 ext = 0;
7152                 break;
7153         case STG_PDF_EXTENDED:
7154                 ext = 1;
7155                 break;
7156         default:
7157                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7158                                       /*sks_valid*/ 1,
7159                                       /*command*/ 1,
7160                                       /*field*/ 2,
7161                                       /*bit_valid*/ 1,
7162                                       /*bit*/ 5);
7163                 ctl_done((union ctl_io *)ctsio);
7164                 return(retval);
7165         }
7166
7167         num_target_ports = 0;
7168         shared_group = (softc->is_single != 0);
7169         mtx_lock(&softc->ctl_lock);
7170         STAILQ_FOREACH(port, &softc->port_list, links) {
7171                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7172                         continue;
7173                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7174                         continue;
7175                 num_target_ports++;
7176                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7177                         shared_group = 1;
7178         }
7179         mtx_unlock(&softc->ctl_lock);
7180         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7181
7182         if (ext)
7183                 total_len = sizeof(struct scsi_target_group_data_extended);
7184         else
7185                 total_len = sizeof(struct scsi_target_group_data);
7186         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7187                 (shared_group + num_ha_groups) +
7188             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7189
7190         alloc_len = scsi_4btoul(cdb->length);
7191
7192         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7193
7194         ctsio->kern_sg_entries = 0;
7195
7196         if (total_len < alloc_len) {
7197                 ctsio->residual = alloc_len - total_len;
7198                 ctsio->kern_data_len = total_len;
7199                 ctsio->kern_total_len = total_len;
7200         } else {
7201                 ctsio->residual = 0;
7202                 ctsio->kern_data_len = alloc_len;
7203                 ctsio->kern_total_len = alloc_len;
7204         }
7205         ctsio->kern_data_resid = 0;
7206         ctsio->kern_rel_offset = 0;
7207
7208         if (ext) {
7209                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7210                     ctsio->kern_data_ptr;
7211                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7212                 rtg_ext_ptr->format_type = 0x10;
7213                 rtg_ext_ptr->implicit_transition_time = 0;
7214                 tpg_desc = &rtg_ext_ptr->groups[0];
7215         } else {
7216                 rtg_ptr = (struct scsi_target_group_data *)
7217                     ctsio->kern_data_ptr;
7218                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7219                 tpg_desc = &rtg_ptr->groups[0];
7220         }
7221
7222         mtx_lock(&softc->ctl_lock);
7223         pg = softc->port_min / softc->port_cnt;
7224         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7225                 /* Some shelf is known to be primary. */
7226                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7227                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7228                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7229                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7230                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7231                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7232                 else
7233                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7234                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7235                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7236                 } else {
7237                         ts = os;
7238                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7239                 }
7240         } else {
7241                 /* No known primary shelf. */
7242                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7243                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7244                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7245                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7246                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7247                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7248                 } else {
7249                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7250                 }
7251         }
7252         if (shared_group) {
7253                 tpg_desc->pref_state = ts;
7254                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7255                     TPG_U_SUP | TPG_T_SUP;
7256                 scsi_ulto2b(1, tpg_desc->target_port_group);
7257                 tpg_desc->status = TPG_IMPLICIT;
7258                 pc = 0;
7259                 STAILQ_FOREACH(port, &softc->port_list, links) {
7260                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7261                                 continue;
7262                         if (!softc->is_single &&
7263                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7264                                 continue;
7265                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7266                                 continue;
7267                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7268                             relative_target_port_identifier);
7269                         pc++;
7270                 }
7271                 tpg_desc->target_port_count = pc;
7272                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7273                     &tpg_desc->descriptors[pc];
7274         }
7275         for (g = 0; g < num_ha_groups; g++) {
7276                 tpg_desc->pref_state = (g == pg) ? ts : os;
7277                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7278                     TPG_U_SUP | TPG_T_SUP;
7279                 scsi_ulto2b(2 + g, 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->targ_port < g * softc->port_cnt ||
7284                             port->targ_port >= (g + 1) * softc->port_cnt)
7285                                 continue;
7286                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7287                                 continue;
7288                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7289                                 continue;
7290                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7291                                 continue;
7292                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7293                             relative_target_port_identifier);
7294                         pc++;
7295                 }
7296                 tpg_desc->target_port_count = pc;
7297                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7298                     &tpg_desc->descriptors[pc];
7299         }
7300         mtx_unlock(&softc->ctl_lock);
7301
7302         ctl_set_success(ctsio);
7303         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7304         ctsio->be_move_done = ctl_config_move_done;
7305         ctl_datamove((union ctl_io *)ctsio);
7306         return(retval);
7307 }
7308
7309 int
7310 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7311 {
7312         struct ctl_lun *lun = CTL_LUN(ctsio);
7313         struct scsi_report_supported_opcodes *cdb;
7314         const struct ctl_cmd_entry *entry, *sentry;
7315         struct scsi_report_supported_opcodes_all *all;
7316         struct scsi_report_supported_opcodes_descr *descr;
7317         struct scsi_report_supported_opcodes_one *one;
7318         int retval;
7319         int alloc_len, total_len;
7320         int opcode, service_action, i, j, num;
7321
7322         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7323
7324         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7325         retval = CTL_RETVAL_COMPLETE;
7326
7327         opcode = cdb->requested_opcode;
7328         service_action = scsi_2btoul(cdb->requested_service_action);
7329         switch (cdb->options & RSO_OPTIONS_MASK) {
7330         case RSO_OPTIONS_ALL:
7331                 num = 0;
7332                 for (i = 0; i < 256; i++) {
7333                         entry = &ctl_cmd_table[i];
7334                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7335                                 for (j = 0; j < 32; j++) {
7336                                         sentry = &((const struct ctl_cmd_entry *)
7337                                             entry->execute)[j];
7338                                         if (ctl_cmd_applicable(
7339                                             lun->be_lun->lun_type, sentry))
7340                                                 num++;
7341                                 }
7342                         } else {
7343                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7344                                     entry))
7345                                         num++;
7346                         }
7347                 }
7348                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7349                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7350                 break;
7351         case RSO_OPTIONS_OC:
7352                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7353                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7354                                               /*sks_valid*/ 1,
7355                                               /*command*/ 1,
7356                                               /*field*/ 2,
7357                                               /*bit_valid*/ 1,
7358                                               /*bit*/ 2);
7359                         ctl_done((union ctl_io *)ctsio);
7360                         return (CTL_RETVAL_COMPLETE);
7361                 }
7362                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7363                 break;
7364         case RSO_OPTIONS_OC_SA:
7365                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7366                     service_action >= 32) {
7367                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7368                                               /*sks_valid*/ 1,
7369                                               /*command*/ 1,
7370                                               /*field*/ 2,
7371                                               /*bit_valid*/ 1,
7372                                               /*bit*/ 2);
7373                         ctl_done((union ctl_io *)ctsio);
7374                         return (CTL_RETVAL_COMPLETE);
7375                 }
7376                 /* FALLTHROUGH */
7377         case RSO_OPTIONS_OC_ASA:
7378                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7379                 break;
7380         default:
7381                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7382                                       /*sks_valid*/ 1,
7383                                       /*command*/ 1,
7384                                       /*field*/ 2,
7385                                       /*bit_valid*/ 1,
7386                                       /*bit*/ 2);
7387                 ctl_done((union ctl_io *)ctsio);
7388                 return (CTL_RETVAL_COMPLETE);
7389         }
7390
7391         alloc_len = scsi_4btoul(cdb->length);
7392
7393         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7394
7395         ctsio->kern_sg_entries = 0;
7396
7397         if (total_len < alloc_len) {
7398                 ctsio->residual = alloc_len - total_len;
7399                 ctsio->kern_data_len = total_len;
7400                 ctsio->kern_total_len = total_len;
7401         } else {
7402                 ctsio->residual = 0;
7403                 ctsio->kern_data_len = alloc_len;
7404                 ctsio->kern_total_len = alloc_len;
7405         }
7406         ctsio->kern_data_resid = 0;
7407         ctsio->kern_rel_offset = 0;
7408
7409         switch (cdb->options & RSO_OPTIONS_MASK) {
7410         case RSO_OPTIONS_ALL:
7411                 all = (struct scsi_report_supported_opcodes_all *)
7412                     ctsio->kern_data_ptr;
7413                 num = 0;
7414                 for (i = 0; i < 256; i++) {
7415                         entry = &ctl_cmd_table[i];
7416                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7417                                 for (j = 0; j < 32; j++) {
7418                                         sentry = &((const struct ctl_cmd_entry *)
7419                                             entry->execute)[j];
7420                                         if (!ctl_cmd_applicable(
7421                                             lun->be_lun->lun_type, sentry))
7422                                                 continue;
7423                                         descr = &all->descr[num++];
7424                                         descr->opcode = i;
7425                                         scsi_ulto2b(j, descr->service_action);
7426                                         descr->flags = RSO_SERVACTV;
7427                                         scsi_ulto2b(sentry->length,
7428                                             descr->cdb_length);
7429                                 }
7430                         } else {
7431                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7432                                     entry))
7433                                         continue;
7434                                 descr = &all->descr[num++];
7435                                 descr->opcode = i;
7436                                 scsi_ulto2b(0, descr->service_action);
7437                                 descr->flags = 0;
7438                                 scsi_ulto2b(entry->length, descr->cdb_length);
7439                         }
7440                 }
7441                 scsi_ulto4b(
7442                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7443                     all->length);
7444                 break;
7445         case RSO_OPTIONS_OC:
7446                 one = (struct scsi_report_supported_opcodes_one *)
7447                     ctsio->kern_data_ptr;
7448                 entry = &ctl_cmd_table[opcode];
7449                 goto fill_one;
7450         case RSO_OPTIONS_OC_SA:
7451                 one = (struct scsi_report_supported_opcodes_one *)
7452                     ctsio->kern_data_ptr;
7453                 entry = &ctl_cmd_table[opcode];
7454                 entry = &((const struct ctl_cmd_entry *)
7455                     entry->execute)[service_action];
7456 fill_one:
7457                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7458                         one->support = 3;
7459                         scsi_ulto2b(entry->length, one->cdb_length);
7460                         one->cdb_usage[0] = opcode;
7461                         memcpy(&one->cdb_usage[1], entry->usage,
7462                             entry->length - 1);
7463                 } else
7464                         one->support = 1;
7465                 break;
7466         case RSO_OPTIONS_OC_ASA:
7467                 one = (struct scsi_report_supported_opcodes_one *)
7468                     ctsio->kern_data_ptr;
7469                 entry = &ctl_cmd_table[opcode];
7470                 if (entry->flags & CTL_CMD_FLAG_SA5) {
7471                         entry = &((const struct ctl_cmd_entry *)
7472                             entry->execute)[service_action];
7473                 } else if (service_action != 0) {
7474                         one->support = 1;
7475                         break;
7476                 }
7477                 goto fill_one;
7478         }
7479
7480         ctl_set_success(ctsio);
7481         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7482         ctsio->be_move_done = ctl_config_move_done;
7483         ctl_datamove((union ctl_io *)ctsio);
7484         return(retval);
7485 }
7486
7487 int
7488 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7489 {
7490         struct scsi_report_supported_tmf *cdb;
7491         struct scsi_report_supported_tmf_ext_data *data;
7492         int retval;
7493         int alloc_len, total_len;
7494
7495         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7496
7497         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7498
7499         retval = CTL_RETVAL_COMPLETE;
7500
7501         if (cdb->options & RST_REPD)
7502                 total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7503         else
7504                 total_len = sizeof(struct scsi_report_supported_tmf_data);
7505         alloc_len = scsi_4btoul(cdb->length);
7506
7507         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7508
7509         ctsio->kern_sg_entries = 0;
7510
7511         if (total_len < alloc_len) {
7512                 ctsio->residual = alloc_len - total_len;
7513                 ctsio->kern_data_len = total_len;
7514                 ctsio->kern_total_len = total_len;
7515         } else {
7516                 ctsio->residual = 0;
7517                 ctsio->kern_data_len = alloc_len;
7518                 ctsio->kern_total_len = alloc_len;
7519         }
7520         ctsio->kern_data_resid = 0;
7521         ctsio->kern_rel_offset = 0;
7522
7523         data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7524         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7525             RST_TRS;
7526         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7527         data->length = total_len - 4;
7528
7529         ctl_set_success(ctsio);
7530         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7531         ctsio->be_move_done = ctl_config_move_done;
7532         ctl_datamove((union ctl_io *)ctsio);
7533         return (retval);
7534 }
7535
7536 int
7537 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7538 {
7539         struct scsi_report_timestamp *cdb;
7540         struct scsi_report_timestamp_data *data;
7541         struct timeval tv;
7542         int64_t timestamp;
7543         int retval;
7544         int alloc_len, total_len;
7545
7546         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7547
7548         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7549
7550         retval = CTL_RETVAL_COMPLETE;
7551
7552         total_len = sizeof(struct scsi_report_timestamp_data);
7553         alloc_len = scsi_4btoul(cdb->length);
7554
7555         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7556
7557         ctsio->kern_sg_entries = 0;
7558
7559         if (total_len < alloc_len) {
7560                 ctsio->residual = alloc_len - total_len;
7561                 ctsio->kern_data_len = total_len;
7562                 ctsio->kern_total_len = total_len;
7563         } else {
7564                 ctsio->residual = 0;
7565                 ctsio->kern_data_len = alloc_len;
7566                 ctsio->kern_total_len = alloc_len;
7567         }
7568         ctsio->kern_data_resid = 0;
7569         ctsio->kern_rel_offset = 0;
7570
7571         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7572         scsi_ulto2b(sizeof(*data) - 2, data->length);
7573         data->origin = RTS_ORIG_OUTSIDE;
7574         getmicrotime(&tv);
7575         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7576         scsi_ulto4b(timestamp >> 16, data->timestamp);
7577         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7578
7579         ctl_set_success(ctsio);
7580         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7581         ctsio->be_move_done = ctl_config_move_done;
7582         ctl_datamove((union ctl_io *)ctsio);
7583         return (retval);
7584 }
7585
7586 int
7587 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7588 {
7589         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7590         struct ctl_lun *lun = CTL_LUN(ctsio);
7591         struct scsi_per_res_in *cdb;
7592         int alloc_len, total_len = 0;
7593         /* struct scsi_per_res_in_rsrv in_data; */
7594         uint64_t key;
7595
7596         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7597
7598         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7599
7600         alloc_len = scsi_2btoul(cdb->length);
7601
7602 retry:
7603         mtx_lock(&lun->lun_lock);
7604         switch (cdb->action) {
7605         case SPRI_RK: /* read keys */
7606                 total_len = sizeof(struct scsi_per_res_in_keys) +
7607                         lun->pr_key_count *
7608                         sizeof(struct scsi_per_res_key);
7609                 break;
7610         case SPRI_RR: /* read reservation */
7611                 if (lun->flags & CTL_LUN_PR_RESERVED)
7612                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7613                 else
7614                         total_len = sizeof(struct scsi_per_res_in_header);
7615                 break;
7616         case SPRI_RC: /* report capabilities */
7617                 total_len = sizeof(struct scsi_per_res_cap);
7618                 break;
7619         case SPRI_RS: /* read full status */
7620                 total_len = sizeof(struct scsi_per_res_in_header) +
7621                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7622                     lun->pr_key_count;
7623                 break;
7624         default:
7625                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7626         }
7627         mtx_unlock(&lun->lun_lock);
7628
7629         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7630
7631         if (total_len < alloc_len) {
7632                 ctsio->residual = alloc_len - total_len;
7633                 ctsio->kern_data_len = total_len;
7634                 ctsio->kern_total_len = total_len;
7635         } else {
7636                 ctsio->residual = 0;
7637                 ctsio->kern_data_len = alloc_len;
7638                 ctsio->kern_total_len = alloc_len;
7639         }
7640
7641         ctsio->kern_data_resid = 0;
7642         ctsio->kern_rel_offset = 0;
7643         ctsio->kern_sg_entries = 0;
7644
7645         mtx_lock(&lun->lun_lock);
7646         switch (cdb->action) {
7647         case SPRI_RK: { // read keys
7648         struct scsi_per_res_in_keys *res_keys;
7649                 int i, key_count;
7650
7651                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7652
7653                 /*
7654                  * We had to drop the lock to allocate our buffer, which
7655                  * leaves time for someone to come in with another
7656                  * persistent reservation.  (That is unlikely, though,
7657                  * since this should be the only persistent reservation
7658                  * command active right now.)
7659                  */
7660                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7661                     (lun->pr_key_count *
7662                      sizeof(struct scsi_per_res_key)))){
7663                         mtx_unlock(&lun->lun_lock);
7664                         free(ctsio->kern_data_ptr, M_CTL);
7665                         printf("%s: reservation length changed, retrying\n",
7666                                __func__);
7667                         goto retry;
7668                 }
7669
7670                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7671
7672                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7673                              lun->pr_key_count, res_keys->header.length);
7674
7675                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7676                         if ((key = ctl_get_prkey(lun, i)) == 0)
7677                                 continue;
7678
7679                         /*
7680                          * We used lun->pr_key_count to calculate the
7681                          * size to allocate.  If it turns out the number of
7682                          * initiators with the registered flag set is
7683                          * larger than that (i.e. they haven't been kept in
7684                          * sync), we've got a problem.
7685                          */
7686                         if (key_count >= lun->pr_key_count) {
7687                                 key_count++;
7688                                 continue;
7689                         }
7690                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7691                         key_count++;
7692                 }
7693                 break;
7694         }
7695         case SPRI_RR: { // read reservation
7696                 struct scsi_per_res_in_rsrv *res;
7697                 int tmp_len, header_only;
7698
7699                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7700
7701                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7702
7703                 if (lun->flags & CTL_LUN_PR_RESERVED)
7704                 {
7705                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7706                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7707                                     res->header.length);
7708                         header_only = 0;
7709                 } else {
7710                         tmp_len = sizeof(struct scsi_per_res_in_header);
7711                         scsi_ulto4b(0, res->header.length);
7712                         header_only = 1;
7713                 }
7714
7715                 /*
7716                  * We had to drop the lock to allocate our buffer, which
7717                  * leaves time for someone to come in with another
7718                  * persistent reservation.  (That is unlikely, though,
7719                  * since this should be the only persistent reservation
7720                  * command active right now.)
7721                  */
7722                 if (tmp_len != total_len) {
7723                         mtx_unlock(&lun->lun_lock);
7724                         free(ctsio->kern_data_ptr, M_CTL);
7725                         printf("%s: reservation status changed, retrying\n",
7726                                __func__);
7727                         goto retry;
7728                 }
7729
7730                 /*
7731                  * No reservation held, so we're done.
7732                  */
7733                 if (header_only != 0)
7734                         break;
7735
7736                 /*
7737                  * If the registration is an All Registrants type, the key
7738                  * is 0, since it doesn't really matter.
7739                  */
7740                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7741                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7742                             res->data.reservation);
7743                 }
7744                 res->data.scopetype = lun->pr_res_type;
7745                 break;
7746         }
7747         case SPRI_RC:     //report capabilities
7748         {
7749                 struct scsi_per_res_cap *res_cap;
7750                 uint16_t type_mask;
7751
7752                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7753                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7754                 res_cap->flags1 = SPRI_CRH;
7755                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7756                 type_mask = SPRI_TM_WR_EX_AR |
7757                             SPRI_TM_EX_AC_RO |
7758                             SPRI_TM_WR_EX_RO |
7759                             SPRI_TM_EX_AC |
7760                             SPRI_TM_WR_EX |
7761                             SPRI_TM_EX_AC_AR;
7762                 scsi_ulto2b(type_mask, res_cap->type_mask);
7763                 break;
7764         }
7765         case SPRI_RS: { // read full status
7766                 struct scsi_per_res_in_full *res_status;
7767                 struct scsi_per_res_in_full_desc *res_desc;
7768                 struct ctl_port *port;
7769                 int i, len;
7770
7771                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7772
7773                 /*
7774                  * We had to drop the lock to allocate our buffer, which
7775                  * leaves time for someone to come in with another
7776                  * persistent reservation.  (That is unlikely, though,
7777                  * since this should be the only persistent reservation
7778                  * command active right now.)
7779                  */
7780                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7781                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7782                      lun->pr_key_count)){
7783                         mtx_unlock(&lun->lun_lock);
7784                         free(ctsio->kern_data_ptr, M_CTL);
7785                         printf("%s: reservation length changed, retrying\n",
7786                                __func__);
7787                         goto retry;
7788                 }
7789
7790                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7791
7792                 res_desc = &res_status->desc[0];
7793                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7794                         if ((key = ctl_get_prkey(lun, i)) == 0)
7795                                 continue;
7796
7797                         scsi_u64to8b(key, res_desc->res_key.key);
7798                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7799                             (lun->pr_res_idx == i ||
7800                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7801                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7802                                 res_desc->scopetype = lun->pr_res_type;
7803                         }
7804                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7805                             res_desc->rel_trgt_port_id);
7806                         len = 0;
7807                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7808                         if (port != NULL)
7809                                 len = ctl_create_iid(port,
7810                                     i % CTL_MAX_INIT_PER_PORT,
7811                                     res_desc->transport_id);
7812                         scsi_ulto4b(len, res_desc->additional_length);
7813                         res_desc = (struct scsi_per_res_in_full_desc *)
7814                             &res_desc->transport_id[len];
7815                 }
7816                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7817                     res_status->header.length);
7818                 break;
7819         }
7820         default:
7821                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7822         }
7823         mtx_unlock(&lun->lun_lock);
7824
7825         ctl_set_success(ctsio);
7826         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7827         ctsio->be_move_done = ctl_config_move_done;
7828         ctl_datamove((union ctl_io *)ctsio);
7829         return (CTL_RETVAL_COMPLETE);
7830 }
7831
7832 /*
7833  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7834  * it should return.
7835  */
7836 static int
7837 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7838                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7839                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7840                 struct scsi_per_res_out_parms* param)
7841 {
7842         union ctl_ha_msg persis_io;
7843         int i;
7844
7845         mtx_lock(&lun->lun_lock);
7846         if (sa_res_key == 0) {
7847                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7848                         /* validate scope and type */
7849                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7850                              SPR_LU_SCOPE) {
7851                                 mtx_unlock(&lun->lun_lock);
7852                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7853                                                       /*sks_valid*/ 1,
7854                                                       /*command*/ 1,
7855                                                       /*field*/ 2,
7856                                                       /*bit_valid*/ 1,
7857                                                       /*bit*/ 4);
7858                                 ctl_done((union ctl_io *)ctsio);
7859                                 return (1);
7860                         }
7861
7862                         if (type>8 || type==2 || type==4 || type==0) {
7863                                 mtx_unlock(&lun->lun_lock);
7864                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7865                                                       /*sks_valid*/ 1,
7866                                                       /*command*/ 1,
7867                                                       /*field*/ 2,
7868                                                       /*bit_valid*/ 1,
7869                                                       /*bit*/ 0);
7870                                 ctl_done((union ctl_io *)ctsio);
7871                                 return (1);
7872                         }
7873
7874                         /*
7875                          * Unregister everybody else and build UA for
7876                          * them
7877                          */
7878                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7879                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7880                                         continue;
7881
7882                                 ctl_clr_prkey(lun, i);
7883                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7884                         }
7885                         lun->pr_key_count = 1;
7886                         lun->pr_res_type = type;
7887                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7888                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7889                                 lun->pr_res_idx = residx;
7890                         lun->pr_generation++;
7891                         mtx_unlock(&lun->lun_lock);
7892
7893                         /* send msg to other side */
7894                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7895                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7896                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7897                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7898                         persis_io.pr.pr_info.res_type = type;
7899                         memcpy(persis_io.pr.pr_info.sa_res_key,
7900                                param->serv_act_res_key,
7901                                sizeof(param->serv_act_res_key));
7902                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7903                             sizeof(persis_io.pr), M_WAITOK);
7904                 } else {
7905                         /* not all registrants */
7906                         mtx_unlock(&lun->lun_lock);
7907                         free(ctsio->kern_data_ptr, M_CTL);
7908                         ctl_set_invalid_field(ctsio,
7909                                               /*sks_valid*/ 1,
7910                                               /*command*/ 0,
7911                                               /*field*/ 8,
7912                                               /*bit_valid*/ 0,
7913                                               /*bit*/ 0);
7914                         ctl_done((union ctl_io *)ctsio);
7915                         return (1);
7916                 }
7917         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7918                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7919                 int found = 0;
7920
7921                 if (res_key == sa_res_key) {
7922                         /* special case */
7923                         /*
7924                          * The spec implies this is not good but doesn't
7925                          * say what to do. There are two choices either
7926                          * generate a res conflict or check condition
7927                          * with illegal field in parameter data. Since
7928                          * that is what is done when the sa_res_key is
7929                          * zero I'll take that approach since this has
7930                          * to do with the sa_res_key.
7931                          */
7932                         mtx_unlock(&lun->lun_lock);
7933                         free(ctsio->kern_data_ptr, M_CTL);
7934                         ctl_set_invalid_field(ctsio,
7935                                               /*sks_valid*/ 1,
7936                                               /*command*/ 0,
7937                                               /*field*/ 8,
7938                                               /*bit_valid*/ 0,
7939                                               /*bit*/ 0);
7940                         ctl_done((union ctl_io *)ctsio);
7941                         return (1);
7942                 }
7943
7944                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7945                         if (ctl_get_prkey(lun, i) != sa_res_key)
7946                                 continue;
7947
7948                         found = 1;
7949                         ctl_clr_prkey(lun, i);
7950                         lun->pr_key_count--;
7951                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7952                 }
7953                 if (!found) {
7954                         mtx_unlock(&lun->lun_lock);
7955                         free(ctsio->kern_data_ptr, M_CTL);
7956                         ctl_set_reservation_conflict(ctsio);
7957                         ctl_done((union ctl_io *)ctsio);
7958                         return (CTL_RETVAL_COMPLETE);
7959                 }
7960                 lun->pr_generation++;
7961                 mtx_unlock(&lun->lun_lock);
7962
7963                 /* send msg to other side */
7964                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7965                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7966                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7967                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7968                 persis_io.pr.pr_info.res_type = type;
7969                 memcpy(persis_io.pr.pr_info.sa_res_key,
7970                        param->serv_act_res_key,
7971                        sizeof(param->serv_act_res_key));
7972                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7973                     sizeof(persis_io.pr), M_WAITOK);
7974         } else {
7975                 /* Reserved but not all registrants */
7976                 /* sa_res_key is res holder */
7977                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7978                         /* validate scope and type */
7979                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7980                              SPR_LU_SCOPE) {
7981                                 mtx_unlock(&lun->lun_lock);
7982                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7983                                                       /*sks_valid*/ 1,
7984                                                       /*command*/ 1,
7985                                                       /*field*/ 2,
7986                                                       /*bit_valid*/ 1,
7987                                                       /*bit*/ 4);
7988                                 ctl_done((union ctl_io *)ctsio);
7989                                 return (1);
7990                         }
7991
7992                         if (type>8 || type==2 || type==4 || type==0) {
7993                                 mtx_unlock(&lun->lun_lock);
7994                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7995                                                       /*sks_valid*/ 1,
7996                                                       /*command*/ 1,
7997                                                       /*field*/ 2,
7998                                                       /*bit_valid*/ 1,
7999                                                       /*bit*/ 0);
8000                                 ctl_done((union ctl_io *)ctsio);
8001                                 return (1);
8002                         }
8003
8004                         /*
8005                          * Do the following:
8006                          * if sa_res_key != res_key remove all
8007                          * registrants w/sa_res_key and generate UA
8008                          * for these registrants(Registrations
8009                          * Preempted) if it wasn't an exclusive
8010                          * reservation generate UA(Reservations
8011                          * Preempted) for all other registered nexuses
8012                          * if the type has changed. Establish the new
8013                          * reservation and holder. If res_key and
8014                          * sa_res_key are the same do the above
8015                          * except don't unregister the res holder.
8016                          */
8017
8018                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8019                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8020                                         continue;
8021
8022                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
8023                                         ctl_clr_prkey(lun, i);
8024                                         lun->pr_key_count--;
8025                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8026                                 } else if (type != lun->pr_res_type &&
8027                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8028                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8029                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8030                                 }
8031                         }
8032                         lun->pr_res_type = type;
8033                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8034                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8035                                 lun->pr_res_idx = residx;
8036                         else
8037                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8038                         lun->pr_generation++;
8039                         mtx_unlock(&lun->lun_lock);
8040
8041                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8042                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8043                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8044                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8045                         persis_io.pr.pr_info.res_type = type;
8046                         memcpy(persis_io.pr.pr_info.sa_res_key,
8047                                param->serv_act_res_key,
8048                                sizeof(param->serv_act_res_key));
8049                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8050                             sizeof(persis_io.pr), M_WAITOK);
8051                 } else {
8052                         /*
8053                          * sa_res_key is not the res holder just
8054                          * remove registrants
8055                          */
8056                         int found=0;
8057
8058                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8059                                 if (sa_res_key != ctl_get_prkey(lun, i))
8060                                         continue;
8061
8062                                 found = 1;
8063                                 ctl_clr_prkey(lun, i);
8064                                 lun->pr_key_count--;
8065                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8066                         }
8067
8068                         if (!found) {
8069                                 mtx_unlock(&lun->lun_lock);
8070                                 free(ctsio->kern_data_ptr, M_CTL);
8071                                 ctl_set_reservation_conflict(ctsio);
8072                                 ctl_done((union ctl_io *)ctsio);
8073                                 return (1);
8074                         }
8075                         lun->pr_generation++;
8076                         mtx_unlock(&lun->lun_lock);
8077
8078                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8079                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8080                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8081                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8082                         persis_io.pr.pr_info.res_type = type;
8083                         memcpy(persis_io.pr.pr_info.sa_res_key,
8084                                param->serv_act_res_key,
8085                                sizeof(param->serv_act_res_key));
8086                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8087                             sizeof(persis_io.pr), M_WAITOK);
8088                 }
8089         }
8090         return (0);
8091 }
8092
8093 static void
8094 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8095 {
8096         uint64_t sa_res_key;
8097         int i;
8098
8099         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8100
8101         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8102          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8103          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8104                 if (sa_res_key == 0) {
8105                         /*
8106                          * Unregister everybody else and build UA for
8107                          * them
8108                          */
8109                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8110                                 if (i == msg->pr.pr_info.residx ||
8111                                     ctl_get_prkey(lun, i) == 0)
8112                                         continue;
8113
8114                                 ctl_clr_prkey(lun, i);
8115                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8116                         }
8117
8118                         lun->pr_key_count = 1;
8119                         lun->pr_res_type = msg->pr.pr_info.res_type;
8120                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8121                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8122                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8123                 } else {
8124                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8125                                 if (sa_res_key == ctl_get_prkey(lun, i))
8126                                         continue;
8127
8128                                 ctl_clr_prkey(lun, i);
8129                                 lun->pr_key_count--;
8130                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8131                         }
8132                 }
8133         } else {
8134                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8135                         if (i == msg->pr.pr_info.residx ||
8136                             ctl_get_prkey(lun, i) == 0)
8137                                 continue;
8138
8139                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8140                                 ctl_clr_prkey(lun, i);
8141                                 lun->pr_key_count--;
8142                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8143                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8144                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8145                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8146                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8147                         }
8148                 }
8149                 lun->pr_res_type = msg->pr.pr_info.res_type;
8150                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8151                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8152                         lun->pr_res_idx = msg->pr.pr_info.residx;
8153                 else
8154                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8155         }
8156         lun->pr_generation++;
8157
8158 }
8159
8160
8161 int
8162 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8163 {
8164         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8165         struct ctl_lun *lun = CTL_LUN(ctsio);
8166         int retval;
8167         u_int32_t param_len;
8168         struct scsi_per_res_out *cdb;
8169         struct scsi_per_res_out_parms* param;
8170         uint32_t residx;
8171         uint64_t res_key, sa_res_key, key;
8172         uint8_t type;
8173         union ctl_ha_msg persis_io;
8174         int    i;
8175
8176         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8177
8178         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8179         retval = CTL_RETVAL_COMPLETE;
8180
8181         /*
8182          * We only support whole-LUN scope.  The scope & type are ignored for
8183          * register, register and ignore existing key and clear.
8184          * We sometimes ignore scope and type on preempts too!!
8185          * Verify reservation type here as well.
8186          */
8187         type = cdb->scope_type & SPR_TYPE_MASK;
8188         if ((cdb->action == SPRO_RESERVE)
8189          || (cdb->action == SPRO_RELEASE)) {
8190                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8191                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8192                                               /*sks_valid*/ 1,
8193                                               /*command*/ 1,
8194                                               /*field*/ 2,
8195                                               /*bit_valid*/ 1,
8196                                               /*bit*/ 4);
8197                         ctl_done((union ctl_io *)ctsio);
8198                         return (CTL_RETVAL_COMPLETE);
8199                 }
8200
8201                 if (type>8 || type==2 || type==4 || type==0) {
8202                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8203                                               /*sks_valid*/ 1,
8204                                               /*command*/ 1,
8205                                               /*field*/ 2,
8206                                               /*bit_valid*/ 1,
8207                                               /*bit*/ 0);
8208                         ctl_done((union ctl_io *)ctsio);
8209                         return (CTL_RETVAL_COMPLETE);
8210                 }
8211         }
8212
8213         param_len = scsi_4btoul(cdb->length);
8214
8215         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8216                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8217                 ctsio->kern_data_len = param_len;
8218                 ctsio->kern_total_len = param_len;
8219                 ctsio->kern_data_resid = 0;
8220                 ctsio->kern_rel_offset = 0;
8221                 ctsio->kern_sg_entries = 0;
8222                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8223                 ctsio->be_move_done = ctl_config_move_done;
8224                 ctl_datamove((union ctl_io *)ctsio);
8225
8226                 return (CTL_RETVAL_COMPLETE);
8227         }
8228
8229         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8230
8231         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8232         res_key = scsi_8btou64(param->res_key.key);
8233         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8234
8235         /*
8236          * Validate the reservation key here except for SPRO_REG_IGNO
8237          * This must be done for all other service actions
8238          */
8239         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8240                 mtx_lock(&lun->lun_lock);
8241                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8242                         if (res_key != key) {
8243                                 /*
8244                                  * The current key passed in doesn't match
8245                                  * the one the initiator previously
8246                                  * registered.
8247                                  */
8248                                 mtx_unlock(&lun->lun_lock);
8249                                 free(ctsio->kern_data_ptr, M_CTL);
8250                                 ctl_set_reservation_conflict(ctsio);
8251                                 ctl_done((union ctl_io *)ctsio);
8252                                 return (CTL_RETVAL_COMPLETE);
8253                         }
8254                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8255                         /*
8256                          * We are not registered
8257                          */
8258                         mtx_unlock(&lun->lun_lock);
8259                         free(ctsio->kern_data_ptr, M_CTL);
8260                         ctl_set_reservation_conflict(ctsio);
8261                         ctl_done((union ctl_io *)ctsio);
8262                         return (CTL_RETVAL_COMPLETE);
8263                 } else if (res_key != 0) {
8264                         /*
8265                          * We are not registered and trying to register but
8266                          * the register key isn't zero.
8267                          */
8268                         mtx_unlock(&lun->lun_lock);
8269                         free(ctsio->kern_data_ptr, M_CTL);
8270                         ctl_set_reservation_conflict(ctsio);
8271                         ctl_done((union ctl_io *)ctsio);
8272                         return (CTL_RETVAL_COMPLETE);
8273                 }
8274                 mtx_unlock(&lun->lun_lock);
8275         }
8276
8277         switch (cdb->action & SPRO_ACTION_MASK) {
8278         case SPRO_REGISTER:
8279         case SPRO_REG_IGNO: {
8280
8281 #if 0
8282                 printf("Registration received\n");
8283 #endif
8284
8285                 /*
8286                  * We don't support any of these options, as we report in
8287                  * the read capabilities request (see
8288                  * ctl_persistent_reserve_in(), above).
8289                  */
8290                 if ((param->flags & SPR_SPEC_I_PT)
8291                  || (param->flags & SPR_ALL_TG_PT)
8292                  || (param->flags & SPR_APTPL)) {
8293                         int bit_ptr;
8294
8295                         if (param->flags & SPR_APTPL)
8296                                 bit_ptr = 0;
8297                         else if (param->flags & SPR_ALL_TG_PT)
8298                                 bit_ptr = 2;
8299                         else /* SPR_SPEC_I_PT */
8300                                 bit_ptr = 3;
8301
8302                         free(ctsio->kern_data_ptr, M_CTL);
8303                         ctl_set_invalid_field(ctsio,
8304                                               /*sks_valid*/ 1,
8305                                               /*command*/ 0,
8306                                               /*field*/ 20,
8307                                               /*bit_valid*/ 1,
8308                                               /*bit*/ bit_ptr);
8309                         ctl_done((union ctl_io *)ctsio);
8310                         return (CTL_RETVAL_COMPLETE);
8311                 }
8312
8313                 mtx_lock(&lun->lun_lock);
8314
8315                 /*
8316                  * The initiator wants to clear the
8317                  * key/unregister.
8318                  */
8319                 if (sa_res_key == 0) {
8320                         if ((res_key == 0
8321                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8322                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8323                           && ctl_get_prkey(lun, residx) == 0)) {
8324                                 mtx_unlock(&lun->lun_lock);
8325                                 goto done;
8326                         }
8327
8328                         ctl_clr_prkey(lun, residx);
8329                         lun->pr_key_count--;
8330
8331                         if (residx == lun->pr_res_idx) {
8332                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8333                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8334
8335                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8336                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8337                                     lun->pr_key_count) {
8338                                         /*
8339                                          * If the reservation is a registrants
8340                                          * only type we need to generate a UA
8341                                          * for other registered inits.  The
8342                                          * sense code should be RESERVATIONS
8343                                          * RELEASED
8344                                          */
8345
8346                                         for (i = softc->init_min; i < softc->init_max; i++){
8347                                                 if (ctl_get_prkey(lun, i) == 0)
8348                                                         continue;
8349                                                 ctl_est_ua(lun, i,
8350                                                     CTL_UA_RES_RELEASE);
8351                                         }
8352                                 }
8353                                 lun->pr_res_type = 0;
8354                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8355                                 if (lun->pr_key_count==0) {
8356                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8357                                         lun->pr_res_type = 0;
8358                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8359                                 }
8360                         }
8361                         lun->pr_generation++;
8362                         mtx_unlock(&lun->lun_lock);
8363
8364                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8365                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8366                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8367                         persis_io.pr.pr_info.residx = residx;
8368                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8369                             sizeof(persis_io.pr), M_WAITOK);
8370                 } else /* sa_res_key != 0 */ {
8371
8372                         /*
8373                          * If we aren't registered currently then increment
8374                          * the key count and set the registered flag.
8375                          */
8376                         ctl_alloc_prkey(lun, residx);
8377                         if (ctl_get_prkey(lun, residx) == 0)
8378                                 lun->pr_key_count++;
8379                         ctl_set_prkey(lun, residx, sa_res_key);
8380                         lun->pr_generation++;
8381                         mtx_unlock(&lun->lun_lock);
8382
8383                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8384                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8385                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8386                         persis_io.pr.pr_info.residx = residx;
8387                         memcpy(persis_io.pr.pr_info.sa_res_key,
8388                                param->serv_act_res_key,
8389                                sizeof(param->serv_act_res_key));
8390                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8391                             sizeof(persis_io.pr), M_WAITOK);
8392                 }
8393
8394                 break;
8395         }
8396         case SPRO_RESERVE:
8397 #if 0
8398                 printf("Reserve executed type %d\n", type);
8399 #endif
8400                 mtx_lock(&lun->lun_lock);
8401                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8402                         /*
8403                          * if this isn't the reservation holder and it's
8404                          * not a "all registrants" type or if the type is
8405                          * different then we have a conflict
8406                          */
8407                         if ((lun->pr_res_idx != residx
8408                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8409                          || lun->pr_res_type != type) {
8410                                 mtx_unlock(&lun->lun_lock);
8411                                 free(ctsio->kern_data_ptr, M_CTL);
8412                                 ctl_set_reservation_conflict(ctsio);
8413                                 ctl_done((union ctl_io *)ctsio);
8414                                 return (CTL_RETVAL_COMPLETE);
8415                         }
8416                         mtx_unlock(&lun->lun_lock);
8417                 } else /* create a reservation */ {
8418                         /*
8419                          * If it's not an "all registrants" type record
8420                          * reservation holder
8421                          */
8422                         if (type != SPR_TYPE_WR_EX_AR
8423                          && type != SPR_TYPE_EX_AC_AR)
8424                                 lun->pr_res_idx = residx; /* Res holder */
8425                         else
8426                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8427
8428                         lun->flags |= CTL_LUN_PR_RESERVED;
8429                         lun->pr_res_type = type;
8430
8431                         mtx_unlock(&lun->lun_lock);
8432
8433                         /* send msg to other side */
8434                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8435                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8436                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8437                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8438                         persis_io.pr.pr_info.res_type = type;
8439                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8440                             sizeof(persis_io.pr), M_WAITOK);
8441                 }
8442                 break;
8443
8444         case SPRO_RELEASE:
8445                 mtx_lock(&lun->lun_lock);
8446                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8447                         /* No reservation exists return good status */
8448                         mtx_unlock(&lun->lun_lock);
8449                         goto done;
8450                 }
8451                 /*
8452                  * Is this nexus a reservation holder?
8453                  */
8454                 if (lun->pr_res_idx != residx
8455                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8456                         /*
8457                          * not a res holder return good status but
8458                          * do nothing
8459                          */
8460                         mtx_unlock(&lun->lun_lock);
8461                         goto done;
8462                 }
8463
8464                 if (lun->pr_res_type != type) {
8465                         mtx_unlock(&lun->lun_lock);
8466                         free(ctsio->kern_data_ptr, M_CTL);
8467                         ctl_set_illegal_pr_release(ctsio);
8468                         ctl_done((union ctl_io *)ctsio);
8469                         return (CTL_RETVAL_COMPLETE);
8470                 }
8471
8472                 /* okay to release */
8473                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8474                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8475                 lun->pr_res_type = 0;
8476
8477                 /*
8478                  * If this isn't an exclusive access reservation and NUAR
8479                  * is not set, generate UA for all other registrants.
8480                  */
8481                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8482                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8483                         for (i = softc->init_min; i < softc->init_max; i++) {
8484                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8485                                         continue;
8486                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8487                         }
8488                 }
8489                 mtx_unlock(&lun->lun_lock);
8490
8491                 /* Send msg to other side */
8492                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8493                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8494                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8495                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8496                      sizeof(persis_io.pr), M_WAITOK);
8497                 break;
8498
8499         case SPRO_CLEAR:
8500                 /* send msg to other side */
8501
8502                 mtx_lock(&lun->lun_lock);
8503                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8504                 lun->pr_res_type = 0;
8505                 lun->pr_key_count = 0;
8506                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8507
8508                 ctl_clr_prkey(lun, residx);
8509                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8510                         if (ctl_get_prkey(lun, i) != 0) {
8511                                 ctl_clr_prkey(lun, i);
8512                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8513                         }
8514                 lun->pr_generation++;
8515                 mtx_unlock(&lun->lun_lock);
8516
8517                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8518                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8519                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8520                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8521                      sizeof(persis_io.pr), M_WAITOK);
8522                 break;
8523
8524         case SPRO_PREEMPT:
8525         case SPRO_PRE_ABO: {
8526                 int nretval;
8527
8528                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8529                                           residx, ctsio, cdb, param);
8530                 if (nretval != 0)
8531                         return (CTL_RETVAL_COMPLETE);
8532                 break;
8533         }
8534         default:
8535                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8536         }
8537
8538 done:
8539         free(ctsio->kern_data_ptr, M_CTL);
8540         ctl_set_success(ctsio);
8541         ctl_done((union ctl_io *)ctsio);
8542
8543         return (retval);
8544 }
8545
8546 /*
8547  * This routine is for handling a message from the other SC pertaining to
8548  * persistent reserve out. All the error checking will have been done
8549  * so only perorming the action need be done here to keep the two
8550  * in sync.
8551  */
8552 static void
8553 ctl_hndl_per_res_out_on_other_sc(union ctl_io *io)
8554 {
8555         struct ctl_softc *softc = CTL_SOFTC(io);
8556         union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg;
8557         struct ctl_lun *lun;
8558         int i;
8559         uint32_t residx, targ_lun;
8560
8561         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8562         mtx_lock(&softc->ctl_lock);
8563         if (targ_lun >= CTL_MAX_LUNS ||
8564             (lun = softc->ctl_luns[targ_lun]) == NULL) {
8565                 mtx_unlock(&softc->ctl_lock);
8566                 return;
8567         }
8568         mtx_lock(&lun->lun_lock);
8569         mtx_unlock(&softc->ctl_lock);
8570         if (lun->flags & CTL_LUN_DISABLED) {
8571                 mtx_unlock(&lun->lun_lock);
8572                 return;
8573         }
8574         residx = ctl_get_initindex(&msg->hdr.nexus);
8575         switch(msg->pr.pr_info.action) {
8576         case CTL_PR_REG_KEY:
8577                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8578                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8579                         lun->pr_key_count++;
8580                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8581                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8582                 lun->pr_generation++;
8583                 break;
8584
8585         case CTL_PR_UNREG_KEY:
8586                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8587                 lun->pr_key_count--;
8588
8589                 /* XXX Need to see if the reservation has been released */
8590                 /* if so do we need to generate UA? */
8591                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8592                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8593                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8594
8595                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8596                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8597                             lun->pr_key_count) {
8598                                 /*
8599                                  * If the reservation is a registrants
8600                                  * only type we need to generate a UA
8601                                  * for other registered inits.  The
8602                                  * sense code should be RESERVATIONS
8603                                  * RELEASED
8604                                  */
8605
8606                                 for (i = softc->init_min; i < softc->init_max; i++) {
8607                                         if (ctl_get_prkey(lun, i) == 0)
8608                                                 continue;
8609
8610                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8611                                 }
8612                         }
8613                         lun->pr_res_type = 0;
8614                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8615                         if (lun->pr_key_count==0) {
8616                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8617                                 lun->pr_res_type = 0;
8618                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8619                         }
8620                 }
8621                 lun->pr_generation++;
8622                 break;
8623
8624         case CTL_PR_RESERVE:
8625                 lun->flags |= CTL_LUN_PR_RESERVED;
8626                 lun->pr_res_type = msg->pr.pr_info.res_type;
8627                 lun->pr_res_idx = msg->pr.pr_info.residx;
8628
8629                 break;
8630
8631         case CTL_PR_RELEASE:
8632                 /*
8633                  * If this isn't an exclusive access reservation and NUAR
8634                  * is not set, generate UA for all other registrants.
8635                  */
8636                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8637                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8638                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8639                         for (i = softc->init_min; i < softc->init_max; i++)
8640                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8641                                         continue;
8642                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8643                 }
8644
8645                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8646                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8647                 lun->pr_res_type = 0;
8648                 break;
8649
8650         case CTL_PR_PREEMPT:
8651                 ctl_pro_preempt_other(lun, msg);
8652                 break;
8653         case CTL_PR_CLEAR:
8654                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8655                 lun->pr_res_type = 0;
8656                 lun->pr_key_count = 0;
8657                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8658
8659                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8660                         if (ctl_get_prkey(lun, i) == 0)
8661                                 continue;
8662                         ctl_clr_prkey(lun, i);
8663                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8664                 }
8665                 lun->pr_generation++;
8666                 break;
8667         }
8668
8669         mtx_unlock(&lun->lun_lock);
8670 }
8671
8672 int
8673 ctl_read_write(struct ctl_scsiio *ctsio)
8674 {
8675         struct ctl_lun *lun = CTL_LUN(ctsio);
8676         struct ctl_lba_len_flags *lbalen;
8677         uint64_t lba;
8678         uint32_t num_blocks;
8679         int flags, retval;
8680         int isread;
8681
8682         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8683
8684         flags = 0;
8685         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8686               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8687         switch (ctsio->cdb[0]) {
8688         case READ_6:
8689         case WRITE_6: {
8690                 struct scsi_rw_6 *cdb;
8691
8692                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8693
8694                 lba = scsi_3btoul(cdb->addr);
8695                 /* only 5 bits are valid in the most significant address byte */
8696                 lba &= 0x1fffff;
8697                 num_blocks = cdb->length;
8698                 /*
8699                  * This is correct according to SBC-2.
8700                  */
8701                 if (num_blocks == 0)
8702                         num_blocks = 256;
8703                 break;
8704         }
8705         case READ_10:
8706         case WRITE_10: {
8707                 struct scsi_rw_10 *cdb;
8708
8709                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8710                 if (cdb->byte2 & SRW10_FUA)
8711                         flags |= CTL_LLF_FUA;
8712                 if (cdb->byte2 & SRW10_DPO)
8713                         flags |= CTL_LLF_DPO;
8714                 lba = scsi_4btoul(cdb->addr);
8715                 num_blocks = scsi_2btoul(cdb->length);
8716                 break;
8717         }
8718         case WRITE_VERIFY_10: {
8719                 struct scsi_write_verify_10 *cdb;
8720
8721                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8722                 flags |= CTL_LLF_FUA;
8723                 if (cdb->byte2 & SWV_DPO)
8724                         flags |= CTL_LLF_DPO;
8725                 lba = scsi_4btoul(cdb->addr);
8726                 num_blocks = scsi_2btoul(cdb->length);
8727                 break;
8728         }
8729         case READ_12:
8730         case WRITE_12: {
8731                 struct scsi_rw_12 *cdb;
8732
8733                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8734                 if (cdb->byte2 & SRW12_FUA)
8735                         flags |= CTL_LLF_FUA;
8736                 if (cdb->byte2 & SRW12_DPO)
8737                         flags |= CTL_LLF_DPO;
8738                 lba = scsi_4btoul(cdb->addr);
8739                 num_blocks = scsi_4btoul(cdb->length);
8740                 break;
8741         }
8742         case WRITE_VERIFY_12: {
8743                 struct scsi_write_verify_12 *cdb;
8744
8745                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8746                 flags |= CTL_LLF_FUA;
8747                 if (cdb->byte2 & SWV_DPO)
8748                         flags |= CTL_LLF_DPO;
8749                 lba = scsi_4btoul(cdb->addr);
8750                 num_blocks = scsi_4btoul(cdb->length);
8751                 break;
8752         }
8753         case READ_16:
8754         case WRITE_16: {
8755                 struct scsi_rw_16 *cdb;
8756
8757                 cdb = (struct scsi_rw_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_4btoul(cdb->length);
8764                 break;
8765         }
8766         case WRITE_ATOMIC_16: {
8767                 struct scsi_write_atomic_16 *cdb;
8768
8769                 if (lun->be_lun->atomicblock == 0) {
8770                         ctl_set_invalid_opcode(ctsio);
8771                         ctl_done((union ctl_io *)ctsio);
8772                         return (CTL_RETVAL_COMPLETE);
8773                 }
8774
8775                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8776                 if (cdb->byte2 & SRW12_FUA)
8777                         flags |= CTL_LLF_FUA;
8778                 if (cdb->byte2 & SRW12_DPO)
8779                         flags |= CTL_LLF_DPO;
8780                 lba = scsi_8btou64(cdb->addr);
8781                 num_blocks = scsi_2btoul(cdb->length);
8782                 if (num_blocks > lun->be_lun->atomicblock) {
8783                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8784                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8785                             /*bit*/ 0);
8786                         ctl_done((union ctl_io *)ctsio);
8787                         return (CTL_RETVAL_COMPLETE);
8788                 }
8789                 break;
8790         }
8791         case WRITE_VERIFY_16: {
8792                 struct scsi_write_verify_16 *cdb;
8793
8794                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8795                 flags |= CTL_LLF_FUA;
8796                 if (cdb->byte2 & SWV_DPO)
8797                         flags |= CTL_LLF_DPO;
8798                 lba = scsi_8btou64(cdb->addr);
8799                 num_blocks = scsi_4btoul(cdb->length);
8800                 break;
8801         }
8802         default:
8803                 /*
8804                  * We got a command we don't support.  This shouldn't
8805                  * happen, commands should be filtered out above us.
8806                  */
8807                 ctl_set_invalid_opcode(ctsio);
8808                 ctl_done((union ctl_io *)ctsio);
8809
8810                 return (CTL_RETVAL_COMPLETE);
8811                 break; /* NOTREACHED */
8812         }
8813
8814         /*
8815          * The first check is to make sure we're in bounds, the second
8816          * check is to catch wrap-around problems.  If the lba + num blocks
8817          * is less than the lba, then we've wrapped around and the block
8818          * range is invalid anyway.
8819          */
8820         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8821          || ((lba + num_blocks) < lba)) {
8822                 ctl_set_lba_out_of_range(ctsio,
8823                     MAX(lba, lun->be_lun->maxlba + 1));
8824                 ctl_done((union ctl_io *)ctsio);
8825                 return (CTL_RETVAL_COMPLETE);
8826         }
8827
8828         /*
8829          * According to SBC-3, a transfer length of 0 is not an error.
8830          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8831          * translates to 256 blocks for those commands.
8832          */
8833         if (num_blocks == 0) {
8834                 ctl_set_success(ctsio);
8835                 ctl_done((union ctl_io *)ctsio);
8836                 return (CTL_RETVAL_COMPLETE);
8837         }
8838
8839         /* Set FUA and/or DPO if caches are disabled. */
8840         if (isread) {
8841                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8842                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8843         } else {
8844                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8845                         flags |= CTL_LLF_FUA;
8846         }
8847
8848         lbalen = (struct ctl_lba_len_flags *)
8849             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8850         lbalen->lba = lba;
8851         lbalen->len = num_blocks;
8852         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8853
8854         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8855         ctsio->kern_rel_offset = 0;
8856
8857         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8858
8859         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8860         return (retval);
8861 }
8862
8863 static int
8864 ctl_cnw_cont(union ctl_io *io)
8865 {
8866         struct ctl_lun *lun = CTL_LUN(io);
8867         struct ctl_scsiio *ctsio;
8868         struct ctl_lba_len_flags *lbalen;
8869         int retval;
8870
8871         ctsio = &io->scsiio;
8872         ctsio->io_hdr.status = CTL_STATUS_NONE;
8873         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8874         lbalen = (struct ctl_lba_len_flags *)
8875             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8876         lbalen->flags &= ~CTL_LLF_COMPARE;
8877         lbalen->flags |= CTL_LLF_WRITE;
8878
8879         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8880         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8881         return (retval);
8882 }
8883
8884 int
8885 ctl_cnw(struct ctl_scsiio *ctsio)
8886 {
8887         struct ctl_lun *lun = CTL_LUN(ctsio);
8888         struct ctl_lba_len_flags *lbalen;
8889         uint64_t lba;
8890         uint32_t num_blocks;
8891         int flags, retval;
8892
8893         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8894
8895         flags = 0;
8896         switch (ctsio->cdb[0]) {
8897         case COMPARE_AND_WRITE: {
8898                 struct scsi_compare_and_write *cdb;
8899
8900                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8901                 if (cdb->byte2 & SRW10_FUA)
8902                         flags |= CTL_LLF_FUA;
8903                 if (cdb->byte2 & SRW10_DPO)
8904                         flags |= CTL_LLF_DPO;
8905                 lba = scsi_8btou64(cdb->addr);
8906                 num_blocks = cdb->length;
8907                 break;
8908         }
8909         default:
8910                 /*
8911                  * We got a command we don't support.  This shouldn't
8912                  * happen, commands should be filtered out above us.
8913                  */
8914                 ctl_set_invalid_opcode(ctsio);
8915                 ctl_done((union ctl_io *)ctsio);
8916
8917                 return (CTL_RETVAL_COMPLETE);
8918                 break; /* NOTREACHED */
8919         }
8920
8921         /*
8922          * The first check is to make sure we're in bounds, the second
8923          * check is to catch wrap-around problems.  If the lba + num blocks
8924          * is less than the lba, then we've wrapped around and the block
8925          * range is invalid anyway.
8926          */
8927         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8928          || ((lba + num_blocks) < lba)) {
8929                 ctl_set_lba_out_of_range(ctsio,
8930                     MAX(lba, lun->be_lun->maxlba + 1));
8931                 ctl_done((union ctl_io *)ctsio);
8932                 return (CTL_RETVAL_COMPLETE);
8933         }
8934
8935         /*
8936          * According to SBC-3, a transfer length of 0 is not an error.
8937          */
8938         if (num_blocks == 0) {
8939                 ctl_set_success(ctsio);
8940                 ctl_done((union ctl_io *)ctsio);
8941                 return (CTL_RETVAL_COMPLETE);
8942         }
8943
8944         /* Set FUA if write cache is disabled. */
8945         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8946                 flags |= CTL_LLF_FUA;
8947
8948         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8949         ctsio->kern_rel_offset = 0;
8950
8951         /*
8952          * Set the IO_CONT flag, so that if this I/O gets passed to
8953          * ctl_data_submit_done(), it'll get passed back to
8954          * ctl_ctl_cnw_cont() for further processing.
8955          */
8956         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8957         ctsio->io_cont = ctl_cnw_cont;
8958
8959         lbalen = (struct ctl_lba_len_flags *)
8960             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8961         lbalen->lba = lba;
8962         lbalen->len = num_blocks;
8963         lbalen->flags = CTL_LLF_COMPARE | flags;
8964
8965         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8966         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8967         return (retval);
8968 }
8969
8970 int
8971 ctl_verify(struct ctl_scsiio *ctsio)
8972 {
8973         struct ctl_lun *lun = CTL_LUN(ctsio);
8974         struct ctl_lba_len_flags *lbalen;
8975         uint64_t lba;
8976         uint32_t num_blocks;
8977         int bytchk, flags;
8978         int retval;
8979
8980         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8981
8982         bytchk = 0;
8983         flags = CTL_LLF_FUA;
8984         switch (ctsio->cdb[0]) {
8985         case VERIFY_10: {
8986                 struct scsi_verify_10 *cdb;
8987
8988                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8989                 if (cdb->byte2 & SVFY_BYTCHK)
8990                         bytchk = 1;
8991                 if (cdb->byte2 & SVFY_DPO)
8992                         flags |= CTL_LLF_DPO;
8993                 lba = scsi_4btoul(cdb->addr);
8994                 num_blocks = scsi_2btoul(cdb->length);
8995                 break;
8996         }
8997         case VERIFY_12: {
8998                 struct scsi_verify_12 *cdb;
8999
9000                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
9001                 if (cdb->byte2 & SVFY_BYTCHK)
9002                         bytchk = 1;
9003                 if (cdb->byte2 & SVFY_DPO)
9004                         flags |= CTL_LLF_DPO;
9005                 lba = scsi_4btoul(cdb->addr);
9006                 num_blocks = scsi_4btoul(cdb->length);
9007                 break;
9008         }
9009         case VERIFY_16: {
9010                 struct scsi_rw_16 *cdb;
9011
9012                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9013                 if (cdb->byte2 & SVFY_BYTCHK)
9014                         bytchk = 1;
9015                 if (cdb->byte2 & SVFY_DPO)
9016                         flags |= CTL_LLF_DPO;
9017                 lba = scsi_8btou64(cdb->addr);
9018                 num_blocks = scsi_4btoul(cdb->length);
9019                 break;
9020         }
9021         default:
9022                 /*
9023                  * We got a command we don't support.  This shouldn't
9024                  * happen, commands should be filtered out above us.
9025                  */
9026                 ctl_set_invalid_opcode(ctsio);
9027                 ctl_done((union ctl_io *)ctsio);
9028                 return (CTL_RETVAL_COMPLETE);
9029         }
9030
9031         /*
9032          * The first check is to make sure we're in bounds, the second
9033          * check is to catch wrap-around problems.  If the lba + num blocks
9034          * is less than the lba, then we've wrapped around and the block
9035          * range is invalid anyway.
9036          */
9037         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9038          || ((lba + num_blocks) < lba)) {
9039                 ctl_set_lba_out_of_range(ctsio,
9040                     MAX(lba, lun->be_lun->maxlba + 1));
9041                 ctl_done((union ctl_io *)ctsio);
9042                 return (CTL_RETVAL_COMPLETE);
9043         }
9044
9045         /*
9046          * According to SBC-3, a transfer length of 0 is not an error.
9047          */
9048         if (num_blocks == 0) {
9049                 ctl_set_success(ctsio);
9050                 ctl_done((union ctl_io *)ctsio);
9051                 return (CTL_RETVAL_COMPLETE);
9052         }
9053
9054         lbalen = (struct ctl_lba_len_flags *)
9055             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9056         lbalen->lba = lba;
9057         lbalen->len = num_blocks;
9058         if (bytchk) {
9059                 lbalen->flags = CTL_LLF_COMPARE | flags;
9060                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9061         } else {
9062                 lbalen->flags = CTL_LLF_VERIFY | flags;
9063                 ctsio->kern_total_len = 0;
9064         }
9065         ctsio->kern_rel_offset = 0;
9066
9067         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9068         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9069         return (retval);
9070 }
9071
9072 int
9073 ctl_report_luns(struct ctl_scsiio *ctsio)
9074 {
9075         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9076         struct ctl_port *port = CTL_PORT(ctsio);
9077         struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio);
9078         struct scsi_report_luns *cdb;
9079         struct scsi_report_luns_data *lun_data;
9080         int num_filled, num_luns, num_port_luns, retval;
9081         uint32_t alloc_len, lun_datalen;
9082         uint32_t initidx, targ_lun_id, lun_id;
9083
9084         retval = CTL_RETVAL_COMPLETE;
9085         cdb = (struct scsi_report_luns *)ctsio->cdb;
9086
9087         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9088
9089         num_luns = 0;
9090         num_port_luns = port->lun_map ? port->lun_map_size : CTL_MAX_LUNS;
9091         mtx_lock(&softc->ctl_lock);
9092         for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) {
9093                 if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX)
9094                         num_luns++;
9095         }
9096         mtx_unlock(&softc->ctl_lock);
9097
9098         switch (cdb->select_report) {
9099         case RPL_REPORT_DEFAULT:
9100         case RPL_REPORT_ALL:
9101         case RPL_REPORT_NONSUBSID:
9102                 break;
9103         case RPL_REPORT_WELLKNOWN:
9104         case RPL_REPORT_ADMIN:
9105         case RPL_REPORT_CONGLOM:
9106                 num_luns = 0;
9107                 break;
9108         default:
9109                 ctl_set_invalid_field(ctsio,
9110                                       /*sks_valid*/ 1,
9111                                       /*command*/ 1,
9112                                       /*field*/ 2,
9113                                       /*bit_valid*/ 0,
9114                                       /*bit*/ 0);
9115                 ctl_done((union ctl_io *)ctsio);
9116                 return (retval);
9117                 break; /* NOTREACHED */
9118         }
9119
9120         alloc_len = scsi_4btoul(cdb->length);
9121         /*
9122          * The initiator has to allocate at least 16 bytes for this request,
9123          * so he can at least get the header and the first LUN.  Otherwise
9124          * we reject the request (per SPC-3 rev 14, section 6.21).
9125          */
9126         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9127             sizeof(struct scsi_report_luns_lundata))) {
9128                 ctl_set_invalid_field(ctsio,
9129                                       /*sks_valid*/ 1,
9130                                       /*command*/ 1,
9131                                       /*field*/ 6,
9132                                       /*bit_valid*/ 0,
9133                                       /*bit*/ 0);
9134                 ctl_done((union ctl_io *)ctsio);
9135                 return (retval);
9136         }
9137
9138         lun_datalen = sizeof(*lun_data) +
9139                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9140
9141         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9142         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9143         ctsio->kern_sg_entries = 0;
9144
9145         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9146
9147         mtx_lock(&softc->ctl_lock);
9148         for (targ_lun_id = 0, num_filled = 0;
9149             targ_lun_id < num_port_luns && num_filled < num_luns;
9150             targ_lun_id++) {
9151                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9152                 if (lun_id == UINT32_MAX)
9153                         continue;
9154                 lun = softc->ctl_luns[lun_id];
9155                 if (lun == NULL)
9156                         continue;
9157
9158                 be64enc(lun_data->luns[num_filled++].lundata,
9159                     ctl_encode_lun(targ_lun_id));
9160
9161                 /*
9162                  * According to SPC-3, rev 14 section 6.21:
9163                  *
9164                  * "The execution of a REPORT LUNS command to any valid and
9165                  * installed logical unit shall clear the REPORTED LUNS DATA
9166                  * HAS CHANGED unit attention condition for all logical
9167                  * units of that target with respect to the requesting
9168                  * initiator. A valid and installed logical unit is one
9169                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9170                  * INQUIRY data (see 6.4.2)."
9171                  *
9172                  * If request_lun is NULL, the LUN this report luns command
9173                  * was issued to is either disabled or doesn't exist. In that
9174                  * case, we shouldn't clear any pending lun change unit
9175                  * attention.
9176                  */
9177                 if (request_lun != NULL) {
9178                         mtx_lock(&lun->lun_lock);
9179                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9180                         mtx_unlock(&lun->lun_lock);
9181                 }
9182         }
9183         mtx_unlock(&softc->ctl_lock);
9184
9185         /*
9186          * It's quite possible that we've returned fewer LUNs than we allocated
9187          * space for.  Trim it.
9188          */
9189         lun_datalen = sizeof(*lun_data) +
9190                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9191
9192         if (lun_datalen < alloc_len) {
9193                 ctsio->residual = alloc_len - lun_datalen;
9194                 ctsio->kern_data_len = lun_datalen;
9195                 ctsio->kern_total_len = lun_datalen;
9196         } else {
9197                 ctsio->residual = 0;
9198                 ctsio->kern_data_len = alloc_len;
9199                 ctsio->kern_total_len = alloc_len;
9200         }
9201         ctsio->kern_data_resid = 0;
9202         ctsio->kern_rel_offset = 0;
9203         ctsio->kern_sg_entries = 0;
9204
9205         /*
9206          * We set this to the actual data length, regardless of how much
9207          * space we actually have to return results.  If the user looks at
9208          * this value, he'll know whether or not he allocated enough space
9209          * and reissue the command if necessary.  We don't support well
9210          * known logical units, so if the user asks for that, return none.
9211          */
9212         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9213
9214         /*
9215          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9216          * this request.
9217          */
9218         ctl_set_success(ctsio);
9219         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9220         ctsio->be_move_done = ctl_config_move_done;
9221         ctl_datamove((union ctl_io *)ctsio);
9222         return (retval);
9223 }
9224
9225 int
9226 ctl_request_sense(struct ctl_scsiio *ctsio)
9227 {
9228         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9229         struct ctl_lun *lun = CTL_LUN(ctsio);
9230         struct scsi_request_sense *cdb;
9231         struct scsi_sense_data *sense_ptr;
9232         uint32_t initidx;
9233         int have_error;
9234         u_int sense_len = SSD_FULL_SIZE;
9235         scsi_sense_data_type sense_format;
9236         ctl_ua_type ua_type;
9237         uint8_t asc = 0, ascq = 0;
9238
9239         cdb = (struct scsi_request_sense *)ctsio->cdb;
9240
9241         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9242
9243         /*
9244          * Determine which sense format the user wants.
9245          */
9246         if (cdb->byte2 & SRS_DESC)
9247                 sense_format = SSD_TYPE_DESC;
9248         else
9249                 sense_format = SSD_TYPE_FIXED;
9250
9251         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9252         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9253         ctsio->kern_sg_entries = 0;
9254
9255         /*
9256          * struct scsi_sense_data, which is currently set to 256 bytes, is
9257          * larger than the largest allowed value for the length field in the
9258          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9259          */
9260         ctsio->residual = 0;
9261         ctsio->kern_data_len = cdb->length;
9262         ctsio->kern_total_len = cdb->length;
9263
9264         ctsio->kern_data_resid = 0;
9265         ctsio->kern_rel_offset = 0;
9266         ctsio->kern_sg_entries = 0;
9267
9268         /*
9269          * If we don't have a LUN, we don't have any pending sense.
9270          */
9271         if (lun == NULL ||
9272             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9273              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9274                 /* "Logical unit not supported" */
9275                 ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9276                     /*current_error*/ 1,
9277                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9278                     /*asc*/ 0x25,
9279                     /*ascq*/ 0x00,
9280                     SSD_ELEM_NONE);
9281                 goto send;
9282         }
9283
9284         have_error = 0;
9285         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9286         /*
9287          * Check for pending sense, and then for pending unit attentions.
9288          * Pending sense gets returned first, then pending unit attentions.
9289          */
9290         mtx_lock(&lun->lun_lock);
9291 #ifdef CTL_WITH_CA
9292         if (ctl_is_set(lun->have_ca, initidx)) {
9293                 scsi_sense_data_type stored_format;
9294
9295                 /*
9296                  * Check to see which sense format was used for the stored
9297                  * sense data.
9298                  */
9299                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9300
9301                 /*
9302                  * If the user requested a different sense format than the
9303                  * one we stored, then we need to convert it to the other
9304                  * format.  If we're going from descriptor to fixed format
9305                  * sense data, we may lose things in translation, depending
9306                  * on what options were used.
9307                  *
9308                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9309                  * for some reason we'll just copy it out as-is.
9310                  */
9311                 if ((stored_format == SSD_TYPE_FIXED)
9312                  && (sense_format == SSD_TYPE_DESC))
9313                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9314                             &lun->pending_sense[initidx],
9315                             (struct scsi_sense_data_desc *)sense_ptr);
9316                 else if ((stored_format == SSD_TYPE_DESC)
9317                       && (sense_format == SSD_TYPE_FIXED))
9318                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9319                             &lun->pending_sense[initidx],
9320                             (struct scsi_sense_data_fixed *)sense_ptr);
9321                 else
9322                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9323                                MIN(sizeof(*sense_ptr),
9324                                sizeof(lun->pending_sense[initidx])));
9325
9326                 ctl_clear_mask(lun->have_ca, initidx);
9327                 have_error = 1;
9328         } else
9329 #endif
9330         if (have_error == 0) {
9331                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9332                     sense_format);
9333                 if (ua_type != CTL_UA_NONE)
9334                         have_error = 1;
9335         }
9336         if (have_error == 0) {
9337                 /*
9338                  * Report informational exception if have one and allowed.
9339                  */
9340                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9341                         asc = lun->ie_asc;
9342                         ascq = lun->ie_ascq;
9343                 }
9344                 ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9345                     /*current_error*/ 1,
9346                     /*sense_key*/ SSD_KEY_NO_SENSE,
9347                     /*asc*/ asc,
9348                     /*ascq*/ ascq,
9349                     SSD_ELEM_NONE);
9350         }
9351         mtx_unlock(&lun->lun_lock);
9352
9353 send:
9354         /*
9355          * We report the SCSI status as OK, since the status of the command
9356          * itself is OK.  We're reporting sense as parameter data.
9357          */
9358         ctl_set_success(ctsio);
9359         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9360         ctsio->be_move_done = ctl_config_move_done;
9361         ctl_datamove((union ctl_io *)ctsio);
9362         return (CTL_RETVAL_COMPLETE);
9363 }
9364
9365 int
9366 ctl_tur(struct ctl_scsiio *ctsio)
9367 {
9368
9369         CTL_DEBUG_PRINT(("ctl_tur\n"));
9370
9371         ctl_set_success(ctsio);
9372         ctl_done((union ctl_io *)ctsio);
9373
9374         return (CTL_RETVAL_COMPLETE);
9375 }
9376
9377 /*
9378  * SCSI VPD page 0x00, the Supported VPD Pages page.
9379  */
9380 static int
9381 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9382 {
9383         struct ctl_lun *lun = CTL_LUN(ctsio);
9384         struct scsi_vpd_supported_pages *pages;
9385         int sup_page_size;
9386         int p;
9387
9388         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9389             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9390         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9391         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9392         ctsio->kern_sg_entries = 0;
9393
9394         if (sup_page_size < alloc_len) {
9395                 ctsio->residual = alloc_len - sup_page_size;
9396                 ctsio->kern_data_len = sup_page_size;
9397                 ctsio->kern_total_len = sup_page_size;
9398         } else {
9399                 ctsio->residual = 0;
9400                 ctsio->kern_data_len = alloc_len;
9401                 ctsio->kern_total_len = alloc_len;
9402         }
9403         ctsio->kern_data_resid = 0;
9404         ctsio->kern_rel_offset = 0;
9405         ctsio->kern_sg_entries = 0;
9406
9407         /*
9408          * The control device is always connected.  The disk device, on the
9409          * other hand, may not be online all the time.  Need to change this
9410          * to figure out whether the disk device is actually online or not.
9411          */
9412         if (lun != NULL)
9413                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9414                                 lun->be_lun->lun_type;
9415         else
9416                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9417
9418         p = 0;
9419         /* Supported VPD pages */
9420         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9421         /* Serial Number */
9422         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9423         /* Device Identification */
9424         pages->page_list[p++] = SVPD_DEVICE_ID;
9425         /* Extended INQUIRY Data */
9426         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9427         /* Mode Page Policy */
9428         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9429         /* SCSI Ports */
9430         pages->page_list[p++] = SVPD_SCSI_PORTS;
9431         /* Third-party Copy */
9432         pages->page_list[p++] = SVPD_SCSI_TPC;
9433         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9434                 /* Block limits */
9435                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9436                 /* Block Device Characteristics */
9437                 pages->page_list[p++] = SVPD_BDC;
9438                 /* Logical Block Provisioning */
9439                 pages->page_list[p++] = SVPD_LBP;
9440         }
9441         pages->length = p;
9442
9443         ctl_set_success(ctsio);
9444         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9445         ctsio->be_move_done = ctl_config_move_done;
9446         ctl_datamove((union ctl_io *)ctsio);
9447         return (CTL_RETVAL_COMPLETE);
9448 }
9449
9450 /*
9451  * SCSI VPD page 0x80, the Unit Serial Number page.
9452  */
9453 static int
9454 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9455 {
9456         struct ctl_lun *lun = CTL_LUN(ctsio);
9457         struct scsi_vpd_unit_serial_number *sn_ptr;
9458         int data_len;
9459
9460         data_len = 4 + CTL_SN_LEN;
9461         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9462         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9463         if (data_len < alloc_len) {
9464                 ctsio->residual = alloc_len - data_len;
9465                 ctsio->kern_data_len = data_len;
9466                 ctsio->kern_total_len = data_len;
9467         } else {
9468                 ctsio->residual = 0;
9469                 ctsio->kern_data_len = alloc_len;
9470                 ctsio->kern_total_len = alloc_len;
9471         }
9472         ctsio->kern_data_resid = 0;
9473         ctsio->kern_rel_offset = 0;
9474         ctsio->kern_sg_entries = 0;
9475
9476         /*
9477          * The control device is always connected.  The disk device, on the
9478          * other hand, may not be online all the time.  Need to change this
9479          * to figure out whether the disk device is actually online or not.
9480          */
9481         if (lun != NULL)
9482                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9483                                   lun->be_lun->lun_type;
9484         else
9485                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9486
9487         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9488         sn_ptr->length = CTL_SN_LEN;
9489         /*
9490          * If we don't have a LUN, we just leave the serial number as
9491          * all spaces.
9492          */
9493         if (lun != NULL) {
9494                 strncpy((char *)sn_ptr->serial_num,
9495                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9496         } else
9497                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9498
9499         ctl_set_success(ctsio);
9500         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9501         ctsio->be_move_done = ctl_config_move_done;
9502         ctl_datamove((union ctl_io *)ctsio);
9503         return (CTL_RETVAL_COMPLETE);
9504 }
9505
9506
9507 /*
9508  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9509  */
9510 static int
9511 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9512 {
9513         struct ctl_lun *lun = CTL_LUN(ctsio);
9514         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9515         int data_len;
9516
9517         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9518         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9519         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9520         ctsio->kern_sg_entries = 0;
9521
9522         if (data_len < alloc_len) {
9523                 ctsio->residual = alloc_len - data_len;
9524                 ctsio->kern_data_len = data_len;
9525                 ctsio->kern_total_len = data_len;
9526         } else {
9527                 ctsio->residual = 0;
9528                 ctsio->kern_data_len = alloc_len;
9529                 ctsio->kern_total_len = alloc_len;
9530         }
9531         ctsio->kern_data_resid = 0;
9532         ctsio->kern_rel_offset = 0;
9533         ctsio->kern_sg_entries = 0;
9534
9535         /*
9536          * The control device is always connected.  The disk device, on the
9537          * other hand, may not be online all the time.
9538          */
9539         if (lun != NULL)
9540                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9541                                      lun->be_lun->lun_type;
9542         else
9543                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9544         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9545         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9546         /*
9547          * We support head of queue, ordered and simple tags.
9548          */
9549         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9550         /*
9551          * Volatile cache supported.
9552          */
9553         eid_ptr->flags3 = SVPD_EID_V_SUP;
9554
9555         /*
9556          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9557          * attention for a particular IT nexus on all LUNs once we report
9558          * it to that nexus once.  This bit is required as of SPC-4.
9559          */
9560         eid_ptr->flags4 = SVPD_EID_LUICLR;
9561
9562         /*
9563          * We support revert to defaults (RTD) bit in MODE SELECT.
9564          */
9565         eid_ptr->flags5 = SVPD_EID_RTD_SUP;
9566
9567         /*
9568          * XXX KDM in order to correctly answer this, we would need
9569          * information from the SIM to determine how much sense data it
9570          * can send.  So this would really be a path inquiry field, most
9571          * likely.  This can be set to a maximum of 252 according to SPC-4,
9572          * but the hardware may or may not be able to support that much.
9573          * 0 just means that the maximum sense data length is not reported.
9574          */
9575         eid_ptr->max_sense_length = 0;
9576
9577         ctl_set_success(ctsio);
9578         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9579         ctsio->be_move_done = ctl_config_move_done;
9580         ctl_datamove((union ctl_io *)ctsio);
9581         return (CTL_RETVAL_COMPLETE);
9582 }
9583
9584 static int
9585 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9586 {
9587         struct ctl_lun *lun = CTL_LUN(ctsio);
9588         struct scsi_vpd_mode_page_policy *mpp_ptr;
9589         int data_len;
9590
9591         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9592             sizeof(struct scsi_vpd_mode_page_policy_descr);
9593
9594         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9595         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9596         ctsio->kern_sg_entries = 0;
9597
9598         if (data_len < alloc_len) {
9599                 ctsio->residual = alloc_len - data_len;
9600                 ctsio->kern_data_len = data_len;
9601                 ctsio->kern_total_len = data_len;
9602         } else {
9603                 ctsio->residual = 0;
9604                 ctsio->kern_data_len = alloc_len;
9605                 ctsio->kern_total_len = alloc_len;
9606         }
9607         ctsio->kern_data_resid = 0;
9608         ctsio->kern_rel_offset = 0;
9609         ctsio->kern_sg_entries = 0;
9610
9611         /*
9612          * The control device is always connected.  The disk device, on the
9613          * other hand, may not be online all the time.
9614          */
9615         if (lun != NULL)
9616                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9617                                      lun->be_lun->lun_type;
9618         else
9619                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9620         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9621         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9622         mpp_ptr->descr[0].page_code = 0x3f;
9623         mpp_ptr->descr[0].subpage_code = 0xff;
9624         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9625
9626         ctl_set_success(ctsio);
9627         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9628         ctsio->be_move_done = ctl_config_move_done;
9629         ctl_datamove((union ctl_io *)ctsio);
9630         return (CTL_RETVAL_COMPLETE);
9631 }
9632
9633 /*
9634  * SCSI VPD page 0x83, the Device Identification page.
9635  */
9636 static int
9637 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9638 {
9639         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9640         struct ctl_port *port = CTL_PORT(ctsio);
9641         struct ctl_lun *lun = CTL_LUN(ctsio);
9642         struct scsi_vpd_device_id *devid_ptr;
9643         struct scsi_vpd_id_descriptor *desc;
9644         int data_len, g;
9645         uint8_t proto;
9646
9647         data_len = sizeof(struct scsi_vpd_device_id) +
9648             sizeof(struct scsi_vpd_id_descriptor) +
9649                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9650             sizeof(struct scsi_vpd_id_descriptor) +
9651                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9652         if (lun && lun->lun_devid)
9653                 data_len += lun->lun_devid->len;
9654         if (port && port->port_devid)
9655                 data_len += port->port_devid->len;
9656         if (port && port->target_devid)
9657                 data_len += port->target_devid->len;
9658
9659         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9660         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9661         ctsio->kern_sg_entries = 0;
9662
9663         if (data_len < alloc_len) {
9664                 ctsio->residual = alloc_len - data_len;
9665                 ctsio->kern_data_len = data_len;
9666                 ctsio->kern_total_len = data_len;
9667         } else {
9668                 ctsio->residual = 0;
9669                 ctsio->kern_data_len = alloc_len;
9670                 ctsio->kern_total_len = alloc_len;
9671         }
9672         ctsio->kern_data_resid = 0;
9673         ctsio->kern_rel_offset = 0;
9674         ctsio->kern_sg_entries = 0;
9675
9676         /*
9677          * The control device is always connected.  The disk device, on the
9678          * other hand, may not be online all the time.
9679          */
9680         if (lun != NULL)
9681                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9682                                      lun->be_lun->lun_type;
9683         else
9684                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9685         devid_ptr->page_code = SVPD_DEVICE_ID;
9686         scsi_ulto2b(data_len - 4, devid_ptr->length);
9687
9688         if (port && port->port_type == CTL_PORT_FC)
9689                 proto = SCSI_PROTO_FC << 4;
9690         else if (port && port->port_type == CTL_PORT_ISCSI)
9691                 proto = SCSI_PROTO_ISCSI << 4;
9692         else
9693                 proto = SCSI_PROTO_SPI << 4;
9694         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9695
9696         /*
9697          * We're using a LUN association here.  i.e., this device ID is a
9698          * per-LUN identifier.
9699          */
9700         if (lun && lun->lun_devid) {
9701                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9702                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9703                     lun->lun_devid->len);
9704         }
9705
9706         /*
9707          * This is for the WWPN which is a port association.
9708          */
9709         if (port && port->port_devid) {
9710                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9711                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9712                     port->port_devid->len);
9713         }
9714
9715         /*
9716          * This is for the Relative Target Port(type 4h) identifier
9717          */
9718         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9719         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9720             SVPD_ID_TYPE_RELTARG;
9721         desc->length = 4;
9722         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9723         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9724             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9725
9726         /*
9727          * This is for the Target Port Group(type 5h) identifier
9728          */
9729         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9730         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9731             SVPD_ID_TYPE_TPORTGRP;
9732         desc->length = 4;
9733         if (softc->is_single ||
9734             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9735                 g = 1;
9736         else
9737                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9738         scsi_ulto2b(g, &desc->identifier[2]);
9739         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9740             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9741
9742         /*
9743          * This is for the Target identifier
9744          */
9745         if (port && port->target_devid) {
9746                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9747         }
9748
9749         ctl_set_success(ctsio);
9750         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9751         ctsio->be_move_done = ctl_config_move_done;
9752         ctl_datamove((union ctl_io *)ctsio);
9753         return (CTL_RETVAL_COMPLETE);
9754 }
9755
9756 static int
9757 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9758 {
9759         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9760         struct ctl_lun *lun = CTL_LUN(ctsio);
9761         struct scsi_vpd_scsi_ports *sp;
9762         struct scsi_vpd_port_designation *pd;
9763         struct scsi_vpd_port_designation_cont *pdc;
9764         struct ctl_port *port;
9765         int data_len, num_target_ports, iid_len, id_len;
9766
9767         num_target_ports = 0;
9768         iid_len = 0;
9769         id_len = 0;
9770         mtx_lock(&softc->ctl_lock);
9771         STAILQ_FOREACH(port, &softc->port_list, links) {
9772                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9773                         continue;
9774                 if (lun != NULL &&
9775                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9776                         continue;
9777                 num_target_ports++;
9778                 if (port->init_devid)
9779                         iid_len += port->init_devid->len;
9780                 if (port->port_devid)
9781                         id_len += port->port_devid->len;
9782         }
9783         mtx_unlock(&softc->ctl_lock);
9784
9785         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9786             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9787              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9788         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9789         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9790         ctsio->kern_sg_entries = 0;
9791
9792         if (data_len < alloc_len) {
9793                 ctsio->residual = alloc_len - data_len;
9794                 ctsio->kern_data_len = data_len;
9795                 ctsio->kern_total_len = data_len;
9796         } else {
9797                 ctsio->residual = 0;
9798                 ctsio->kern_data_len = alloc_len;
9799                 ctsio->kern_total_len = alloc_len;
9800         }
9801         ctsio->kern_data_resid = 0;
9802         ctsio->kern_rel_offset = 0;
9803         ctsio->kern_sg_entries = 0;
9804
9805         /*
9806          * The control device is always connected.  The disk device, on the
9807          * other hand, may not be online all the time.  Need to change this
9808          * to figure out whether the disk device is actually online or not.
9809          */
9810         if (lun != NULL)
9811                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9812                                   lun->be_lun->lun_type;
9813         else
9814                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9815
9816         sp->page_code = SVPD_SCSI_PORTS;
9817         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9818             sp->page_length);
9819         pd = &sp->design[0];
9820
9821         mtx_lock(&softc->ctl_lock);
9822         STAILQ_FOREACH(port, &softc->port_list, links) {
9823                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9824                         continue;
9825                 if (lun != NULL &&
9826                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9827                         continue;
9828                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9829                 if (port->init_devid) {
9830                         iid_len = port->init_devid->len;
9831                         memcpy(pd->initiator_transportid,
9832                             port->init_devid->data, port->init_devid->len);
9833                 } else
9834                         iid_len = 0;
9835                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9836                 pdc = (struct scsi_vpd_port_designation_cont *)
9837                     (&pd->initiator_transportid[iid_len]);
9838                 if (port->port_devid) {
9839                         id_len = port->port_devid->len;
9840                         memcpy(pdc->target_port_descriptors,
9841                             port->port_devid->data, port->port_devid->len);
9842                 } else
9843                         id_len = 0;
9844                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9845                 pd = (struct scsi_vpd_port_designation *)
9846                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9847         }
9848         mtx_unlock(&softc->ctl_lock);
9849
9850         ctl_set_success(ctsio);
9851         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9852         ctsio->be_move_done = ctl_config_move_done;
9853         ctl_datamove((union ctl_io *)ctsio);
9854         return (CTL_RETVAL_COMPLETE);
9855 }
9856
9857 static int
9858 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9859 {
9860         struct ctl_lun *lun = CTL_LUN(ctsio);
9861         struct scsi_vpd_block_limits *bl_ptr;
9862         uint64_t ival;
9863
9864         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9865         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9866         ctsio->kern_sg_entries = 0;
9867
9868         if (sizeof(*bl_ptr) < alloc_len) {
9869                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
9870                 ctsio->kern_data_len = sizeof(*bl_ptr);
9871                 ctsio->kern_total_len = sizeof(*bl_ptr);
9872         } else {
9873                 ctsio->residual = 0;
9874                 ctsio->kern_data_len = alloc_len;
9875                 ctsio->kern_total_len = alloc_len;
9876         }
9877         ctsio->kern_data_resid = 0;
9878         ctsio->kern_rel_offset = 0;
9879         ctsio->kern_sg_entries = 0;
9880
9881         /*
9882          * The control device is always connected.  The disk device, on the
9883          * other hand, may not be online all the time.  Need to change this
9884          * to figure out whether the disk device is actually online or not.
9885          */
9886         if (lun != NULL)
9887                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9888                                   lun->be_lun->lun_type;
9889         else
9890                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9891
9892         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9893         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9894         bl_ptr->max_cmp_write_len = 0xff;
9895         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9896         if (lun != NULL) {
9897                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9898                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9899                         ival = 0xffffffff;
9900                         ctl_get_opt_number(&lun->be_lun->options,
9901                             "unmap_max_lba", &ival);
9902                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9903                         ival = 0xffffffff;
9904                         ctl_get_opt_number(&lun->be_lun->options,
9905                             "unmap_max_descr", &ival);
9906                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9907                         if (lun->be_lun->ublockexp != 0) {
9908                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9909                                     bl_ptr->opt_unmap_grain);
9910                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9911                                     bl_ptr->unmap_grain_align);
9912                         }
9913                 }
9914                 scsi_ulto4b(lun->be_lun->atomicblock,
9915                     bl_ptr->max_atomic_transfer_length);
9916                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9917                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9918                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9919                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9920                 ival = UINT64_MAX;
9921                 ctl_get_opt_number(&lun->be_lun->options, "write_same_max_lba", &ival);
9922                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9923         }
9924
9925         ctl_set_success(ctsio);
9926         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9927         ctsio->be_move_done = ctl_config_move_done;
9928         ctl_datamove((union ctl_io *)ctsio);
9929         return (CTL_RETVAL_COMPLETE);
9930 }
9931
9932 static int
9933 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9934 {
9935         struct ctl_lun *lun = CTL_LUN(ctsio);
9936         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9937         const char *value;
9938         u_int i;
9939
9940         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9941         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9942         ctsio->kern_sg_entries = 0;
9943
9944         if (sizeof(*bdc_ptr) < alloc_len) {
9945                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
9946                 ctsio->kern_data_len = sizeof(*bdc_ptr);
9947                 ctsio->kern_total_len = sizeof(*bdc_ptr);
9948         } else {
9949                 ctsio->residual = 0;
9950                 ctsio->kern_data_len = alloc_len;
9951                 ctsio->kern_total_len = alloc_len;
9952         }
9953         ctsio->kern_data_resid = 0;
9954         ctsio->kern_rel_offset = 0;
9955         ctsio->kern_sg_entries = 0;
9956
9957         /*
9958          * The control device is always connected.  The disk device, on the
9959          * other hand, may not be online all the time.  Need to change this
9960          * to figure out whether the disk device is actually online or not.
9961          */
9962         if (lun != NULL)
9963                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9964                                   lun->be_lun->lun_type;
9965         else
9966                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9967         bdc_ptr->page_code = SVPD_BDC;
9968         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9969         if (lun != NULL &&
9970             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9971                 i = strtol(value, NULL, 0);
9972         else
9973                 i = CTL_DEFAULT_ROTATION_RATE;
9974         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9975         if (lun != NULL &&
9976             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9977                 i = strtol(value, NULL, 0);
9978         else
9979                 i = 0;
9980         bdc_ptr->wab_wac_ff = (i & 0x0f);
9981         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9982
9983         ctl_set_success(ctsio);
9984         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9985         ctsio->be_move_done = ctl_config_move_done;
9986         ctl_datamove((union ctl_io *)ctsio);
9987         return (CTL_RETVAL_COMPLETE);
9988 }
9989
9990 static int
9991 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9992 {
9993         struct ctl_lun *lun = CTL_LUN(ctsio);
9994         struct scsi_vpd_logical_block_prov *lbp_ptr;
9995         const char *value;
9996
9997         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9998         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9999         ctsio->kern_sg_entries = 0;
10000
10001         if (sizeof(*lbp_ptr) < alloc_len) {
10002                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10003                 ctsio->kern_data_len = sizeof(*lbp_ptr);
10004                 ctsio->kern_total_len = sizeof(*lbp_ptr);
10005         } else {
10006                 ctsio->residual = 0;
10007                 ctsio->kern_data_len = alloc_len;
10008                 ctsio->kern_total_len = alloc_len;
10009         }
10010         ctsio->kern_data_resid = 0;
10011         ctsio->kern_rel_offset = 0;
10012         ctsio->kern_sg_entries = 0;
10013
10014         /*
10015          * The control device is always connected.  The disk device, on the
10016          * other hand, may not be online all the time.  Need to change this
10017          * to figure out whether the disk device is actually online or not.
10018          */
10019         if (lun != NULL)
10020                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10021                                   lun->be_lun->lun_type;
10022         else
10023                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10024
10025         lbp_ptr->page_code = SVPD_LBP;
10026         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10027         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10028         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10029                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10030                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10031                 value = ctl_get_opt(&lun->be_lun->options, "provisioning_type");
10032                 if (value != NULL) {
10033                         if (strcmp(value, "resource") == 0)
10034                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10035                         else if (strcmp(value, "thin") == 0)
10036                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
10037                 } else
10038                         lbp_ptr->prov_type = SVPD_LBP_THIN;
10039         }
10040
10041         ctl_set_success(ctsio);
10042         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10043         ctsio->be_move_done = ctl_config_move_done;
10044         ctl_datamove((union ctl_io *)ctsio);
10045         return (CTL_RETVAL_COMPLETE);
10046 }
10047
10048 /*
10049  * INQUIRY with the EVPD bit set.
10050  */
10051 static int
10052 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10053 {
10054         struct ctl_lun *lun = CTL_LUN(ctsio);
10055         struct scsi_inquiry *cdb;
10056         int alloc_len, retval;
10057
10058         cdb = (struct scsi_inquiry *)ctsio->cdb;
10059         alloc_len = scsi_2btoul(cdb->length);
10060
10061         switch (cdb->page_code) {
10062         case SVPD_SUPPORTED_PAGES:
10063                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10064                 break;
10065         case SVPD_UNIT_SERIAL_NUMBER:
10066                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10067                 break;
10068         case SVPD_DEVICE_ID:
10069                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10070                 break;
10071         case SVPD_EXTENDED_INQUIRY_DATA:
10072                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10073                 break;
10074         case SVPD_MODE_PAGE_POLICY:
10075                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10076                 break;
10077         case SVPD_SCSI_PORTS:
10078                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10079                 break;
10080         case SVPD_SCSI_TPC:
10081                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10082                 break;
10083         case SVPD_BLOCK_LIMITS:
10084                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10085                         goto err;
10086                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10087                 break;
10088         case SVPD_BDC:
10089                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10090                         goto err;
10091                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10092                 break;
10093         case SVPD_LBP:
10094                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10095                         goto err;
10096                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10097                 break;
10098         default:
10099 err:
10100                 ctl_set_invalid_field(ctsio,
10101                                       /*sks_valid*/ 1,
10102                                       /*command*/ 1,
10103                                       /*field*/ 2,
10104                                       /*bit_valid*/ 0,
10105                                       /*bit*/ 0);
10106                 ctl_done((union ctl_io *)ctsio);
10107                 retval = CTL_RETVAL_COMPLETE;
10108                 break;
10109         }
10110
10111         return (retval);
10112 }
10113
10114 /*
10115  * Standard INQUIRY data.
10116  */
10117 static int
10118 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10119 {
10120         struct ctl_softc *softc = CTL_SOFTC(ctsio);
10121         struct ctl_port *port = CTL_PORT(ctsio);
10122         struct ctl_lun *lun = CTL_LUN(ctsio);
10123         struct scsi_inquiry_data *inq_ptr;
10124         struct scsi_inquiry *cdb;
10125         char *val;
10126         uint32_t alloc_len, data_len;
10127         ctl_port_type port_type;
10128
10129         port_type = port->port_type;
10130         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10131                 port_type = CTL_PORT_SCSI;
10132
10133         cdb = (struct scsi_inquiry *)ctsio->cdb;
10134         alloc_len = scsi_2btoul(cdb->length);
10135
10136         /*
10137          * We malloc the full inquiry data size here and fill it
10138          * in.  If the user only asks for less, we'll give him
10139          * that much.
10140          */
10141         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10142         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10143         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10144         ctsio->kern_sg_entries = 0;
10145         ctsio->kern_data_resid = 0;
10146         ctsio->kern_rel_offset = 0;
10147
10148         if (data_len < alloc_len) {
10149                 ctsio->residual = alloc_len - data_len;
10150                 ctsio->kern_data_len = data_len;
10151                 ctsio->kern_total_len = data_len;
10152         } else {
10153                 ctsio->residual = 0;
10154                 ctsio->kern_data_len = alloc_len;
10155                 ctsio->kern_total_len = alloc_len;
10156         }
10157
10158         if (lun != NULL) {
10159                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10160                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10161                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10162                             lun->be_lun->lun_type;
10163                 } else {
10164                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10165                             lun->be_lun->lun_type;
10166                 }
10167                 if (lun->flags & CTL_LUN_REMOVABLE)
10168                         inq_ptr->dev_qual2 |= SID_RMB;
10169         } else
10170                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10171
10172         /* RMB in byte 2 is 0 */
10173         inq_ptr->version = SCSI_REV_SPC5;
10174
10175         /*
10176          * According to SAM-3, even if a device only supports a single
10177          * level of LUN addressing, it should still set the HISUP bit:
10178          *
10179          * 4.9.1 Logical unit numbers overview
10180          *
10181          * All logical unit number formats described in this standard are
10182          * hierarchical in structure even when only a single level in that
10183          * hierarchy is used. The HISUP bit shall be set to one in the
10184          * standard INQUIRY data (see SPC-2) when any logical unit number
10185          * format described in this standard is used.  Non-hierarchical
10186          * formats are outside the scope of this standard.
10187          *
10188          * Therefore we set the HiSup bit here.
10189          *
10190          * The response format is 2, per SPC-3.
10191          */
10192         inq_ptr->response_format = SID_HiSup | 2;
10193
10194         inq_ptr->additional_length = data_len -
10195             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10196         CTL_DEBUG_PRINT(("additional_length = %d\n",
10197                          inq_ptr->additional_length));
10198
10199         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10200         if (port_type == CTL_PORT_SCSI)
10201                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10202         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10203         inq_ptr->flags = SID_CmdQue;
10204         if (port_type == CTL_PORT_SCSI)
10205                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10206
10207         /*
10208          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10209          * We have 8 bytes for the vendor name, and 16 bytes for the device
10210          * name and 4 bytes for the revision.
10211          */
10212         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10213             "vendor")) == NULL) {
10214                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10215         } else {
10216                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10217                 strncpy(inq_ptr->vendor, val,
10218                     min(sizeof(inq_ptr->vendor), strlen(val)));
10219         }
10220         if (lun == NULL) {
10221                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10222                     sizeof(inq_ptr->product));
10223         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10224                 switch (lun->be_lun->lun_type) {
10225                 case T_DIRECT:
10226                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10227                             sizeof(inq_ptr->product));
10228                         break;
10229                 case T_PROCESSOR:
10230                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10231                             sizeof(inq_ptr->product));
10232                         break;
10233                 case T_CDROM:
10234                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10235                             sizeof(inq_ptr->product));
10236                         break;
10237                 default:
10238                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10239                             sizeof(inq_ptr->product));
10240                         break;
10241                 }
10242         } else {
10243                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10244                 strncpy(inq_ptr->product, val,
10245                     min(sizeof(inq_ptr->product), strlen(val)));
10246         }
10247
10248         /*
10249          * XXX make this a macro somewhere so it automatically gets
10250          * incremented when we make changes.
10251          */
10252         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10253             "revision")) == NULL) {
10254                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10255         } else {
10256                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10257                 strncpy(inq_ptr->revision, val,
10258                     min(sizeof(inq_ptr->revision), strlen(val)));
10259         }
10260
10261         /*
10262          * For parallel SCSI, we support double transition and single
10263          * transition clocking.  We also support QAS (Quick Arbitration
10264          * and Selection) and Information Unit transfers on both the
10265          * control and array devices.
10266          */
10267         if (port_type == CTL_PORT_SCSI)
10268                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10269                                     SID_SPI_IUS;
10270
10271         /* SAM-6 (no version claimed) */
10272         scsi_ulto2b(0x00C0, inq_ptr->version1);
10273         /* SPC-5 (no version claimed) */
10274         scsi_ulto2b(0x05C0, inq_ptr->version2);
10275         if (port_type == CTL_PORT_FC) {
10276                 /* FCP-2 ANSI INCITS.350:2003 */
10277                 scsi_ulto2b(0x0917, inq_ptr->version3);
10278         } else if (port_type == CTL_PORT_SCSI) {
10279                 /* SPI-4 ANSI INCITS.362:200x */
10280                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10281         } else if (port_type == CTL_PORT_ISCSI) {
10282                 /* iSCSI (no version claimed) */
10283                 scsi_ulto2b(0x0960, inq_ptr->version3);
10284         } else if (port_type == CTL_PORT_SAS) {
10285                 /* SAS (no version claimed) */
10286                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10287         }
10288
10289         if (lun == NULL) {
10290                 /* SBC-4 (no version claimed) */
10291                 scsi_ulto2b(0x0600, inq_ptr->version4);
10292         } else {
10293                 switch (lun->be_lun->lun_type) {
10294                 case T_DIRECT:
10295                         /* SBC-4 (no version claimed) */
10296                         scsi_ulto2b(0x0600, inq_ptr->version4);
10297                         break;
10298                 case T_PROCESSOR:
10299                         break;
10300                 case T_CDROM:
10301                         /* MMC-6 (no version claimed) */
10302                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10303                         break;
10304                 default:
10305                         break;
10306                 }
10307         }
10308
10309         ctl_set_success(ctsio);
10310         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10311         ctsio->be_move_done = ctl_config_move_done;
10312         ctl_datamove((union ctl_io *)ctsio);
10313         return (CTL_RETVAL_COMPLETE);
10314 }
10315
10316 int
10317 ctl_inquiry(struct ctl_scsiio *ctsio)
10318 {
10319         struct scsi_inquiry *cdb;
10320         int retval;
10321
10322         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10323
10324         cdb = (struct scsi_inquiry *)ctsio->cdb;
10325         if (cdb->byte2 & SI_EVPD)
10326                 retval = ctl_inquiry_evpd(ctsio);
10327         else if (cdb->page_code == 0)
10328                 retval = ctl_inquiry_std(ctsio);
10329         else {
10330                 ctl_set_invalid_field(ctsio,
10331                                       /*sks_valid*/ 1,
10332                                       /*command*/ 1,
10333                                       /*field*/ 2,
10334                                       /*bit_valid*/ 0,
10335                                       /*bit*/ 0);
10336                 ctl_done((union ctl_io *)ctsio);
10337                 return (CTL_RETVAL_COMPLETE);
10338         }
10339
10340         return (retval);
10341 }
10342
10343 int
10344 ctl_get_config(struct ctl_scsiio *ctsio)
10345 {
10346         struct ctl_lun *lun = CTL_LUN(ctsio);
10347         struct scsi_get_config_header *hdr;
10348         struct scsi_get_config_feature *feature;
10349         struct scsi_get_config *cdb;
10350         uint32_t alloc_len, data_len;
10351         int rt, starting;
10352
10353         cdb = (struct scsi_get_config *)ctsio->cdb;
10354         rt = (cdb->rt & SGC_RT_MASK);
10355         starting = scsi_2btoul(cdb->starting_feature);
10356         alloc_len = scsi_2btoul(cdb->length);
10357
10358         data_len = sizeof(struct scsi_get_config_header) +
10359             sizeof(struct scsi_get_config_feature) + 8 +
10360             sizeof(struct scsi_get_config_feature) + 8 +
10361             sizeof(struct scsi_get_config_feature) + 4 +
10362             sizeof(struct scsi_get_config_feature) + 4 +
10363             sizeof(struct scsi_get_config_feature) + 8 +
10364             sizeof(struct scsi_get_config_feature) +
10365             sizeof(struct scsi_get_config_feature) + 4 +
10366             sizeof(struct scsi_get_config_feature) + 4 +
10367             sizeof(struct scsi_get_config_feature) + 4 +
10368             sizeof(struct scsi_get_config_feature) + 4 +
10369             sizeof(struct scsi_get_config_feature) + 4 +
10370             sizeof(struct scsi_get_config_feature) + 4;
10371         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10372         ctsio->kern_sg_entries = 0;
10373         ctsio->kern_data_resid = 0;
10374         ctsio->kern_rel_offset = 0;
10375
10376         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10377         if (lun->flags & CTL_LUN_NO_MEDIA)
10378                 scsi_ulto2b(0x0000, hdr->current_profile);
10379         else
10380                 scsi_ulto2b(0x0010, hdr->current_profile);
10381         feature = (struct scsi_get_config_feature *)(hdr + 1);
10382
10383         if (starting > 0x003b)
10384                 goto done;
10385         if (starting > 0x003a)
10386                 goto f3b;
10387         if (starting > 0x002b)
10388                 goto f3a;
10389         if (starting > 0x002a)
10390                 goto f2b;
10391         if (starting > 0x001f)
10392                 goto f2a;
10393         if (starting > 0x001e)
10394                 goto f1f;
10395         if (starting > 0x001d)
10396                 goto f1e;
10397         if (starting > 0x0010)
10398                 goto f1d;
10399         if (starting > 0x0003)
10400                 goto f10;
10401         if (starting > 0x0002)
10402                 goto f3;
10403         if (starting > 0x0001)
10404                 goto f2;
10405         if (starting > 0x0000)
10406                 goto f1;
10407
10408         /* Profile List */
10409         scsi_ulto2b(0x0000, feature->feature_code);
10410         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10411         feature->add_length = 8;
10412         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10413         feature->feature_data[2] = 0x00;
10414         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10415         feature->feature_data[6] = 0x01;
10416         feature = (struct scsi_get_config_feature *)
10417             &feature->feature_data[feature->add_length];
10418
10419 f1:     /* Core */
10420         scsi_ulto2b(0x0001, feature->feature_code);
10421         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10422         feature->add_length = 8;
10423         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10424         feature->feature_data[4] = 0x03;
10425         feature = (struct scsi_get_config_feature *)
10426             &feature->feature_data[feature->add_length];
10427
10428 f2:     /* Morphing */
10429         scsi_ulto2b(0x0002, feature->feature_code);
10430         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10431         feature->add_length = 4;
10432         feature->feature_data[0] = 0x02;
10433         feature = (struct scsi_get_config_feature *)
10434             &feature->feature_data[feature->add_length];
10435
10436 f3:     /* Removable Medium */
10437         scsi_ulto2b(0x0003, feature->feature_code);
10438         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10439         feature->add_length = 4;
10440         feature->feature_data[0] = 0x39;
10441         feature = (struct scsi_get_config_feature *)
10442             &feature->feature_data[feature->add_length];
10443
10444         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10445                 goto done;
10446
10447 f10:    /* Random Read */
10448         scsi_ulto2b(0x0010, feature->feature_code);
10449         feature->flags = 0x00;
10450         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10451                 feature->flags |= SGC_F_CURRENT;
10452         feature->add_length = 8;
10453         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10454         scsi_ulto2b(1, &feature->feature_data[4]);
10455         feature->feature_data[6] = 0x00;
10456         feature = (struct scsi_get_config_feature *)
10457             &feature->feature_data[feature->add_length];
10458
10459 f1d:    /* Multi-Read */
10460         scsi_ulto2b(0x001D, feature->feature_code);
10461         feature->flags = 0x00;
10462         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10463                 feature->flags |= SGC_F_CURRENT;
10464         feature->add_length = 0;
10465         feature = (struct scsi_get_config_feature *)
10466             &feature->feature_data[feature->add_length];
10467
10468 f1e:    /* CD Read */
10469         scsi_ulto2b(0x001E, feature->feature_code);
10470         feature->flags = 0x00;
10471         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10472                 feature->flags |= SGC_F_CURRENT;
10473         feature->add_length = 4;
10474         feature->feature_data[0] = 0x00;
10475         feature = (struct scsi_get_config_feature *)
10476             &feature->feature_data[feature->add_length];
10477
10478 f1f:    /* DVD Read */
10479         scsi_ulto2b(0x001F, feature->feature_code);
10480         feature->flags = 0x08;
10481         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10482                 feature->flags |= SGC_F_CURRENT;
10483         feature->add_length = 4;
10484         feature->feature_data[0] = 0x01;
10485         feature->feature_data[2] = 0x03;
10486         feature = (struct scsi_get_config_feature *)
10487             &feature->feature_data[feature->add_length];
10488
10489 f2a:    /* DVD+RW */
10490         scsi_ulto2b(0x002A, feature->feature_code);
10491         feature->flags = 0x04;
10492         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10493                 feature->flags |= SGC_F_CURRENT;
10494         feature->add_length = 4;
10495         feature->feature_data[0] = 0x00;
10496         feature->feature_data[1] = 0x00;
10497         feature = (struct scsi_get_config_feature *)
10498             &feature->feature_data[feature->add_length];
10499
10500 f2b:    /* DVD+R */
10501         scsi_ulto2b(0x002B, feature->feature_code);
10502         feature->flags = 0x00;
10503         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10504                 feature->flags |= SGC_F_CURRENT;
10505         feature->add_length = 4;
10506         feature->feature_data[0] = 0x00;
10507         feature = (struct scsi_get_config_feature *)
10508             &feature->feature_data[feature->add_length];
10509
10510 f3a:    /* DVD+RW Dual Layer */
10511         scsi_ulto2b(0x003A, feature->feature_code);
10512         feature->flags = 0x00;
10513         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10514                 feature->flags |= SGC_F_CURRENT;
10515         feature->add_length = 4;
10516         feature->feature_data[0] = 0x00;
10517         feature->feature_data[1] = 0x00;
10518         feature = (struct scsi_get_config_feature *)
10519             &feature->feature_data[feature->add_length];
10520
10521 f3b:    /* DVD+R Dual Layer */
10522         scsi_ulto2b(0x003B, feature->feature_code);
10523         feature->flags = 0x00;
10524         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10525                 feature->flags |= SGC_F_CURRENT;
10526         feature->add_length = 4;
10527         feature->feature_data[0] = 0x00;
10528         feature = (struct scsi_get_config_feature *)
10529             &feature->feature_data[feature->add_length];
10530
10531 done:
10532         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10533         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10534                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10535                 if (scsi_2btoul(feature->feature_code) == starting)
10536                         feature = (struct scsi_get_config_feature *)
10537                             &feature->feature_data[feature->add_length];
10538                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10539         }
10540         scsi_ulto4b(data_len - 4, hdr->data_length);
10541         if (data_len < alloc_len) {
10542                 ctsio->residual = alloc_len - data_len;
10543                 ctsio->kern_data_len = data_len;
10544                 ctsio->kern_total_len = data_len;
10545         } else {
10546                 ctsio->residual = 0;
10547                 ctsio->kern_data_len = alloc_len;
10548                 ctsio->kern_total_len = alloc_len;
10549         }
10550
10551         ctl_set_success(ctsio);
10552         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10553         ctsio->be_move_done = ctl_config_move_done;
10554         ctl_datamove((union ctl_io *)ctsio);
10555         return (CTL_RETVAL_COMPLETE);
10556 }
10557
10558 int
10559 ctl_get_event_status(struct ctl_scsiio *ctsio)
10560 {
10561         struct scsi_get_event_status_header *hdr;
10562         struct scsi_get_event_status *cdb;
10563         uint32_t alloc_len, data_len;
10564         int notif_class;
10565
10566         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10567         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10568                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10569                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10570                 ctl_done((union ctl_io *)ctsio);
10571                 return (CTL_RETVAL_COMPLETE);
10572         }
10573         notif_class = cdb->notif_class;
10574         alloc_len = scsi_2btoul(cdb->length);
10575
10576         data_len = sizeof(struct scsi_get_event_status_header);
10577         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10578         ctsio->kern_sg_entries = 0;
10579         ctsio->kern_data_resid = 0;
10580         ctsio->kern_rel_offset = 0;
10581
10582         if (data_len < alloc_len) {
10583                 ctsio->residual = alloc_len - data_len;
10584                 ctsio->kern_data_len = data_len;
10585                 ctsio->kern_total_len = data_len;
10586         } else {
10587                 ctsio->residual = 0;
10588                 ctsio->kern_data_len = alloc_len;
10589                 ctsio->kern_total_len = alloc_len;
10590         }
10591
10592         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10593         scsi_ulto2b(0, hdr->descr_length);
10594         hdr->nea_class = SGESN_NEA;
10595         hdr->supported_class = 0;
10596
10597         ctl_set_success(ctsio);
10598         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10599         ctsio->be_move_done = ctl_config_move_done;
10600         ctl_datamove((union ctl_io *)ctsio);
10601         return (CTL_RETVAL_COMPLETE);
10602 }
10603
10604 int
10605 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10606 {
10607         struct scsi_mechanism_status_header *hdr;
10608         struct scsi_mechanism_status *cdb;
10609         uint32_t alloc_len, data_len;
10610
10611         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10612         alloc_len = scsi_2btoul(cdb->length);
10613
10614         data_len = sizeof(struct scsi_mechanism_status_header);
10615         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10616         ctsio->kern_sg_entries = 0;
10617         ctsio->kern_data_resid = 0;
10618         ctsio->kern_rel_offset = 0;
10619
10620         if (data_len < alloc_len) {
10621                 ctsio->residual = alloc_len - data_len;
10622                 ctsio->kern_data_len = data_len;
10623                 ctsio->kern_total_len = data_len;
10624         } else {
10625                 ctsio->residual = 0;
10626                 ctsio->kern_data_len = alloc_len;
10627                 ctsio->kern_total_len = alloc_len;
10628         }
10629
10630         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10631         hdr->state1 = 0x00;
10632         hdr->state2 = 0xe0;
10633         scsi_ulto3b(0, hdr->lba);
10634         hdr->slots_num = 0;
10635         scsi_ulto2b(0, hdr->slots_length);
10636
10637         ctl_set_success(ctsio);
10638         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10639         ctsio->be_move_done = ctl_config_move_done;
10640         ctl_datamove((union ctl_io *)ctsio);
10641         return (CTL_RETVAL_COMPLETE);
10642 }
10643
10644 static void
10645 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10646 {
10647
10648         lba += 150;
10649         buf[0] = 0;
10650         buf[1] = bin2bcd((lba / 75) / 60);
10651         buf[2] = bin2bcd((lba / 75) % 60);
10652         buf[3] = bin2bcd(lba % 75);
10653 }
10654
10655 int
10656 ctl_read_toc(struct ctl_scsiio *ctsio)
10657 {
10658         struct ctl_lun *lun = CTL_LUN(ctsio);
10659         struct scsi_read_toc_hdr *hdr;
10660         struct scsi_read_toc_type01_descr *descr;
10661         struct scsi_read_toc *cdb;
10662         uint32_t alloc_len, data_len;
10663         int format, msf;
10664
10665         cdb = (struct scsi_read_toc *)ctsio->cdb;
10666         msf = (cdb->byte2 & CD_MSF) != 0;
10667         format = cdb->format;
10668         alloc_len = scsi_2btoul(cdb->data_len);
10669
10670         data_len = sizeof(struct scsi_read_toc_hdr);
10671         if (format == 0)
10672                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10673         else
10674                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10675         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10676         ctsio->kern_sg_entries = 0;
10677         ctsio->kern_data_resid = 0;
10678         ctsio->kern_rel_offset = 0;
10679
10680         if (data_len < alloc_len) {
10681                 ctsio->residual = alloc_len - data_len;
10682                 ctsio->kern_data_len = data_len;
10683                 ctsio->kern_total_len = data_len;
10684         } else {
10685                 ctsio->residual = 0;
10686                 ctsio->kern_data_len = alloc_len;
10687                 ctsio->kern_total_len = alloc_len;
10688         }
10689
10690         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10691         if (format == 0) {
10692                 scsi_ulto2b(0x12, hdr->data_length);
10693                 hdr->first = 1;
10694                 hdr->last = 1;
10695                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10696                 descr->addr_ctl = 0x14;
10697                 descr->track_number = 1;
10698                 if (msf)
10699                         ctl_ultomsf(0, descr->track_start);
10700                 else
10701                         scsi_ulto4b(0, descr->track_start);
10702                 descr++;
10703                 descr->addr_ctl = 0x14;
10704                 descr->track_number = 0xaa;
10705                 if (msf)
10706                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10707                 else
10708                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10709         } else {
10710                 scsi_ulto2b(0x0a, hdr->data_length);
10711                 hdr->first = 1;
10712                 hdr->last = 1;
10713                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10714                 descr->addr_ctl = 0x14;
10715                 descr->track_number = 1;
10716                 if (msf)
10717                         ctl_ultomsf(0, descr->track_start);
10718                 else
10719                         scsi_ulto4b(0, descr->track_start);
10720         }
10721
10722         ctl_set_success(ctsio);
10723         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10724         ctsio->be_move_done = ctl_config_move_done;
10725         ctl_datamove((union ctl_io *)ctsio);
10726         return (CTL_RETVAL_COMPLETE);
10727 }
10728
10729 /*
10730  * For known CDB types, parse the LBA and length.
10731  */
10732 static int
10733 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10734 {
10735         if (io->io_hdr.io_type != CTL_IO_SCSI)
10736                 return (1);
10737
10738         switch (io->scsiio.cdb[0]) {
10739         case COMPARE_AND_WRITE: {
10740                 struct scsi_compare_and_write *cdb;
10741
10742                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10743
10744                 *lba = scsi_8btou64(cdb->addr);
10745                 *len = cdb->length;
10746                 break;
10747         }
10748         case READ_6:
10749         case WRITE_6: {
10750                 struct scsi_rw_6 *cdb;
10751
10752                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10753
10754                 *lba = scsi_3btoul(cdb->addr);
10755                 /* only 5 bits are valid in the most significant address byte */
10756                 *lba &= 0x1fffff;
10757                 *len = cdb->length;
10758                 break;
10759         }
10760         case READ_10:
10761         case WRITE_10: {
10762                 struct scsi_rw_10 *cdb;
10763
10764                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10765
10766                 *lba = scsi_4btoul(cdb->addr);
10767                 *len = scsi_2btoul(cdb->length);
10768                 break;
10769         }
10770         case WRITE_VERIFY_10: {
10771                 struct scsi_write_verify_10 *cdb;
10772
10773                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10774
10775                 *lba = scsi_4btoul(cdb->addr);
10776                 *len = scsi_2btoul(cdb->length);
10777                 break;
10778         }
10779         case READ_12:
10780         case WRITE_12: {
10781                 struct scsi_rw_12 *cdb;
10782
10783                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10784
10785                 *lba = scsi_4btoul(cdb->addr);
10786                 *len = scsi_4btoul(cdb->length);
10787                 break;
10788         }
10789         case WRITE_VERIFY_12: {
10790                 struct scsi_write_verify_12 *cdb;
10791
10792                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10793
10794                 *lba = scsi_4btoul(cdb->addr);
10795                 *len = scsi_4btoul(cdb->length);
10796                 break;
10797         }
10798         case READ_16:
10799         case WRITE_16: {
10800                 struct scsi_rw_16 *cdb;
10801
10802                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10803
10804                 *lba = scsi_8btou64(cdb->addr);
10805                 *len = scsi_4btoul(cdb->length);
10806                 break;
10807         }
10808         case WRITE_ATOMIC_16: {
10809                 struct scsi_write_atomic_16 *cdb;
10810
10811                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10812
10813                 *lba = scsi_8btou64(cdb->addr);
10814                 *len = scsi_2btoul(cdb->length);
10815                 break;
10816         }
10817         case WRITE_VERIFY_16: {
10818                 struct scsi_write_verify_16 *cdb;
10819
10820                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10821
10822                 *lba = scsi_8btou64(cdb->addr);
10823                 *len = scsi_4btoul(cdb->length);
10824                 break;
10825         }
10826         case WRITE_SAME_10: {
10827                 struct scsi_write_same_10 *cdb;
10828
10829                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10830
10831                 *lba = scsi_4btoul(cdb->addr);
10832                 *len = scsi_2btoul(cdb->length);
10833                 break;
10834         }
10835         case WRITE_SAME_16: {
10836                 struct scsi_write_same_16 *cdb;
10837
10838                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10839
10840                 *lba = scsi_8btou64(cdb->addr);
10841                 *len = scsi_4btoul(cdb->length);
10842                 break;
10843         }
10844         case VERIFY_10: {
10845                 struct scsi_verify_10 *cdb;
10846
10847                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10848
10849                 *lba = scsi_4btoul(cdb->addr);
10850                 *len = scsi_2btoul(cdb->length);
10851                 break;
10852         }
10853         case VERIFY_12: {
10854                 struct scsi_verify_12 *cdb;
10855
10856                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10857
10858                 *lba = scsi_4btoul(cdb->addr);
10859                 *len = scsi_4btoul(cdb->length);
10860                 break;
10861         }
10862         case VERIFY_16: {
10863                 struct scsi_verify_16 *cdb;
10864
10865                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10866
10867                 *lba = scsi_8btou64(cdb->addr);
10868                 *len = scsi_4btoul(cdb->length);
10869                 break;
10870         }
10871         case UNMAP: {
10872                 *lba = 0;
10873                 *len = UINT64_MAX;
10874                 break;
10875         }
10876         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10877                 struct scsi_get_lba_status *cdb;
10878
10879                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10880                 *lba = scsi_8btou64(cdb->addr);
10881                 *len = UINT32_MAX;
10882                 break;
10883         }
10884         default:
10885                 return (1);
10886                 break; /* NOTREACHED */
10887         }
10888
10889         return (0);
10890 }
10891
10892 static ctl_action
10893 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10894     bool seq)
10895 {
10896         uint64_t endlba1, endlba2;
10897
10898         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10899         endlba2 = lba2 + len2 - 1;
10900
10901         if ((endlba1 < lba2) || (endlba2 < lba1))
10902                 return (CTL_ACTION_PASS);
10903         else
10904                 return (CTL_ACTION_BLOCK);
10905 }
10906
10907 static int
10908 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10909 {
10910         struct ctl_ptr_len_flags *ptrlen;
10911         struct scsi_unmap_desc *buf, *end, *range;
10912         uint64_t lba;
10913         uint32_t len;
10914
10915         /* If not UNMAP -- go other way. */
10916         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10917             io->scsiio.cdb[0] != UNMAP)
10918                 return (CTL_ACTION_ERROR);
10919
10920         /* If UNMAP without data -- block and wait for data. */
10921         ptrlen = (struct ctl_ptr_len_flags *)
10922             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10923         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10924             ptrlen->ptr == NULL)
10925                 return (CTL_ACTION_BLOCK);
10926
10927         /* UNMAP with data -- check for collision. */
10928         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10929         end = buf + ptrlen->len / sizeof(*buf);
10930         for (range = buf; range < end; range++) {
10931                 lba = scsi_8btou64(range->lba);
10932                 len = scsi_4btoul(range->length);
10933                 if ((lba < lba2 + len2) && (lba + len > lba2))
10934                         return (CTL_ACTION_BLOCK);
10935         }
10936         return (CTL_ACTION_PASS);
10937 }
10938
10939 static ctl_action
10940 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10941 {
10942         uint64_t lba1, lba2;
10943         uint64_t len1, len2;
10944         int retval;
10945
10946         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10947                 return (CTL_ACTION_ERROR);
10948
10949         retval = ctl_extent_check_unmap(io1, lba2, len2);
10950         if (retval != CTL_ACTION_ERROR)
10951                 return (retval);
10952
10953         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10954                 return (CTL_ACTION_ERROR);
10955
10956         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10957                 seq = FALSE;
10958         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10959 }
10960
10961 static ctl_action
10962 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10963 {
10964         uint64_t lba1, lba2;
10965         uint64_t len1, len2;
10966
10967         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10968                 return (CTL_ACTION_PASS);
10969         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10970                 return (CTL_ACTION_ERROR);
10971         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10972                 return (CTL_ACTION_ERROR);
10973
10974         if (lba1 + len1 == lba2)
10975                 return (CTL_ACTION_BLOCK);
10976         return (CTL_ACTION_PASS);
10977 }
10978
10979 static ctl_action
10980 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10981     union ctl_io *ooa_io)
10982 {
10983         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10984         const ctl_serialize_action *serialize_row;
10985
10986         /*
10987          * The initiator attempted multiple untagged commands at the same
10988          * time.  Can't do that.
10989          */
10990         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10991          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10992          && ((pending_io->io_hdr.nexus.targ_port ==
10993               ooa_io->io_hdr.nexus.targ_port)
10994           && (pending_io->io_hdr.nexus.initid ==
10995               ooa_io->io_hdr.nexus.initid))
10996          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10997               CTL_FLAG_STATUS_SENT)) == 0))
10998                 return (CTL_ACTION_OVERLAP);
10999
11000         /*
11001          * The initiator attempted to send multiple tagged commands with
11002          * the same ID.  (It's fine if different initiators have the same
11003          * tag ID.)
11004          *
11005          * Even if all of those conditions are true, we don't kill the I/O
11006          * if the command ahead of us has been aborted.  We won't end up
11007          * sending it to the FETD, and it's perfectly legal to resend a
11008          * command with the same tag number as long as the previous
11009          * instance of this tag number has been aborted somehow.
11010          */
11011         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11012          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11013          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11014          && ((pending_io->io_hdr.nexus.targ_port ==
11015               ooa_io->io_hdr.nexus.targ_port)
11016           && (pending_io->io_hdr.nexus.initid ==
11017               ooa_io->io_hdr.nexus.initid))
11018          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11019               CTL_FLAG_STATUS_SENT)) == 0))
11020                 return (CTL_ACTION_OVERLAP_TAG);
11021
11022         /*
11023          * If we get a head of queue tag, SAM-3 says that we should
11024          * immediately execute it.
11025          *
11026          * What happens if this command would normally block for some other
11027          * reason?  e.g. a request sense with a head of queue tag
11028          * immediately after a write.  Normally that would block, but this
11029          * will result in its getting executed immediately...
11030          *
11031          * We currently return "pass" instead of "skip", so we'll end up
11032          * going through the rest of the queue to check for overlapped tags.
11033          *
11034          * XXX KDM check for other types of blockage first??
11035          */
11036         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11037                 return (CTL_ACTION_PASS);
11038
11039         /*
11040          * Ordered tags have to block until all items ahead of them
11041          * have completed.  If we get called with an ordered tag, we always
11042          * block, if something else is ahead of us in the queue.
11043          */
11044         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11045                 return (CTL_ACTION_BLOCK);
11046
11047         /*
11048          * Simple tags get blocked until all head of queue and ordered tags
11049          * ahead of them have completed.  I'm lumping untagged commands in
11050          * with simple tags here.  XXX KDM is that the right thing to do?
11051          */
11052         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11053           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11054          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11055           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11056                 return (CTL_ACTION_BLOCK);
11057
11058         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11059         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
11060             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
11061              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
11062              pending_io->scsiio.cdb[1], pending_io));
11063         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11064         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
11065                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
11066         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
11067             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
11068              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
11069              ooa_io->scsiio.cdb[1], ooa_io));
11070
11071         serialize_row = ctl_serialize_table[ooa_entry->seridx];
11072
11073         switch (serialize_row[pending_entry->seridx]) {
11074         case CTL_SER_BLOCK:
11075                 return (CTL_ACTION_BLOCK);
11076         case CTL_SER_EXTENT:
11077                 return (ctl_extent_check(ooa_io, pending_io,
11078                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11079         case CTL_SER_EXTENTOPT:
11080                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
11081                     SCP_QUEUE_ALG_UNRESTRICTED)
11082                         return (ctl_extent_check(ooa_io, pending_io,
11083                             (lun->be_lun &&
11084                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11085                 return (CTL_ACTION_PASS);
11086         case CTL_SER_EXTENTSEQ:
11087                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
11088                         return (ctl_extent_check_seq(ooa_io, pending_io));
11089                 return (CTL_ACTION_PASS);
11090         case CTL_SER_PASS:
11091                 return (CTL_ACTION_PASS);
11092         case CTL_SER_BLOCKOPT:
11093                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
11094                     SCP_QUEUE_ALG_UNRESTRICTED)
11095                         return (CTL_ACTION_BLOCK);
11096                 return (CTL_ACTION_PASS);
11097         case CTL_SER_SKIP:
11098                 return (CTL_ACTION_SKIP);
11099         default:
11100                 panic("%s: Invalid serialization value %d for %d => %d",
11101                     __func__, serialize_row[pending_entry->seridx],
11102                     pending_entry->seridx, ooa_entry->seridx);
11103         }
11104
11105         return (CTL_ACTION_ERROR);
11106 }
11107
11108 /*
11109  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11110  * Assumptions:
11111  * - pending_io is generally either incoming, or on the blocked queue
11112  * - starting I/O is the I/O we want to start the check with.
11113  */
11114 static ctl_action
11115 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11116               union ctl_io *starting_io)
11117 {
11118         union ctl_io *ooa_io;
11119         ctl_action action;
11120
11121         mtx_assert(&lun->lun_lock, MA_OWNED);
11122
11123         /*
11124          * Run back along the OOA queue, starting with the current
11125          * blocked I/O and going through every I/O before it on the
11126          * queue.  If starting_io is NULL, we'll just end up returning
11127          * CTL_ACTION_PASS.
11128          */
11129         for (ooa_io = starting_io; ooa_io != NULL;
11130              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11131              ooa_links)){
11132
11133                 /*
11134                  * This routine just checks to see whether
11135                  * cur_blocked is blocked by ooa_io, which is ahead
11136                  * of it in the queue.  It doesn't queue/dequeue
11137                  * cur_blocked.
11138                  */
11139                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11140                 switch (action) {
11141                 case CTL_ACTION_BLOCK:
11142                 case CTL_ACTION_OVERLAP:
11143                 case CTL_ACTION_OVERLAP_TAG:
11144                 case CTL_ACTION_SKIP:
11145                 case CTL_ACTION_ERROR:
11146                         return (action);
11147                         break; /* NOTREACHED */
11148                 case CTL_ACTION_PASS:
11149                         break;
11150                 default:
11151                         panic("%s: Invalid action %d\n", __func__, action);
11152                 }
11153         }
11154
11155         return (CTL_ACTION_PASS);
11156 }
11157
11158 /*
11159  * Assumptions:
11160  * - An I/O has just completed, and has been removed from the per-LUN OOA
11161  *   queue, so some items on the blocked queue may now be unblocked.
11162  */
11163 static int
11164 ctl_check_blocked(struct ctl_lun *lun)
11165 {
11166         struct ctl_softc *softc = lun->ctl_softc;
11167         union ctl_io *cur_blocked, *next_blocked;
11168
11169         mtx_assert(&lun->lun_lock, MA_OWNED);
11170
11171         /*
11172          * Run forward from the head of the blocked queue, checking each
11173          * entry against the I/Os prior to it on the OOA queue to see if
11174          * there is still any blockage.
11175          *
11176          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11177          * with our removing a variable on it while it is traversing the
11178          * list.
11179          */
11180         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11181              cur_blocked != NULL; cur_blocked = next_blocked) {
11182                 union ctl_io *prev_ooa;
11183                 ctl_action action;
11184
11185                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11186                                                           blocked_links);
11187
11188                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11189                                                       ctl_ooaq, ooa_links);
11190
11191                 /*
11192                  * If cur_blocked happens to be the first item in the OOA
11193                  * queue now, prev_ooa will be NULL, and the action
11194                  * returned will just be CTL_ACTION_PASS.
11195                  */
11196                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11197
11198                 switch (action) {
11199                 case CTL_ACTION_BLOCK:
11200                         /* Nothing to do here, still blocked */
11201                         break;
11202                 case CTL_ACTION_OVERLAP:
11203                 case CTL_ACTION_OVERLAP_TAG:
11204                         /*
11205                          * This shouldn't happen!  In theory we've already
11206                          * checked this command for overlap...
11207                          */
11208                         break;
11209                 case CTL_ACTION_PASS:
11210                 case CTL_ACTION_SKIP: {
11211                         const struct ctl_cmd_entry *entry;
11212
11213                         /*
11214                          * The skip case shouldn't happen, this transaction
11215                          * should have never made it onto the blocked queue.
11216                          */
11217                         /*
11218                          * This I/O is no longer blocked, we can remove it
11219                          * from the blocked queue.  Since this is a TAILQ
11220                          * (doubly linked list), we can do O(1) removals
11221                          * from any place on the list.
11222                          */
11223                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11224                                      blocked_links);
11225                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11226
11227                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11228                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11229                                 /*
11230                                  * Need to send IO back to original side to
11231                                  * run
11232                                  */
11233                                 union ctl_ha_msg msg_info;
11234
11235                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11236                                 msg_info.hdr.original_sc =
11237                                         cur_blocked->io_hdr.original_sc;
11238                                 msg_info.hdr.serializing_sc = cur_blocked;
11239                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11240                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11241                                     sizeof(msg_info.hdr), M_NOWAIT);
11242                                 break;
11243                         }
11244                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11245
11246                         /*
11247                          * Check this I/O for LUN state changes that may
11248                          * have happened while this command was blocked.
11249                          * The LUN state may have been changed by a command
11250                          * ahead of us in the queue, so we need to re-check
11251                          * for any states that can be caused by SCSI
11252                          * commands.
11253                          */
11254                         if (ctl_scsiio_lun_check(lun, entry,
11255                                                  &cur_blocked->scsiio) == 0) {
11256                                 cur_blocked->io_hdr.flags |=
11257                                                       CTL_FLAG_IS_WAS_ON_RTR;
11258                                 ctl_enqueue_rtr(cur_blocked);
11259                         } else
11260                                 ctl_done(cur_blocked);
11261                         break;
11262                 }
11263                 default:
11264                         /*
11265                          * This probably shouldn't happen -- we shouldn't
11266                          * get CTL_ACTION_ERROR, or anything else.
11267                          */
11268                         break;
11269                 }
11270         }
11271
11272         return (CTL_RETVAL_COMPLETE);
11273 }
11274
11275 /*
11276  * This routine (with one exception) checks LUN flags that can be set by
11277  * commands ahead of us in the OOA queue.  These flags have to be checked
11278  * when a command initially comes in, and when we pull a command off the
11279  * blocked queue and are preparing to execute it.  The reason we have to
11280  * check these flags for commands on the blocked queue is that the LUN
11281  * state may have been changed by a command ahead of us while we're on the
11282  * blocked queue.
11283  *
11284  * Ordering is somewhat important with these checks, so please pay
11285  * careful attention to the placement of any new checks.
11286  */
11287 static int
11288 ctl_scsiio_lun_check(struct ctl_lun *lun,
11289     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11290 {
11291         struct ctl_softc *softc = lun->ctl_softc;
11292         int retval;
11293         uint32_t residx;
11294
11295         retval = 0;
11296
11297         mtx_assert(&lun->lun_lock, MA_OWNED);
11298
11299         /*
11300          * If this shelf is a secondary shelf controller, we may have to
11301          * reject some commands disallowed by HA mode and link state.
11302          */
11303         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11304                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11305                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11306                         ctl_set_lun_unavail(ctsio);
11307                         retval = 1;
11308                         goto bailout;
11309                 }
11310                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11311                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11312                         ctl_set_lun_transit(ctsio);
11313                         retval = 1;
11314                         goto bailout;
11315                 }
11316                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11317                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11318                         ctl_set_lun_standby(ctsio);
11319                         retval = 1;
11320                         goto bailout;
11321                 }
11322
11323                 /* The rest of checks are only done on executing side */
11324                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11325                         goto bailout;
11326         }
11327
11328         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11329                 if (lun->be_lun &&
11330                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11331                         ctl_set_hw_write_protected(ctsio);
11332                         retval = 1;
11333                         goto bailout;
11334                 }
11335                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11336                         ctl_set_sense(ctsio, /*current_error*/ 1,
11337                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11338                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11339                         retval = 1;
11340                         goto bailout;
11341                 }
11342         }
11343
11344         /*
11345          * Check for a reservation conflict.  If this command isn't allowed
11346          * even on reserved LUNs, and if this initiator isn't the one who
11347          * reserved us, reject the command with a reservation conflict.
11348          */
11349         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11350         if ((lun->flags & CTL_LUN_RESERVED)
11351          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11352                 if (lun->res_idx != residx) {
11353                         ctl_set_reservation_conflict(ctsio);
11354                         retval = 1;
11355                         goto bailout;
11356                 }
11357         }
11358
11359         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11360             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11361                 /* No reservation or command is allowed. */;
11362         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11363             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11364              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11365              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11366                 /* The command is allowed for Write Exclusive resv. */;
11367         } else {
11368                 /*
11369                  * if we aren't registered or it's a res holder type
11370                  * reservation and this isn't the res holder then set a
11371                  * conflict.
11372                  */
11373                 if (ctl_get_prkey(lun, residx) == 0 ||
11374                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11375                         ctl_set_reservation_conflict(ctsio);
11376                         retval = 1;
11377                         goto bailout;
11378                 }
11379         }
11380
11381         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11382                 if (lun->flags & CTL_LUN_EJECTED)
11383                         ctl_set_lun_ejected(ctsio);
11384                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11385                         if (lun->flags & CTL_LUN_REMOVABLE)
11386                                 ctl_set_lun_no_media(ctsio);
11387                         else
11388                                 ctl_set_lun_int_reqd(ctsio);
11389                 } else if (lun->flags & CTL_LUN_STOPPED)
11390                         ctl_set_lun_stopped(ctsio);
11391                 else
11392                         goto bailout;
11393                 retval = 1;
11394                 goto bailout;
11395         }
11396
11397 bailout:
11398         return (retval);
11399 }
11400
11401 static void
11402 ctl_failover_io(union ctl_io *io, int have_lock)
11403 {
11404         ctl_set_busy(&io->scsiio);
11405         ctl_done(io);
11406 }
11407
11408 static void
11409 ctl_failover_lun(union ctl_io *rio)
11410 {
11411         struct ctl_softc *softc = CTL_SOFTC(rio);
11412         struct ctl_lun *lun;
11413         struct ctl_io_hdr *io, *next_io;
11414         uint32_t targ_lun;
11415
11416         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11417         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11418
11419         /* Find and lock the LUN. */
11420         mtx_lock(&softc->ctl_lock);
11421         if (targ_lun > CTL_MAX_LUNS ||
11422             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11423                 mtx_unlock(&softc->ctl_lock);
11424                 return;
11425         }
11426         mtx_lock(&lun->lun_lock);
11427         mtx_unlock(&softc->ctl_lock);
11428         if (lun->flags & CTL_LUN_DISABLED) {
11429                 mtx_unlock(&lun->lun_lock);
11430                 return;
11431         }
11432
11433         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11434                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11435                         /* We are master */
11436                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11437                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11438                                         io->flags |= CTL_FLAG_ABORT;
11439                                         io->flags |= CTL_FLAG_FAILOVER;
11440                                 } else { /* This can be only due to DATAMOVE */
11441                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11442                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11443                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11444                                         io->port_status = 31340;
11445                                         ctl_enqueue_isc((union ctl_io *)io);
11446                                 }
11447                         }
11448                         /* We are slave */
11449                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11450                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11451                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11452                                         io->flags |= CTL_FLAG_FAILOVER;
11453                                 } else {
11454                                         ctl_set_busy(&((union ctl_io *)io)->
11455                                             scsiio);
11456                                         ctl_done((union ctl_io *)io);
11457                                 }
11458                         }
11459                 }
11460         } else { /* SERIALIZE modes */
11461                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11462                     next_io) {
11463                         /* We are master */
11464                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11465                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11466                                     blocked_links);
11467                                 io->flags &= ~CTL_FLAG_BLOCKED;
11468                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11469                                 ctl_free_io((union ctl_io *)io);
11470                         }
11471                 }
11472                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11473                         /* We are master */
11474                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11475                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11476                                 ctl_free_io((union ctl_io *)io);
11477                         }
11478                         /* We are slave */
11479                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11480                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11481                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11482                                         ctl_set_busy(&((union ctl_io *)io)->
11483                                             scsiio);
11484                                         ctl_done((union ctl_io *)io);
11485                                 }
11486                         }
11487                 }
11488                 ctl_check_blocked(lun);
11489         }
11490         mtx_unlock(&lun->lun_lock);
11491 }
11492
11493 static int
11494 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11495 {
11496         struct ctl_lun *lun;
11497         const struct ctl_cmd_entry *entry;
11498         uint32_t initidx, targ_lun;
11499         int retval = 0;
11500
11501         lun = NULL;
11502         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11503         if (targ_lun < CTL_MAX_LUNS)
11504                 lun = softc->ctl_luns[targ_lun];
11505         if (lun) {
11506                 /*
11507                  * If the LUN is invalid, pretend that it doesn't exist.
11508                  * It will go away as soon as all pending I/O has been
11509                  * completed.
11510                  */
11511                 mtx_lock(&lun->lun_lock);
11512                 if (lun->flags & CTL_LUN_DISABLED) {
11513                         mtx_unlock(&lun->lun_lock);
11514                         lun = NULL;
11515                 }
11516         }
11517         CTL_LUN(ctsio) = lun;
11518         if (lun) {
11519                 CTL_BACKEND_LUN(ctsio) = lun->be_lun;
11520
11521                 /*
11522                  * Every I/O goes into the OOA queue for a particular LUN,
11523                  * and stays there until completion.
11524                  */
11525 #ifdef CTL_TIME_IO
11526                 if (TAILQ_EMPTY(&lun->ooa_queue))
11527                         lun->idle_time += getsbinuptime() - lun->last_busy;
11528 #endif
11529                 TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11530         }
11531
11532         /* Get command entry and return error if it is unsuppotyed. */
11533         entry = ctl_validate_command(ctsio);
11534         if (entry == NULL) {
11535                 if (lun)
11536                         mtx_unlock(&lun->lun_lock);
11537                 return (retval);
11538         }
11539
11540         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11541         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11542
11543         /*
11544          * Check to see whether we can send this command to LUNs that don't
11545          * exist.  This should pretty much only be the case for inquiry
11546          * and request sense.  Further checks, below, really require having
11547          * a LUN, so we can't really check the command anymore.  Just put
11548          * it on the rtr queue.
11549          */
11550         if (lun == NULL) {
11551                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11552                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11553                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11554                         return (retval);
11555                 }
11556
11557                 ctl_set_unsupported_lun(ctsio);
11558                 ctl_done((union ctl_io *)ctsio);
11559                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11560                 return (retval);
11561         } else {
11562                 /*
11563                  * Make sure we support this particular command on this LUN.
11564                  * e.g., we don't support writes to the control LUN.
11565                  */
11566                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11567                         mtx_unlock(&lun->lun_lock);
11568                         ctl_set_invalid_opcode(ctsio);
11569                         ctl_done((union ctl_io *)ctsio);
11570                         return (retval);
11571                 }
11572         }
11573
11574         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11575
11576 #ifdef CTL_WITH_CA
11577         /*
11578          * If we've got a request sense, it'll clear the contingent
11579          * allegiance condition.  Otherwise, if we have a CA condition for
11580          * this initiator, clear it, because it sent down a command other
11581          * than request sense.
11582          */
11583         if ((ctsio->cdb[0] != REQUEST_SENSE)
11584          && (ctl_is_set(lun->have_ca, initidx)))
11585                 ctl_clear_mask(lun->have_ca, initidx);
11586 #endif
11587
11588         /*
11589          * If the command has this flag set, it handles its own unit
11590          * attention reporting, we shouldn't do anything.  Otherwise we
11591          * check for any pending unit attentions, and send them back to the
11592          * initiator.  We only do this when a command initially comes in,
11593          * not when we pull it off the blocked queue.
11594          *
11595          * According to SAM-3, section 5.3.2, the order that things get
11596          * presented back to the host is basically unit attentions caused
11597          * by some sort of reset event, busy status, reservation conflicts
11598          * or task set full, and finally any other status.
11599          *
11600          * One issue here is that some of the unit attentions we report
11601          * don't fall into the "reset" category (e.g. "reported luns data
11602          * has changed").  So reporting it here, before the reservation
11603          * check, may be technically wrong.  I guess the only thing to do
11604          * would be to check for and report the reset events here, and then
11605          * check for the other unit attention types after we check for a
11606          * reservation conflict.
11607          *
11608          * XXX KDM need to fix this
11609          */
11610         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11611                 ctl_ua_type ua_type;
11612                 u_int sense_len = 0;
11613
11614                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11615                     &sense_len, SSD_TYPE_NONE);
11616                 if (ua_type != CTL_UA_NONE) {
11617                         mtx_unlock(&lun->lun_lock);
11618                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11619                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11620                         ctsio->sense_len = sense_len;
11621                         ctl_done((union ctl_io *)ctsio);
11622                         return (retval);
11623                 }
11624         }
11625
11626
11627         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11628                 mtx_unlock(&lun->lun_lock);
11629                 ctl_done((union ctl_io *)ctsio);
11630                 return (retval);
11631         }
11632
11633         /*
11634          * XXX CHD this is where we want to send IO to other side if
11635          * this LUN is secondary on this SC. We will need to make a copy
11636          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11637          * the copy we send as FROM_OTHER.
11638          * We also need to stuff the address of the original IO so we can
11639          * find it easily. Something similar will need be done on the other
11640          * side so when we are done we can find the copy.
11641          */
11642         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11643             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11644             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11645                 union ctl_ha_msg msg_info;
11646                 int isc_retval;
11647
11648                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11649                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11650                 mtx_unlock(&lun->lun_lock);
11651
11652                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11653                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11654                 msg_info.hdr.serializing_sc = NULL;
11655                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11656                 msg_info.scsi.tag_num = ctsio->tag_num;
11657                 msg_info.scsi.tag_type = ctsio->tag_type;
11658                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11659                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11660
11661                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11662                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11663                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11664                         ctl_set_busy(ctsio);
11665                         ctl_done((union ctl_io *)ctsio);
11666                         return (retval);
11667                 }
11668                 return (retval);
11669         }
11670
11671         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11672                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11673                               ctl_ooaq, ooa_links))) {
11674         case CTL_ACTION_BLOCK:
11675                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11676                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11677                                   blocked_links);
11678                 mtx_unlock(&lun->lun_lock);
11679                 return (retval);
11680         case CTL_ACTION_PASS:
11681         case CTL_ACTION_SKIP:
11682                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11683                 mtx_unlock(&lun->lun_lock);
11684                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11685                 break;
11686         case CTL_ACTION_OVERLAP:
11687                 mtx_unlock(&lun->lun_lock);
11688                 ctl_set_overlapped_cmd(ctsio);
11689                 ctl_done((union ctl_io *)ctsio);
11690                 break;
11691         case CTL_ACTION_OVERLAP_TAG:
11692                 mtx_unlock(&lun->lun_lock);
11693                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11694                 ctl_done((union ctl_io *)ctsio);
11695                 break;
11696         case CTL_ACTION_ERROR:
11697         default:
11698                 mtx_unlock(&lun->lun_lock);
11699                 ctl_set_internal_failure(ctsio,
11700                                          /*sks_valid*/ 0,
11701                                          /*retry_count*/ 0);
11702                 ctl_done((union ctl_io *)ctsio);
11703                 break;
11704         }
11705         return (retval);
11706 }
11707
11708 const struct ctl_cmd_entry *
11709 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11710 {
11711         const struct ctl_cmd_entry *entry;
11712         int service_action;
11713
11714         entry = &ctl_cmd_table[ctsio->cdb[0]];
11715         if (sa)
11716                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11717         if (entry->flags & CTL_CMD_FLAG_SA5) {
11718                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11719                 entry = &((const struct ctl_cmd_entry *)
11720                     entry->execute)[service_action];
11721         }
11722         return (entry);
11723 }
11724
11725 const struct ctl_cmd_entry *
11726 ctl_validate_command(struct ctl_scsiio *ctsio)
11727 {
11728         const struct ctl_cmd_entry *entry;
11729         int i, sa;
11730         uint8_t diff;
11731
11732         entry = ctl_get_cmd_entry(ctsio, &sa);
11733         if (entry->execute == NULL) {
11734                 if (sa)
11735                         ctl_set_invalid_field(ctsio,
11736                                               /*sks_valid*/ 1,
11737                                               /*command*/ 1,
11738                                               /*field*/ 1,
11739                                               /*bit_valid*/ 1,
11740                                               /*bit*/ 4);
11741                 else
11742                         ctl_set_invalid_opcode(ctsio);
11743                 ctl_done((union ctl_io *)ctsio);
11744                 return (NULL);
11745         }
11746         KASSERT(entry->length > 0,
11747             ("Not defined length for command 0x%02x/0x%02x",
11748              ctsio->cdb[0], ctsio->cdb[1]));
11749         for (i = 1; i < entry->length; i++) {
11750                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11751                 if (diff == 0)
11752                         continue;
11753                 ctl_set_invalid_field(ctsio,
11754                                       /*sks_valid*/ 1,
11755                                       /*command*/ 1,
11756                                       /*field*/ i,
11757                                       /*bit_valid*/ 1,
11758                                       /*bit*/ fls(diff) - 1);
11759                 ctl_done((union ctl_io *)ctsio);
11760                 return (NULL);
11761         }
11762         return (entry);
11763 }
11764
11765 static int
11766 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11767 {
11768
11769         switch (lun_type) {
11770         case T_DIRECT:
11771                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11772                         return (0);
11773                 break;
11774         case T_PROCESSOR:
11775                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11776                         return (0);
11777                 break;
11778         case T_CDROM:
11779                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11780                         return (0);
11781                 break;
11782         default:
11783                 return (0);
11784         }
11785         return (1);
11786 }
11787
11788 static int
11789 ctl_scsiio(struct ctl_scsiio *ctsio)
11790 {
11791         int retval;
11792         const struct ctl_cmd_entry *entry;
11793
11794         retval = CTL_RETVAL_COMPLETE;
11795
11796         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11797
11798         entry = ctl_get_cmd_entry(ctsio, NULL);
11799
11800         /*
11801          * If this I/O has been aborted, just send it straight to
11802          * ctl_done() without executing it.
11803          */
11804         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11805                 ctl_done((union ctl_io *)ctsio);
11806                 goto bailout;
11807         }
11808
11809         /*
11810          * All the checks should have been handled by ctl_scsiio_precheck().
11811          * We should be clear now to just execute the I/O.
11812          */
11813         retval = entry->execute(ctsio);
11814
11815 bailout:
11816         return (retval);
11817 }
11818
11819 /*
11820  * Since we only implement one target right now, a bus reset simply resets
11821  * our single target.
11822  */
11823 static int
11824 ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11825 {
11826         return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11827 }
11828
11829 static int
11830 ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11831                  ctl_ua_type ua_type)
11832 {
11833         struct ctl_port *port = CTL_PORT(io);
11834         struct ctl_lun *lun;
11835         int retval;
11836
11837         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11838                 union ctl_ha_msg msg_info;
11839
11840                 msg_info.hdr.nexus = io->io_hdr.nexus;
11841                 if (ua_type==CTL_UA_TARG_RESET)
11842                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11843                 else
11844                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11845                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11846                 msg_info.hdr.original_sc = NULL;
11847                 msg_info.hdr.serializing_sc = NULL;
11848                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11849                     sizeof(msg_info.task), M_WAITOK);
11850         }
11851         retval = 0;
11852
11853         mtx_lock(&softc->ctl_lock);
11854         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11855                 if (port != NULL &&
11856                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
11857                         continue;
11858                 retval += ctl_do_lun_reset(lun, io, ua_type);
11859         }
11860         mtx_unlock(&softc->ctl_lock);
11861         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11862         return (retval);
11863 }
11864
11865 /*
11866  * The LUN should always be set.  The I/O is optional, and is used to
11867  * distinguish between I/Os sent by this initiator, and by other
11868  * initiators.  We set unit attention for initiators other than this one.
11869  * SAM-3 is vague on this point.  It does say that a unit attention should
11870  * be established for other initiators when a LUN is reset (see section
11871  * 5.7.3), but it doesn't specifically say that the unit attention should
11872  * be established for this particular initiator when a LUN is reset.  Here
11873  * is the relevant text, from SAM-3 rev 8:
11874  *
11875  * 5.7.2 When a SCSI initiator port aborts its own tasks
11876  *
11877  * When a SCSI initiator port causes its own task(s) to be aborted, no
11878  * notification that the task(s) have been aborted shall be returned to
11879  * the SCSI initiator port other than the completion response for the
11880  * command or task management function action that caused the task(s) to
11881  * be aborted and notification(s) associated with related effects of the
11882  * action (e.g., a reset unit attention condition).
11883  *
11884  * XXX KDM for now, we're setting unit attention for all initiators.
11885  */
11886 static int
11887 ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11888 {
11889         union ctl_io *xio;
11890 #if 0
11891         uint32_t initidx;
11892 #endif
11893         int i;
11894
11895         mtx_lock(&lun->lun_lock);
11896         /*
11897          * Run through the OOA queue and abort each I/O.
11898          */
11899         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11900              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11901                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11902         }
11903
11904         /*
11905          * This version sets unit attention for every
11906          */
11907 #if 0
11908         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11909         ctl_est_ua_all(lun, initidx, ua_type);
11910 #else
11911         ctl_est_ua_all(lun, -1, ua_type);
11912 #endif
11913
11914         /*
11915          * A reset (any kind, really) clears reservations established with
11916          * RESERVE/RELEASE.  It does not clear reservations established
11917          * with PERSISTENT RESERVE OUT, but we don't support that at the
11918          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11919          * reservations made with the RESERVE/RELEASE commands, because
11920          * those commands are obsolete in SPC-3.
11921          */
11922         lun->flags &= ~CTL_LUN_RESERVED;
11923
11924 #ifdef CTL_WITH_CA
11925         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11926                 ctl_clear_mask(lun->have_ca, i);
11927 #endif
11928         lun->prevent_count = 0;
11929         if (lun->prevent) {
11930                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
11931                         ctl_clear_mask(lun->prevent, i);
11932         }
11933         mtx_unlock(&lun->lun_lock);
11934
11935         return (0);
11936 }
11937
11938 static int
11939 ctl_lun_reset(struct ctl_softc *softc, union ctl_io *io)
11940 {
11941         struct ctl_lun *lun;
11942         uint32_t targ_lun;
11943         int retval;
11944
11945         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11946         mtx_lock(&softc->ctl_lock);
11947         if (targ_lun >= CTL_MAX_LUNS ||
11948             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11949                 mtx_unlock(&softc->ctl_lock);
11950                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11951                 return (1);
11952         }
11953         retval = ctl_do_lun_reset(lun, io, CTL_UA_LUN_RESET);
11954         mtx_unlock(&softc->ctl_lock);
11955         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11956
11957         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11958                 union ctl_ha_msg msg_info;
11959
11960                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11961                 msg_info.hdr.nexus = io->io_hdr.nexus;
11962                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11963                 msg_info.hdr.original_sc = NULL;
11964                 msg_info.hdr.serializing_sc = NULL;
11965                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11966                     sizeof(msg_info.task), M_WAITOK);
11967         }
11968         return (retval);
11969 }
11970
11971 static void
11972 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11973     int other_sc)
11974 {
11975         union ctl_io *xio;
11976
11977         mtx_assert(&lun->lun_lock, MA_OWNED);
11978
11979         /*
11980          * Run through the OOA queue and attempt to find the given I/O.
11981          * The target port, initiator ID, tag type and tag number have to
11982          * match the values that we got from the initiator.  If we have an
11983          * untagged command to abort, simply abort the first untagged command
11984          * we come to.  We only allow one untagged command at a time of course.
11985          */
11986         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11987              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11988
11989                 if ((targ_port == UINT32_MAX ||
11990                      targ_port == xio->io_hdr.nexus.targ_port) &&
11991                     (init_id == UINT32_MAX ||
11992                      init_id == xio->io_hdr.nexus.initid)) {
11993                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11994                             init_id != xio->io_hdr.nexus.initid)
11995                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11996                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11997                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11998                                 union ctl_ha_msg msg_info;
11999
12000                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
12001                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12002                                 msg_info.task.tag_num = xio->scsiio.tag_num;
12003                                 msg_info.task.tag_type = xio->scsiio.tag_type;
12004                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12005                                 msg_info.hdr.original_sc = NULL;
12006                                 msg_info.hdr.serializing_sc = NULL;
12007                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12008                                     sizeof(msg_info.task), M_NOWAIT);
12009                         }
12010                 }
12011         }
12012 }
12013
12014 static int
12015 ctl_abort_task_set(union ctl_io *io)
12016 {
12017         struct ctl_softc *softc = CTL_SOFTC(io);
12018         struct ctl_lun *lun;
12019         uint32_t targ_lun;
12020
12021         /*
12022          * Look up the LUN.
12023          */
12024         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12025         mtx_lock(&softc->ctl_lock);
12026         if (targ_lun >= CTL_MAX_LUNS ||
12027             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12028                 mtx_unlock(&softc->ctl_lock);
12029                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12030                 return (1);
12031         }
12032
12033         mtx_lock(&lun->lun_lock);
12034         mtx_unlock(&softc->ctl_lock);
12035         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12036                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12037                     io->io_hdr.nexus.initid,
12038                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12039         } else { /* CTL_TASK_CLEAR_TASK_SET */
12040                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12041                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12042         }
12043         mtx_unlock(&lun->lun_lock);
12044         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12045         return (0);
12046 }
12047
12048 static int
12049 ctl_i_t_nexus_reset(union ctl_io *io)
12050 {
12051         struct ctl_softc *softc = CTL_SOFTC(io);
12052         struct ctl_lun *lun;
12053         uint32_t initidx;
12054
12055         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12056                 union ctl_ha_msg msg_info;
12057
12058                 msg_info.hdr.nexus = io->io_hdr.nexus;
12059                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
12060                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12061                 msg_info.hdr.original_sc = NULL;
12062                 msg_info.hdr.serializing_sc = NULL;
12063                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12064                     sizeof(msg_info.task), M_WAITOK);
12065         }
12066
12067         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12068         mtx_lock(&softc->ctl_lock);
12069         STAILQ_FOREACH(lun, &softc->lun_list, links) {
12070                 mtx_lock(&lun->lun_lock);
12071                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12072                     io->io_hdr.nexus.initid, 1);
12073 #ifdef CTL_WITH_CA
12074                 ctl_clear_mask(lun->have_ca, initidx);
12075 #endif
12076                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
12077                         lun->flags &= ~CTL_LUN_RESERVED;
12078                 if (lun->prevent && ctl_is_set(lun->prevent, initidx)) {
12079                         ctl_clear_mask(lun->prevent, initidx);
12080                         lun->prevent_count--;
12081                 }
12082                 ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
12083                 mtx_unlock(&lun->lun_lock);
12084         }
12085         mtx_unlock(&softc->ctl_lock);
12086         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12087         return (0);
12088 }
12089
12090 static int
12091 ctl_abort_task(union ctl_io *io)
12092 {
12093         struct ctl_softc *softc = CTL_SOFTC(io);
12094         union ctl_io *xio;
12095         struct ctl_lun *lun;
12096 #if 0
12097         struct sbuf sb;
12098         char printbuf[128];
12099 #endif
12100         int found;
12101         uint32_t targ_lun;
12102
12103         found = 0;
12104
12105         /*
12106          * Look up the LUN.
12107          */
12108         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12109         mtx_lock(&softc->ctl_lock);
12110         if (targ_lun >= CTL_MAX_LUNS ||
12111             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12112                 mtx_unlock(&softc->ctl_lock);
12113                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12114                 return (1);
12115         }
12116
12117 #if 0
12118         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12119                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12120 #endif
12121
12122         mtx_lock(&lun->lun_lock);
12123         mtx_unlock(&softc->ctl_lock);
12124         /*
12125          * Run through the OOA queue and attempt to find the given I/O.
12126          * The target port, initiator ID, tag type and tag number have to
12127          * match the values that we got from the initiator.  If we have an
12128          * untagged command to abort, simply abort the first untagged command
12129          * we come to.  We only allow one untagged command at a time of course.
12130          */
12131         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12132              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12133 #if 0
12134                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12135
12136                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12137                             lun->lun, xio->scsiio.tag_num,
12138                             xio->scsiio.tag_type,
12139                             (xio->io_hdr.blocked_links.tqe_prev
12140                             == NULL) ? "" : " BLOCKED",
12141                             (xio->io_hdr.flags &
12142                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12143                             (xio->io_hdr.flags &
12144                             CTL_FLAG_ABORT) ? " ABORT" : "",
12145                             (xio->io_hdr.flags &
12146                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12147                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12148                 sbuf_finish(&sb);
12149                 printf("%s\n", sbuf_data(&sb));
12150 #endif
12151
12152                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12153                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12154                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12155                         continue;
12156
12157                 /*
12158                  * If the abort says that the task is untagged, the
12159                  * task in the queue must be untagged.  Otherwise,
12160                  * we just check to see whether the tag numbers
12161                  * match.  This is because the QLogic firmware
12162                  * doesn't pass back the tag type in an abort
12163                  * request.
12164                  */
12165 #if 0
12166                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12167                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12168                  || (xio->scsiio.tag_num == io->taskio.tag_num))
12169 #endif
12170                 /*
12171                  * XXX KDM we've got problems with FC, because it
12172                  * doesn't send down a tag type with aborts.  So we
12173                  * can only really go by the tag number...
12174                  * This may cause problems with parallel SCSI.
12175                  * Need to figure that out!!
12176                  */
12177                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
12178                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12179                         found = 1;
12180                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12181                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12182                                 union ctl_ha_msg msg_info;
12183
12184                                 msg_info.hdr.nexus = io->io_hdr.nexus;
12185                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12186                                 msg_info.task.tag_num = io->taskio.tag_num;
12187                                 msg_info.task.tag_type = io->taskio.tag_type;
12188                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12189                                 msg_info.hdr.original_sc = NULL;
12190                                 msg_info.hdr.serializing_sc = NULL;
12191 #if 0
12192                                 printf("Sent Abort to other side\n");
12193 #endif
12194                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12195                                     sizeof(msg_info.task), M_NOWAIT);
12196                         }
12197 #if 0
12198                         printf("ctl_abort_task: found I/O to abort\n");
12199 #endif
12200                 }
12201         }
12202         mtx_unlock(&lun->lun_lock);
12203
12204         if (found == 0) {
12205                 /*
12206                  * This isn't really an error.  It's entirely possible for
12207                  * the abort and command completion to cross on the wire.
12208                  * This is more of an informative/diagnostic error.
12209                  */
12210 #if 0
12211                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12212                        "%u:%u:%u tag %d type %d\n",
12213                        io->io_hdr.nexus.initid,
12214                        io->io_hdr.nexus.targ_port,
12215                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12216                        io->taskio.tag_type);
12217 #endif
12218         }
12219         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12220         return (0);
12221 }
12222
12223 static int
12224 ctl_query_task(union ctl_io *io, int task_set)
12225 {
12226         struct ctl_softc *softc = CTL_SOFTC(io);
12227         union ctl_io *xio;
12228         struct ctl_lun *lun;
12229         int found = 0;
12230         uint32_t targ_lun;
12231
12232         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12233         mtx_lock(&softc->ctl_lock);
12234         if (targ_lun >= CTL_MAX_LUNS ||
12235             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12236                 mtx_unlock(&softc->ctl_lock);
12237                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12238                 return (1);
12239         }
12240         mtx_lock(&lun->lun_lock);
12241         mtx_unlock(&softc->ctl_lock);
12242         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12243              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12244
12245                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12246                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12247                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12248                         continue;
12249
12250                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12251                         found = 1;
12252                         break;
12253                 }
12254         }
12255         mtx_unlock(&lun->lun_lock);
12256         if (found)
12257                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12258         else
12259                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12260         return (0);
12261 }
12262
12263 static int
12264 ctl_query_async_event(union ctl_io *io)
12265 {
12266         struct ctl_softc *softc = CTL_SOFTC(io);
12267         struct ctl_lun *lun;
12268         ctl_ua_type ua;
12269         uint32_t targ_lun, initidx;
12270
12271         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12272         mtx_lock(&softc->ctl_lock);
12273         if (targ_lun >= CTL_MAX_LUNS ||
12274             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12275                 mtx_unlock(&softc->ctl_lock);
12276                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12277                 return (1);
12278         }
12279         mtx_lock(&lun->lun_lock);
12280         mtx_unlock(&softc->ctl_lock);
12281         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12282         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12283         mtx_unlock(&lun->lun_lock);
12284         if (ua != CTL_UA_NONE)
12285                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12286         else
12287                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12288         return (0);
12289 }
12290
12291 static void
12292 ctl_run_task(union ctl_io *io)
12293 {
12294         struct ctl_softc *softc = CTL_SOFTC(io);
12295         int retval = 1;
12296
12297         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12298         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12299             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12300         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12301         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12302         switch (io->taskio.task_action) {
12303         case CTL_TASK_ABORT_TASK:
12304                 retval = ctl_abort_task(io);
12305                 break;
12306         case CTL_TASK_ABORT_TASK_SET:
12307         case CTL_TASK_CLEAR_TASK_SET:
12308                 retval = ctl_abort_task_set(io);
12309                 break;
12310         case CTL_TASK_CLEAR_ACA:
12311                 break;
12312         case CTL_TASK_I_T_NEXUS_RESET:
12313                 retval = ctl_i_t_nexus_reset(io);
12314                 break;
12315         case CTL_TASK_LUN_RESET:
12316                 retval = ctl_lun_reset(softc, io);
12317                 break;
12318         case CTL_TASK_TARGET_RESET:
12319                 retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
12320                 break;
12321         case CTL_TASK_BUS_RESET:
12322                 retval = ctl_bus_reset(softc, io);
12323                 break;
12324         case CTL_TASK_PORT_LOGIN:
12325                 break;
12326         case CTL_TASK_PORT_LOGOUT:
12327                 break;
12328         case CTL_TASK_QUERY_TASK:
12329                 retval = ctl_query_task(io, 0);
12330                 break;
12331         case CTL_TASK_QUERY_TASK_SET:
12332                 retval = ctl_query_task(io, 1);
12333                 break;
12334         case CTL_TASK_QUERY_ASYNC_EVENT:
12335                 retval = ctl_query_async_event(io);
12336                 break;
12337         default:
12338                 printf("%s: got unknown task management event %d\n",
12339                        __func__, io->taskio.task_action);
12340                 break;
12341         }
12342         if (retval == 0)
12343                 io->io_hdr.status = CTL_SUCCESS;
12344         else
12345                 io->io_hdr.status = CTL_ERROR;
12346         ctl_done(io);
12347 }
12348
12349 /*
12350  * For HA operation.  Handle commands that come in from the other
12351  * controller.
12352  */
12353 static void
12354 ctl_handle_isc(union ctl_io *io)
12355 {
12356         struct ctl_softc *softc = CTL_SOFTC(io);
12357         struct ctl_lun *lun;
12358         const struct ctl_cmd_entry *entry;
12359         uint32_t targ_lun;
12360
12361         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12362         switch (io->io_hdr.msg_type) {
12363         case CTL_MSG_SERIALIZE:
12364                 ctl_serialize_other_sc_cmd(&io->scsiio);
12365                 break;
12366         case CTL_MSG_R2R:               /* Only used in SER_ONLY mode. */
12367                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12368                 if (targ_lun >= CTL_MAX_LUNS ||
12369                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12370                         ctl_done(io);
12371                         break;
12372                 }
12373                 mtx_lock(&lun->lun_lock);
12374                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
12375                         mtx_unlock(&lun->lun_lock);
12376                         ctl_done(io);
12377                         break;
12378                 }
12379                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12380                 mtx_unlock(&lun->lun_lock);
12381                 ctl_enqueue_rtr(io);
12382                 break;
12383         case CTL_MSG_FINISH_IO:
12384                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12385                         ctl_done(io);
12386                         break;
12387                 }
12388                 if (targ_lun >= CTL_MAX_LUNS ||
12389                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12390                         ctl_free_io(io);
12391                         break;
12392                 }
12393                 mtx_lock(&lun->lun_lock);
12394                 TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12395                 ctl_check_blocked(lun);
12396                 mtx_unlock(&lun->lun_lock);
12397                 ctl_free_io(io);
12398                 break;
12399         case CTL_MSG_PERS_ACTION:
12400                 ctl_hndl_per_res_out_on_other_sc(io);
12401                 ctl_free_io(io);
12402                 break;
12403         case CTL_MSG_BAD_JUJU:
12404                 ctl_done(io);
12405                 break;
12406         case CTL_MSG_DATAMOVE:          /* Only used in XFER mode */
12407                 ctl_datamove_remote(io);
12408                 break;
12409         case CTL_MSG_DATAMOVE_DONE:     /* Only used in XFER mode */
12410                 io->scsiio.be_move_done(io);
12411                 break;
12412         case CTL_MSG_FAILOVER:
12413                 ctl_failover_lun(io);
12414                 ctl_free_io(io);
12415                 break;
12416         default:
12417                 printf("%s: Invalid message type %d\n",
12418                        __func__, io->io_hdr.msg_type);
12419                 ctl_free_io(io);
12420                 break;
12421         }
12422
12423 }
12424
12425
12426 /*
12427  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12428  * there is no match.
12429  */
12430 static ctl_lun_error_pattern
12431 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12432 {
12433         const struct ctl_cmd_entry *entry;
12434         ctl_lun_error_pattern filtered_pattern, pattern;
12435
12436         pattern = desc->error_pattern;
12437
12438         /*
12439          * XXX KDM we need more data passed into this function to match a
12440          * custom pattern, and we actually need to implement custom pattern
12441          * matching.
12442          */
12443         if (pattern & CTL_LUN_PAT_CMD)
12444                 return (CTL_LUN_PAT_CMD);
12445
12446         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12447                 return (CTL_LUN_PAT_ANY);
12448
12449         entry = ctl_get_cmd_entry(ctsio, NULL);
12450
12451         filtered_pattern = entry->pattern & pattern;
12452
12453         /*
12454          * If the user requested specific flags in the pattern (e.g.
12455          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12456          * flags.
12457          *
12458          * If the user did not specify any flags, it doesn't matter whether
12459          * or not the command supports the flags.
12460          */
12461         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12462              (pattern & ~CTL_LUN_PAT_MASK))
12463                 return (CTL_LUN_PAT_NONE);
12464
12465         /*
12466          * If the user asked for a range check, see if the requested LBA
12467          * range overlaps with this command's LBA range.
12468          */
12469         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12470                 uint64_t lba1;
12471                 uint64_t len1;
12472                 ctl_action action;
12473                 int retval;
12474
12475                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12476                 if (retval != 0)
12477                         return (CTL_LUN_PAT_NONE);
12478
12479                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12480                                               desc->lba_range.len, FALSE);
12481                 /*
12482                  * A "pass" means that the LBA ranges don't overlap, so
12483                  * this doesn't match the user's range criteria.
12484                  */
12485                 if (action == CTL_ACTION_PASS)
12486                         return (CTL_LUN_PAT_NONE);
12487         }
12488
12489         return (filtered_pattern);
12490 }
12491
12492 static void
12493 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12494 {
12495         struct ctl_error_desc *desc, *desc2;
12496
12497         mtx_assert(&lun->lun_lock, MA_OWNED);
12498
12499         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12500                 ctl_lun_error_pattern pattern;
12501                 /*
12502                  * Check to see whether this particular command matches
12503                  * the pattern in the descriptor.
12504                  */
12505                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12506                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12507                         continue;
12508
12509                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12510                 case CTL_LUN_INJ_ABORTED:
12511                         ctl_set_aborted(&io->scsiio);
12512                         break;
12513                 case CTL_LUN_INJ_MEDIUM_ERR:
12514                         ctl_set_medium_error(&io->scsiio,
12515                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12516                              CTL_FLAG_DATA_OUT);
12517                         break;
12518                 case CTL_LUN_INJ_UA:
12519                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12520                          * OCCURRED */
12521                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12522                         break;
12523                 case CTL_LUN_INJ_CUSTOM:
12524                         /*
12525                          * We're assuming the user knows what he is doing.
12526                          * Just copy the sense information without doing
12527                          * checks.
12528                          */
12529                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12530                               MIN(sizeof(desc->custom_sense),
12531                                   sizeof(io->scsiio.sense_data)));
12532                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12533                         io->scsiio.sense_len = SSD_FULL_SIZE;
12534                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12535                         break;
12536                 case CTL_LUN_INJ_NONE:
12537                 default:
12538                         /*
12539                          * If this is an error injection type we don't know
12540                          * about, clear the continuous flag (if it is set)
12541                          * so it will get deleted below.
12542                          */
12543                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12544                         break;
12545                 }
12546                 /*
12547                  * By default, each error injection action is a one-shot
12548                  */
12549                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12550                         continue;
12551
12552                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12553
12554                 free(desc, M_CTL);
12555         }
12556 }
12557
12558 #ifdef CTL_IO_DELAY
12559 static void
12560 ctl_datamove_timer_wakeup(void *arg)
12561 {
12562         union ctl_io *io;
12563
12564         io = (union ctl_io *)arg;
12565
12566         ctl_datamove(io);
12567 }
12568 #endif /* CTL_IO_DELAY */
12569
12570 void
12571 ctl_datamove(union ctl_io *io)
12572 {
12573         void (*fe_datamove)(union ctl_io *io);
12574
12575         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12576
12577         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12578
12579 #ifdef CTL_TIME_IO
12580         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12581                 char str[256];
12582                 char path_str[64];
12583                 struct sbuf sb;
12584
12585                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12586                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12587
12588                 sbuf_cat(&sb, path_str);
12589                 switch (io->io_hdr.io_type) {
12590                 case CTL_IO_SCSI:
12591                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12592                         sbuf_printf(&sb, "\n");
12593                         sbuf_cat(&sb, path_str);
12594                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12595                                     io->scsiio.tag_num, io->scsiio.tag_type);
12596                         break;
12597                 case CTL_IO_TASK:
12598                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12599                                     "Tag Type: %d\n", io->taskio.task_action,
12600                                     io->taskio.tag_num, io->taskio.tag_type);
12601                         break;
12602                 default:
12603                         panic("%s: Invalid CTL I/O type %d\n",
12604                             __func__, io->io_hdr.io_type);
12605                 }
12606                 sbuf_cat(&sb, path_str);
12607                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12608                             (intmax_t)time_uptime - io->io_hdr.start_time);
12609                 sbuf_finish(&sb);
12610                 printf("%s", sbuf_data(&sb));
12611         }
12612 #endif /* CTL_TIME_IO */
12613
12614 #ifdef CTL_IO_DELAY
12615         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12616                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12617         } else {
12618                 if ((lun != NULL)
12619                  && (lun->delay_info.datamove_delay > 0)) {
12620
12621                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12622                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12623                         callout_reset(&io->io_hdr.delay_callout,
12624                                       lun->delay_info.datamove_delay * hz,
12625                                       ctl_datamove_timer_wakeup, io);
12626                         if (lun->delay_info.datamove_type ==
12627                             CTL_DELAY_TYPE_ONESHOT)
12628                                 lun->delay_info.datamove_delay = 0;
12629                         return;
12630                 }
12631         }
12632 #endif
12633
12634         /*
12635          * This command has been aborted.  Set the port status, so we fail
12636          * the data move.
12637          */
12638         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12639                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12640                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12641                        io->io_hdr.nexus.targ_port,
12642                        io->io_hdr.nexus.targ_lun);
12643                 io->io_hdr.port_status = 31337;
12644                 /*
12645                  * Note that the backend, in this case, will get the
12646                  * callback in its context.  In other cases it may get
12647                  * called in the frontend's interrupt thread context.
12648                  */
12649                 io->scsiio.be_move_done(io);
12650                 return;
12651         }
12652
12653         /* Don't confuse frontend with zero length data move. */
12654         if (io->scsiio.kern_data_len == 0) {
12655                 io->scsiio.be_move_done(io);
12656                 return;
12657         }
12658
12659         fe_datamove = CTL_PORT(io)->fe_datamove;
12660         fe_datamove(io);
12661 }
12662
12663 static void
12664 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12665 {
12666         union ctl_ha_msg msg;
12667 #ifdef CTL_TIME_IO
12668         struct bintime cur_bt;
12669 #endif
12670
12671         memset(&msg, 0, sizeof(msg));
12672         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12673         msg.hdr.original_sc = io;
12674         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12675         msg.hdr.nexus = io->io_hdr.nexus;
12676         msg.hdr.status = io->io_hdr.status;
12677         msg.scsi.tag_num = io->scsiio.tag_num;
12678         msg.scsi.tag_type = io->scsiio.tag_type;
12679         msg.scsi.scsi_status = io->scsiio.scsi_status;
12680         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12681                io->scsiio.sense_len);
12682         msg.scsi.sense_len = io->scsiio.sense_len;
12683         msg.scsi.sense_residual = io->scsiio.sense_residual;
12684         msg.scsi.fetd_status = io->io_hdr.port_status;
12685         msg.scsi.residual = io->scsiio.residual;
12686         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12687         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12688                 ctl_failover_io(io, /*have_lock*/ have_lock);
12689                 return;
12690         }
12691         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12692             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12693             msg.scsi.sense_len, M_WAITOK);
12694
12695 #ifdef CTL_TIME_IO
12696         getbinuptime(&cur_bt);
12697         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12698         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12699 #endif
12700         io->io_hdr.num_dmas++;
12701 }
12702
12703 /*
12704  * The DMA to the remote side is done, now we need to tell the other side
12705  * we're done so it can continue with its data movement.
12706  */
12707 static void
12708 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12709 {
12710         union ctl_io *io;
12711         uint32_t i;
12712
12713         io = rq->context;
12714
12715         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12716                 printf("%s: ISC DMA write failed with error %d", __func__,
12717                        rq->ret);
12718                 ctl_set_internal_failure(&io->scsiio,
12719                                          /*sks_valid*/ 1,
12720                                          /*retry_count*/ rq->ret);
12721         }
12722
12723         ctl_dt_req_free(rq);
12724
12725         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12726                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12727         free(io->io_hdr.remote_sglist, M_CTL);
12728         io->io_hdr.remote_sglist = NULL;
12729         io->io_hdr.local_sglist = NULL;
12730
12731         /*
12732          * The data is in local and remote memory, so now we need to send
12733          * status (good or back) back to the other side.
12734          */
12735         ctl_send_datamove_done(io, /*have_lock*/ 0);
12736 }
12737
12738 /*
12739  * We've moved the data from the host/controller into local memory.  Now we
12740  * need to push it over to the remote controller's memory.
12741  */
12742 static int
12743 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12744 {
12745         int retval;
12746
12747         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12748                                           ctl_datamove_remote_write_cb);
12749         return (retval);
12750 }
12751
12752 static void
12753 ctl_datamove_remote_write(union ctl_io *io)
12754 {
12755         int retval;
12756         void (*fe_datamove)(union ctl_io *io);
12757
12758         /*
12759          * - Get the data from the host/HBA into local memory.
12760          * - DMA memory from the local controller to the remote controller.
12761          * - Send status back to the remote controller.
12762          */
12763
12764         retval = ctl_datamove_remote_sgl_setup(io);
12765         if (retval != 0)
12766                 return;
12767
12768         /* Switch the pointer over so the FETD knows what to do */
12769         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12770
12771         /*
12772          * Use a custom move done callback, since we need to send completion
12773          * back to the other controller, not to the backend on this side.
12774          */
12775         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12776
12777         fe_datamove = CTL_PORT(io)->fe_datamove;
12778         fe_datamove(io);
12779 }
12780
12781 static int
12782 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12783 {
12784 #if 0
12785         char str[256];
12786         char path_str[64];
12787         struct sbuf sb;
12788 #endif
12789         uint32_t i;
12790
12791         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12792                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12793         free(io->io_hdr.remote_sglist, M_CTL);
12794         io->io_hdr.remote_sglist = NULL;
12795         io->io_hdr.local_sglist = NULL;
12796
12797 #if 0
12798         scsi_path_string(io, path_str, sizeof(path_str));
12799         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12800         sbuf_cat(&sb, path_str);
12801         scsi_command_string(&io->scsiio, NULL, &sb);
12802         sbuf_printf(&sb, "\n");
12803         sbuf_cat(&sb, path_str);
12804         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12805                     io->scsiio.tag_num, io->scsiio.tag_type);
12806         sbuf_cat(&sb, path_str);
12807         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12808                     io->io_hdr.flags, io->io_hdr.status);
12809         sbuf_finish(&sb);
12810         printk("%s", sbuf_data(&sb));
12811 #endif
12812
12813
12814         /*
12815          * The read is done, now we need to send status (good or bad) back
12816          * to the other side.
12817          */
12818         ctl_send_datamove_done(io, /*have_lock*/ 0);
12819
12820         return (0);
12821 }
12822
12823 static void
12824 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12825 {
12826         union ctl_io *io;
12827         void (*fe_datamove)(union ctl_io *io);
12828
12829         io = rq->context;
12830
12831         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12832                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12833                        rq->ret);
12834                 ctl_set_internal_failure(&io->scsiio,
12835                                          /*sks_valid*/ 1,
12836                                          /*retry_count*/ rq->ret);
12837         }
12838
12839         ctl_dt_req_free(rq);
12840
12841         /* Switch the pointer over so the FETD knows what to do */
12842         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12843
12844         /*
12845          * Use a custom move done callback, since we need to send completion
12846          * back to the other controller, not to the backend on this side.
12847          */
12848         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12849
12850         /* XXX KDM add checks like the ones in ctl_datamove? */
12851
12852         fe_datamove = CTL_PORT(io)->fe_datamove;
12853         fe_datamove(io);
12854 }
12855
12856 static int
12857 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12858 {
12859         struct ctl_sg_entry *local_sglist;
12860         uint32_t len_to_go;
12861         int retval;
12862         int i;
12863
12864         retval = 0;
12865         local_sglist = io->io_hdr.local_sglist;
12866         len_to_go = io->scsiio.kern_data_len;
12867
12868         /*
12869          * The difficult thing here is that the size of the various
12870          * S/G segments may be different than the size from the
12871          * remote controller.  That'll make it harder when DMAing
12872          * the data back to the other side.
12873          */
12874         for (i = 0; len_to_go > 0; i++) {
12875                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12876                 local_sglist[i].addr =
12877                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12878
12879                 len_to_go -= local_sglist[i].len;
12880         }
12881         /*
12882          * Reset the number of S/G entries accordingly.  The original
12883          * number of S/G entries is available in rem_sg_entries.
12884          */
12885         io->scsiio.kern_sg_entries = i;
12886
12887 #if 0
12888         printf("%s: kern_sg_entries = %d\n", __func__,
12889                io->scsiio.kern_sg_entries);
12890         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12891                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12892                        local_sglist[i].addr, local_sglist[i].len);
12893 #endif
12894
12895         return (retval);
12896 }
12897
12898 static int
12899 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12900                          ctl_ha_dt_cb callback)
12901 {
12902         struct ctl_ha_dt_req *rq;
12903         struct ctl_sg_entry *remote_sglist, *local_sglist;
12904         uint32_t local_used, remote_used, total_used;
12905         int i, j, isc_ret;
12906
12907         rq = ctl_dt_req_alloc();
12908
12909         /*
12910          * If we failed to allocate the request, and if the DMA didn't fail
12911          * anyway, set busy status.  This is just a resource allocation
12912          * failure.
12913          */
12914         if ((rq == NULL)
12915          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12916              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12917                 ctl_set_busy(&io->scsiio);
12918
12919         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12920             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12921
12922                 if (rq != NULL)
12923                         ctl_dt_req_free(rq);
12924
12925                 /*
12926                  * The data move failed.  We need to return status back
12927                  * to the other controller.  No point in trying to DMA
12928                  * data to the remote controller.
12929                  */
12930
12931                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12932
12933                 return (1);
12934         }
12935
12936         local_sglist = io->io_hdr.local_sglist;
12937         remote_sglist = io->io_hdr.remote_sglist;
12938         local_used = 0;
12939         remote_used = 0;
12940         total_used = 0;
12941
12942         /*
12943          * Pull/push the data over the wire from/to the other controller.
12944          * This takes into account the possibility that the local and
12945          * remote sglists may not be identical in terms of the size of
12946          * the elements and the number of elements.
12947          *
12948          * One fundamental assumption here is that the length allocated for
12949          * both the local and remote sglists is identical.  Otherwise, we've
12950          * essentially got a coding error of some sort.
12951          */
12952         isc_ret = CTL_HA_STATUS_SUCCESS;
12953         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12954                 uint32_t cur_len;
12955                 uint8_t *tmp_ptr;
12956
12957                 rq->command = command;
12958                 rq->context = io;
12959
12960                 /*
12961                  * Both pointers should be aligned.  But it is possible
12962                  * that the allocation length is not.  They should both
12963                  * also have enough slack left over at the end, though,
12964                  * to round up to the next 8 byte boundary.
12965                  */
12966                 cur_len = MIN(local_sglist[i].len - local_used,
12967                               remote_sglist[j].len - remote_used);
12968                 rq->size = cur_len;
12969
12970                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12971                 tmp_ptr += local_used;
12972
12973 #if 0
12974                 /* Use physical addresses when talking to ISC hardware */
12975                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12976                         /* XXX KDM use busdma */
12977                         rq->local = vtophys(tmp_ptr);
12978                 } else
12979                         rq->local = tmp_ptr;
12980 #else
12981                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12982                     ("HA does not support BUS_ADDR"));
12983                 rq->local = tmp_ptr;
12984 #endif
12985
12986                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12987                 tmp_ptr += remote_used;
12988                 rq->remote = tmp_ptr;
12989
12990                 rq->callback = NULL;
12991
12992                 local_used += cur_len;
12993                 if (local_used >= local_sglist[i].len) {
12994                         i++;
12995                         local_used = 0;
12996                 }
12997
12998                 remote_used += cur_len;
12999                 if (remote_used >= remote_sglist[j].len) {
13000                         j++;
13001                         remote_used = 0;
13002                 }
13003                 total_used += cur_len;
13004
13005                 if (total_used >= io->scsiio.kern_data_len)
13006                         rq->callback = callback;
13007
13008 #if 0
13009                 printf("%s: %s: local %p remote %p size %d\n", __func__,
13010                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13011                        rq->local, rq->remote, rq->size);
13012 #endif
13013
13014                 isc_ret = ctl_dt_single(rq);
13015                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
13016                         break;
13017         }
13018         if (isc_ret != CTL_HA_STATUS_WAIT) {
13019                 rq->ret = isc_ret;
13020                 callback(rq);
13021         }
13022
13023         return (0);
13024 }
13025
13026 static void
13027 ctl_datamove_remote_read(union ctl_io *io)
13028 {
13029         int retval;
13030         uint32_t i;
13031
13032         /*
13033          * This will send an error to the other controller in the case of a
13034          * failure.
13035          */
13036         retval = ctl_datamove_remote_sgl_setup(io);
13037         if (retval != 0)
13038                 return;
13039
13040         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13041                                           ctl_datamove_remote_read_cb);
13042         if (retval != 0) {
13043                 /*
13044                  * Make sure we free memory if there was an error..  The
13045                  * ctl_datamove_remote_xfer() function will send the
13046                  * datamove done message, or call the callback with an
13047                  * error if there is a problem.
13048                  */
13049                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13050                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13051                 free(io->io_hdr.remote_sglist, M_CTL);
13052                 io->io_hdr.remote_sglist = NULL;
13053                 io->io_hdr.local_sglist = NULL;
13054         }
13055 }
13056
13057 /*
13058  * Process a datamove request from the other controller.  This is used for
13059  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13060  * first.  Once that is complete, the data gets DMAed into the remote
13061  * controller's memory.  For reads, we DMA from the remote controller's
13062  * memory into our memory first, and then move it out to the FETD.
13063  */
13064 static void
13065 ctl_datamove_remote(union ctl_io *io)
13066 {
13067
13068         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
13069
13070         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13071                 ctl_failover_io(io, /*have_lock*/ 0);
13072                 return;
13073         }
13074
13075         /*
13076          * Note that we look for an aborted I/O here, but don't do some of
13077          * the other checks that ctl_datamove() normally does.
13078          * We don't need to run the datamove delay code, since that should
13079          * have been done if need be on the other controller.
13080          */
13081         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13082                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
13083                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
13084                        io->io_hdr.nexus.targ_port,
13085                        io->io_hdr.nexus.targ_lun);
13086                 io->io_hdr.port_status = 31338;
13087                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13088                 return;
13089         }
13090
13091         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
13092                 ctl_datamove_remote_write(io);
13093         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
13094                 ctl_datamove_remote_read(io);
13095         else {
13096                 io->io_hdr.port_status = 31339;
13097                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13098         }
13099 }
13100
13101 static void
13102 ctl_process_done(union ctl_io *io)
13103 {
13104         struct ctl_softc *softc = CTL_SOFTC(io);
13105         struct ctl_port *port = CTL_PORT(io);
13106         struct ctl_lun *lun = CTL_LUN(io);
13107         void (*fe_done)(union ctl_io *io);
13108         union ctl_ha_msg msg;
13109
13110         CTL_DEBUG_PRINT(("ctl_process_done\n"));
13111         fe_done = port->fe_done;
13112
13113 #ifdef CTL_TIME_IO
13114         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13115                 char str[256];
13116                 char path_str[64];
13117                 struct sbuf sb;
13118
13119                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
13120                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13121
13122                 sbuf_cat(&sb, path_str);
13123                 switch (io->io_hdr.io_type) {
13124                 case CTL_IO_SCSI:
13125                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13126                         sbuf_printf(&sb, "\n");
13127                         sbuf_cat(&sb, path_str);
13128                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13129                                     io->scsiio.tag_num, io->scsiio.tag_type);
13130                         break;
13131                 case CTL_IO_TASK:
13132                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13133                                     "Tag Type: %d\n", io->taskio.task_action,
13134                                     io->taskio.tag_num, io->taskio.tag_type);
13135                         break;
13136                 default:
13137                         panic("%s: Invalid CTL I/O type %d\n",
13138                             __func__, io->io_hdr.io_type);
13139                 }
13140                 sbuf_cat(&sb, path_str);
13141                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13142                             (intmax_t)time_uptime - io->io_hdr.start_time);
13143                 sbuf_finish(&sb);
13144                 printf("%s", sbuf_data(&sb));
13145         }
13146 #endif /* CTL_TIME_IO */
13147
13148         switch (io->io_hdr.io_type) {
13149         case CTL_IO_SCSI:
13150                 break;
13151         case CTL_IO_TASK:
13152                 if (ctl_debug & CTL_DEBUG_INFO)
13153                         ctl_io_error_print(io, NULL);
13154                 fe_done(io);
13155                 return;
13156         default:
13157                 panic("%s: Invalid CTL I/O type %d\n",
13158                     __func__, io->io_hdr.io_type);
13159         }
13160
13161         if (lun == NULL) {
13162                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13163                                  io->io_hdr.nexus.targ_mapped_lun));
13164                 goto bailout;
13165         }
13166
13167         mtx_lock(&lun->lun_lock);
13168
13169         /*
13170          * Check to see if we have any informational exception and status
13171          * of this command can be modified to report it in form of either
13172          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
13173          */
13174         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
13175             io->io_hdr.status == CTL_SUCCESS &&
13176             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
13177                 uint8_t mrie = lun->MODE_IE.mrie;
13178                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
13179                     (lun->MODE_VER.byte3 & SMS_VER_PER));
13180                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
13181                      mrie == SIEP_MRIE_REC_UNCOND ||
13182                      mrie == SIEP_MRIE_NO_SENSE) &&
13183                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
13184                      CTL_CMD_FLAG_NO_SENSE) == 0) {
13185                         ctl_set_sense(&io->scsiio,
13186                               /*current_error*/ 1,
13187                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
13188                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
13189                               /*asc*/ lun->ie_asc,
13190                               /*ascq*/ lun->ie_ascq,
13191                               SSD_ELEM_NONE);
13192                         lun->ie_reported = 1;
13193                 }
13194         } else if (lun->ie_reported < 0)
13195                 lun->ie_reported = 0;
13196
13197         /*
13198          * Check to see if we have any errors to inject here.  We only
13199          * inject errors for commands that don't already have errors set.
13200          */
13201         if (!STAILQ_EMPTY(&lun->error_list) &&
13202             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13203             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13204                 ctl_inject_error(lun, io);
13205
13206         /*
13207          * XXX KDM how do we treat commands that aren't completed
13208          * successfully?
13209          *
13210          * XXX KDM should we also track I/O latency?
13211          */
13212         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13213             io->io_hdr.io_type == CTL_IO_SCSI) {
13214                 int type;
13215 #ifdef CTL_TIME_IO
13216                 struct bintime bt;
13217
13218                 getbinuptime(&bt);
13219                 bintime_sub(&bt, &io->io_hdr.start_bt);
13220 #endif
13221                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13222                     CTL_FLAG_DATA_IN)
13223                         type = CTL_STATS_READ;
13224                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13225                     CTL_FLAG_DATA_OUT)
13226                         type = CTL_STATS_WRITE;
13227                 else
13228                         type = CTL_STATS_NO_IO;
13229
13230 #ifdef CTL_LEGACY_STATS
13231                 uint32_t targ_port = port->targ_port;
13232                 lun->legacy_stats.ports[targ_port].bytes[type] +=
13233                     io->scsiio.kern_total_len;
13234                 lun->legacy_stats.ports[targ_port].operations[type] ++;
13235                 lun->legacy_stats.ports[targ_port].num_dmas[type] +=
13236                     io->io_hdr.num_dmas;
13237 #ifdef CTL_TIME_IO
13238                 bintime_add(&lun->legacy_stats.ports[targ_port].dma_time[type],
13239                    &io->io_hdr.dma_bt);
13240                 bintime_add(&lun->legacy_stats.ports[targ_port].time[type],
13241                     &bt);
13242 #endif
13243 #endif /* CTL_LEGACY_STATS */
13244
13245                 lun->stats.bytes[type] += io->scsiio.kern_total_len;
13246                 lun->stats.operations[type] ++;
13247                 lun->stats.dmas[type] += io->io_hdr.num_dmas;
13248 #ifdef CTL_TIME_IO
13249                 bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt);
13250                 bintime_add(&lun->stats.time[type], &bt);
13251 #endif
13252
13253                 mtx_lock(&port->port_lock);
13254                 port->stats.bytes[type] += io->scsiio.kern_total_len;
13255                 port->stats.operations[type] ++;
13256                 port->stats.dmas[type] += io->io_hdr.num_dmas;
13257 #ifdef CTL_TIME_IO
13258                 bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt);
13259                 bintime_add(&port->stats.time[type], &bt);
13260 #endif
13261                 mtx_unlock(&port->port_lock);
13262         }
13263
13264         /*
13265          * Remove this from the OOA queue.
13266          */
13267         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13268 #ifdef CTL_TIME_IO
13269         if (TAILQ_EMPTY(&lun->ooa_queue))
13270                 lun->last_busy = getsbinuptime();
13271 #endif
13272
13273         /*
13274          * Run through the blocked queue on this LUN and see if anything
13275          * has become unblocked, now that this transaction is done.
13276          */
13277         ctl_check_blocked(lun);
13278
13279         /*
13280          * If the LUN has been invalidated, free it if there is nothing
13281          * left on its OOA queue.
13282          */
13283         if ((lun->flags & CTL_LUN_INVALID)
13284          && TAILQ_EMPTY(&lun->ooa_queue)) {
13285                 mtx_unlock(&lun->lun_lock);
13286                 mtx_lock(&softc->ctl_lock);
13287                 ctl_free_lun(lun);
13288                 mtx_unlock(&softc->ctl_lock);
13289         } else
13290                 mtx_unlock(&lun->lun_lock);
13291
13292 bailout:
13293
13294         /*
13295          * If this command has been aborted, make sure we set the status
13296          * properly.  The FETD is responsible for freeing the I/O and doing
13297          * whatever it needs to do to clean up its state.
13298          */
13299         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13300                 ctl_set_task_aborted(&io->scsiio);
13301
13302         /*
13303          * If enabled, print command error status.
13304          */
13305         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13306             (ctl_debug & CTL_DEBUG_INFO) != 0)
13307                 ctl_io_error_print(io, NULL);
13308
13309         /*
13310          * Tell the FETD or the other shelf controller we're done with this
13311          * command.  Note that only SCSI commands get to this point.  Task
13312          * management commands are completed above.
13313          */
13314         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13315             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13316                 memset(&msg, 0, sizeof(msg));
13317                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13318                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13319                 msg.hdr.nexus = io->io_hdr.nexus;
13320                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13321                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13322                     M_WAITOK);
13323         }
13324
13325         fe_done(io);
13326 }
13327
13328 #ifdef CTL_WITH_CA
13329 /*
13330  * Front end should call this if it doesn't do autosense.  When the request
13331  * sense comes back in from the initiator, we'll dequeue this and send it.
13332  */
13333 int
13334 ctl_queue_sense(union ctl_io *io)
13335 {
13336         struct ctl_softc *softc = CTL_SOFTC(io);
13337         struct ctl_port *port = CTL_PORT(io);
13338         struct ctl_lun *lun;
13339         uint32_t initidx, targ_lun;
13340
13341         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13342
13343         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13344
13345         /*
13346          * LUN lookup will likely move to the ctl_work_thread() once we
13347          * have our new queueing infrastructure (that doesn't put things on
13348          * a per-LUN queue initially).  That is so that we can handle
13349          * things like an INQUIRY to a LUN that we don't have enabled.  We
13350          * can't deal with that right now.
13351          * If we don't have a LUN for this, just toss the sense information.
13352          */
13353         mtx_lock(&softc->ctl_lock);
13354         if (targ_lun >= CTL_MAX_LUNS ||
13355             (lun = softc->ctl_luns[targ_lun]) == NULL) {
13356                 mtx_unlock(&softc->ctl_lock);
13357                 goto bailout;
13358         }
13359         mtx_lock(&lun->lun_lock);
13360         mtx_unlock(&softc->ctl_lock);
13361
13362         /*
13363          * Already have CA set for this LUN...toss the sense information.
13364          */
13365         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13366         if (ctl_is_set(lun->have_ca, initidx)) {
13367                 mtx_unlock(&lun->lun_lock);
13368                 goto bailout;
13369         }
13370
13371         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13372                MIN(sizeof(lun->pending_sense[initidx]),
13373                sizeof(io->scsiio.sense_data)));
13374         ctl_set_mask(lun->have_ca, initidx);
13375         mtx_unlock(&lun->lun_lock);
13376
13377 bailout:
13378         ctl_free_io(io);
13379         return (CTL_RETVAL_COMPLETE);
13380 }
13381 #endif
13382
13383 /*
13384  * Primary command inlet from frontend ports.  All SCSI and task I/O
13385  * requests must go through this function.
13386  */
13387 int
13388 ctl_queue(union ctl_io *io)
13389 {
13390         struct ctl_port *port = CTL_PORT(io);
13391
13392         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13393
13394 #ifdef CTL_TIME_IO
13395         io->io_hdr.start_time = time_uptime;
13396         getbinuptime(&io->io_hdr.start_bt);
13397 #endif /* CTL_TIME_IO */
13398
13399         /* Map FE-specific LUN ID into global one. */
13400         io->io_hdr.nexus.targ_mapped_lun =
13401             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13402
13403         switch (io->io_hdr.io_type) {
13404         case CTL_IO_SCSI:
13405         case CTL_IO_TASK:
13406                 if (ctl_debug & CTL_DEBUG_CDB)
13407                         ctl_io_print(io);
13408                 ctl_enqueue_incoming(io);
13409                 break;
13410         default:
13411                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13412                 return (EINVAL);
13413         }
13414
13415         return (CTL_RETVAL_COMPLETE);
13416 }
13417
13418 #ifdef CTL_IO_DELAY
13419 static void
13420 ctl_done_timer_wakeup(void *arg)
13421 {
13422         union ctl_io *io;
13423
13424         io = (union ctl_io *)arg;
13425         ctl_done(io);
13426 }
13427 #endif /* CTL_IO_DELAY */
13428
13429 void
13430 ctl_serseq_done(union ctl_io *io)
13431 {
13432         struct ctl_lun *lun = CTL_LUN(io);;
13433
13434         if (lun->be_lun == NULL ||
13435             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13436                 return;
13437         mtx_lock(&lun->lun_lock);
13438         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13439         ctl_check_blocked(lun);
13440         mtx_unlock(&lun->lun_lock);
13441 }
13442
13443 void
13444 ctl_done(union ctl_io *io)
13445 {
13446
13447         /*
13448          * Enable this to catch duplicate completion issues.
13449          */
13450 #if 0
13451         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13452                 printf("%s: type %d msg %d cdb %x iptl: "
13453                        "%u:%u:%u tag 0x%04x "
13454                        "flag %#x status %x\n",
13455                         __func__,
13456                         io->io_hdr.io_type,
13457                         io->io_hdr.msg_type,
13458                         io->scsiio.cdb[0],
13459                         io->io_hdr.nexus.initid,
13460                         io->io_hdr.nexus.targ_port,
13461                         io->io_hdr.nexus.targ_lun,
13462                         (io->io_hdr.io_type ==
13463                         CTL_IO_TASK) ?
13464                         io->taskio.tag_num :
13465                         io->scsiio.tag_num,
13466                         io->io_hdr.flags,
13467                         io->io_hdr.status);
13468         } else
13469                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13470 #endif
13471
13472         /*
13473          * This is an internal copy of an I/O, and should not go through
13474          * the normal done processing logic.
13475          */
13476         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13477                 return;
13478
13479 #ifdef CTL_IO_DELAY
13480         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13481                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13482         } else {
13483                 struct ctl_lun *lun = CTL_LUN(io);
13484
13485                 if ((lun != NULL)
13486                  && (lun->delay_info.done_delay > 0)) {
13487
13488                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13489                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13490                         callout_reset(&io->io_hdr.delay_callout,
13491                                       lun->delay_info.done_delay * hz,
13492                                       ctl_done_timer_wakeup, io);
13493                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13494                                 lun->delay_info.done_delay = 0;
13495                         return;
13496                 }
13497         }
13498 #endif /* CTL_IO_DELAY */
13499
13500         ctl_enqueue_done(io);
13501 }
13502
13503 static void
13504 ctl_work_thread(void *arg)
13505 {
13506         struct ctl_thread *thr = (struct ctl_thread *)arg;
13507         struct ctl_softc *softc = thr->ctl_softc;
13508         union ctl_io *io;
13509         int retval;
13510
13511         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13512
13513         for (;;) {
13514                 /*
13515                  * We handle the queues in this order:
13516                  * - ISC
13517                  * - done queue (to free up resources, unblock other commands)
13518                  * - RtR queue
13519                  * - incoming queue
13520                  *
13521                  * If those queues are empty, we break out of the loop and
13522                  * go to sleep.
13523                  */
13524                 mtx_lock(&thr->queue_lock);
13525                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13526                 if (io != NULL) {
13527                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13528                         mtx_unlock(&thr->queue_lock);
13529                         ctl_handle_isc(io);
13530                         continue;
13531                 }
13532                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13533                 if (io != NULL) {
13534                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13535                         /* clear any blocked commands, call fe_done */
13536                         mtx_unlock(&thr->queue_lock);
13537                         ctl_process_done(io);
13538                         continue;
13539                 }
13540                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13541                 if (io != NULL) {
13542                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13543                         mtx_unlock(&thr->queue_lock);
13544                         if (io->io_hdr.io_type == CTL_IO_TASK)
13545                                 ctl_run_task(io);
13546                         else
13547                                 ctl_scsiio_precheck(softc, &io->scsiio);
13548                         continue;
13549                 }
13550                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13551                 if (io != NULL) {
13552                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13553                         mtx_unlock(&thr->queue_lock);
13554                         retval = ctl_scsiio(&io->scsiio);
13555                         if (retval != CTL_RETVAL_COMPLETE)
13556                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13557                         continue;
13558                 }
13559
13560                 /* Sleep until we have something to do. */
13561                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13562         }
13563 }
13564
13565 static void
13566 ctl_lun_thread(void *arg)
13567 {
13568         struct ctl_softc *softc = (struct ctl_softc *)arg;
13569         struct ctl_be_lun *be_lun;
13570
13571         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13572
13573         for (;;) {
13574                 mtx_lock(&softc->ctl_lock);
13575                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13576                 if (be_lun != NULL) {
13577                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13578                         mtx_unlock(&softc->ctl_lock);
13579                         ctl_create_lun(be_lun);
13580                         continue;
13581                 }
13582
13583                 /* Sleep until we have something to do. */
13584                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13585                     PDROP | PRIBIO, "-", 0);
13586         }
13587 }
13588
13589 static void
13590 ctl_thresh_thread(void *arg)
13591 {
13592         struct ctl_softc *softc = (struct ctl_softc *)arg;
13593         struct ctl_lun *lun;
13594         struct ctl_logical_block_provisioning_page *page;
13595         const char *attr;
13596         union ctl_ha_msg msg;
13597         uint64_t thres, val;
13598         int i, e, set;
13599
13600         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13601
13602         for (;;) {
13603                 mtx_lock(&softc->ctl_lock);
13604                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13605                         if ((lun->flags & CTL_LUN_DISABLED) ||
13606                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13607                             lun->backend->lun_attr == NULL)
13608                                 continue;
13609                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13610                             softc->ha_mode == CTL_HA_MODE_XFER)
13611                                 continue;
13612                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13613                                 continue;
13614                         e = 0;
13615                         page = &lun->MODE_LBP;
13616                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13617                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13618                                         continue;
13619                                 thres = scsi_4btoul(page->descr[i].count);
13620                                 thres <<= CTL_LBP_EXPONENT;
13621                                 switch (page->descr[i].resource) {
13622                                 case 0x01:
13623                                         attr = "blocksavail";
13624                                         break;
13625                                 case 0x02:
13626                                         attr = "blocksused";
13627                                         break;
13628                                 case 0xf1:
13629                                         attr = "poolblocksavail";
13630                                         break;
13631                                 case 0xf2:
13632                                         attr = "poolblocksused";
13633                                         break;
13634                                 default:
13635                                         continue;
13636                                 }
13637                                 mtx_unlock(&softc->ctl_lock); // XXX
13638                                 val = lun->backend->lun_attr(
13639                                     lun->be_lun->be_lun, attr);
13640                                 mtx_lock(&softc->ctl_lock);
13641                                 if (val == UINT64_MAX)
13642                                         continue;
13643                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13644                                     == SLBPPD_ARMING_INC)
13645                                         e = (val >= thres);
13646                                 else
13647                                         e = (val <= thres);
13648                                 if (e)
13649                                         break;
13650                         }
13651                         mtx_lock(&lun->lun_lock);
13652                         if (e) {
13653                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13654                                     (uint8_t *)page, lun->ua_tpt_info);
13655                                 if (lun->lasttpt == 0 ||
13656                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13657                                         lun->lasttpt = time_uptime;
13658                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13659                                         set = 1;
13660                                 } else
13661                                         set = 0;
13662                         } else {
13663                                 lun->lasttpt = 0;
13664                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13665                                 set = -1;
13666                         }
13667                         mtx_unlock(&lun->lun_lock);
13668                         if (set != 0 &&
13669                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13670                                 /* Send msg to other side. */
13671                                 bzero(&msg.ua, sizeof(msg.ua));
13672                                 msg.hdr.msg_type = CTL_MSG_UA;
13673                                 msg.hdr.nexus.initid = -1;
13674                                 msg.hdr.nexus.targ_port = -1;
13675                                 msg.hdr.nexus.targ_lun = lun->lun;
13676                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13677                                 msg.ua.ua_all = 1;
13678                                 msg.ua.ua_set = (set > 0);
13679                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13680                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13681                                 mtx_unlock(&softc->ctl_lock); // XXX
13682                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13683                                     sizeof(msg.ua), M_WAITOK);
13684                                 mtx_lock(&softc->ctl_lock);
13685                         }
13686                 }
13687                 mtx_unlock(&softc->ctl_lock);
13688                 pause("-", CTL_LBP_PERIOD * hz);
13689         }
13690 }
13691
13692 static void
13693 ctl_enqueue_incoming(union ctl_io *io)
13694 {
13695         struct ctl_softc *softc = CTL_SOFTC(io);
13696         struct ctl_thread *thr;
13697         u_int idx;
13698
13699         idx = (io->io_hdr.nexus.targ_port * 127 +
13700                io->io_hdr.nexus.initid) % worker_threads;
13701         thr = &softc->threads[idx];
13702         mtx_lock(&thr->queue_lock);
13703         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13704         mtx_unlock(&thr->queue_lock);
13705         wakeup(thr);
13706 }
13707
13708 static void
13709 ctl_enqueue_rtr(union ctl_io *io)
13710 {
13711         struct ctl_softc *softc = CTL_SOFTC(io);
13712         struct ctl_thread *thr;
13713
13714         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13715         mtx_lock(&thr->queue_lock);
13716         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13717         mtx_unlock(&thr->queue_lock);
13718         wakeup(thr);
13719 }
13720
13721 static void
13722 ctl_enqueue_done(union ctl_io *io)
13723 {
13724         struct ctl_softc *softc = CTL_SOFTC(io);
13725         struct ctl_thread *thr;
13726
13727         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13728         mtx_lock(&thr->queue_lock);
13729         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13730         mtx_unlock(&thr->queue_lock);
13731         wakeup(thr);
13732 }
13733
13734 static void
13735 ctl_enqueue_isc(union ctl_io *io)
13736 {
13737         struct ctl_softc *softc = CTL_SOFTC(io);
13738         struct ctl_thread *thr;
13739
13740         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13741         mtx_lock(&thr->queue_lock);
13742         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13743         mtx_unlock(&thr->queue_lock);
13744         wakeup(thr);
13745 }
13746
13747 /*
13748  *  vim: ts=8
13749  */