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 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 406     &worker_threads, 1, "Number of worker threads");
 407 static int ctl_debug = CTL_DEBUG_NONE;
 408 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
 409     &ctl_debug, 0, "Enabled debug flags");
 410 static int ctl_lun_map_size = 1024;
 411 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, lun_map_size, CTLFLAG_RWTUN,
 412     &ctl_lun_map_size, 0, "Size of per-port LUN map (max LUN + 1)");
 413
 414 /*
 415  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 416  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 417  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
 418  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
 419  */
 420 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
 421
 422 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 423                                   int param);
 424 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 425 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
 426 static int ctl_init(void);
 427 static int ctl_shutdown(void);
 428 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 429 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 430 static void ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 431 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 432                               struct ctl_ooa *ooa_hdr,
 433                               struct ctl_ooa_entry *kern_entries);
 434 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 435                      struct thread *td);
 436 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 437                          struct ctl_be_lun *be_lun);
 438 static int ctl_free_lun(struct ctl_lun *lun);
 439 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 440
 441 static int ctl_do_mode_select(union ctl_io *io);
 442 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 443                            uint64_t res_key, uint64_t sa_res_key,
 444                            uint8_t type, uint32_t residx,
 445                            struct ctl_scsiio *ctsio,
 446                            struct scsi_per_res_out *cdb,
 447                            struct scsi_per_res_out_parms* param);
 448 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 449                                   union ctl_ha_msg *msg);
 450 static void ctl_hndl_per_res_out_on_other_sc(union ctl_io *io);
 451 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 452 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 453 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 454 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 455 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 456 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 457                                          int alloc_len);
 458 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 459                                          int alloc_len);
 460 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 461 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 462 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 463 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 464 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 465 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
 466     bool seq);
 467 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
 468 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 469     union ctl_io *pending_io, union ctl_io *ooa_io);
 470 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 471                                 union ctl_io *starting_io);
 472 static int ctl_check_blocked(struct ctl_lun *lun);
 473 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
 474                                 const struct ctl_cmd_entry *entry,
 475                                 struct ctl_scsiio *ctsio);
 476 static void ctl_failover_lun(union ctl_io *io);
 477 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 478                                struct ctl_scsiio *ctsio);
 479 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 480
 481 static int ctl_target_reset(union ctl_io *io);
 482 static void ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx,
 483                          ctl_ua_type ua_type);
 484 static int ctl_lun_reset(union ctl_io *io);
 485 static int ctl_abort_task(union ctl_io *io);
 486 static int ctl_abort_task_set(union ctl_io *io);
 487 static int ctl_query_task(union ctl_io *io, int task_set);
 488 static void ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
 489                               ctl_ua_type ua_type);
 490 static int ctl_i_t_nexus_reset(union ctl_io *io);
 491 static int ctl_query_async_event(union ctl_io *io);
 492 static void ctl_run_task(union ctl_io *io);
 493 #ifdef CTL_IO_DELAY
 494 static void ctl_datamove_timer_wakeup(void *arg);
 495 static void ctl_done_timer_wakeup(void *arg);
 496 #endif /* CTL_IO_DELAY */
 497
 498 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 499 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 500 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 501 static void ctl_datamove_remote_write(union ctl_io *io);
 502 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 503 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 504 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 505 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 506                                     ctl_ha_dt_cb callback);
 507 static void ctl_datamove_remote_read(union ctl_io *io);
 508 static void ctl_datamove_remote(union ctl_io *io);
 509 static void ctl_process_done(union ctl_io *io);
 510 static void ctl_lun_thread(void *arg);
 511 static void ctl_thresh_thread(void *arg);
 512 static void ctl_work_thread(void *arg);
 513 static void ctl_enqueue_incoming(union ctl_io *io);
 514 static void ctl_enqueue_rtr(union ctl_io *io);
 515 static void ctl_enqueue_done(union ctl_io *io);
 516 static void ctl_enqueue_isc(union ctl_io *io);
 517 static const struct ctl_cmd_entry *
 518     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 519 static const struct ctl_cmd_entry *
 520     ctl_validate_command(struct ctl_scsiio *ctsio);
 521 static int ctl_cmd_applicable(uint8_t lun_type,
 522     const struct ctl_cmd_entry *entry);
 523 static int ctl_ha_init(void);
 524 static int ctl_ha_shutdown(void);
 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         .init = ctl_ha_init,
 567         .shutdown = ctl_ha_shutdown,
 568 };
 569
 570 static int
 571 ctl_ha_init(void)
 572 {
 573         struct ctl_softc *softc = control_softc;
 574
 575         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
 576                             &softc->othersc_pool) != 0)
 577                 return (ENOMEM);
 578         if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
 579                 ctl_pool_free(softc->othersc_pool);
 580                 return (EIO);
 581         }
 582         if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
 583             != CTL_HA_STATUS_SUCCESS) {
 584                 ctl_ha_msg_destroy(softc);
 585                 ctl_pool_free(softc->othersc_pool);
 586                 return (EIO);
 587         }
 588         return (0);
 589 };
 590
 591 static int
 592 ctl_ha_shutdown(void)
 593 {
 594         struct ctl_softc *softc = control_softc;
 595         struct ctl_port *port;
 596
 597         ctl_ha_msg_shutdown(softc);
 598         if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL) != CTL_HA_STATUS_SUCCESS)
 599                 return (EIO);
 600         if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
 601                 return (EIO);
 602         ctl_pool_free(softc->othersc_pool);
 603         while ((port = STAILQ_FIRST(&ha_frontend.port_list)) != NULL) {
 604                 ctl_port_deregister(port);
 605                 free(port->port_name, M_CTL);
 606                 free(port, M_CTL);
 607         }
 608         return (0);
 609 };
 610
 611 static void
 612 ctl_ha_datamove(union ctl_io *io)
 613 {
 614         struct ctl_lun *lun = CTL_LUN(io);
 615         struct ctl_sg_entry *sgl;
 616         union ctl_ha_msg msg;
 617         uint32_t sg_entries_sent;
 618         int do_sg_copy, i, j;
 619
 620         memset(&msg.dt, 0, sizeof(msg.dt));
 621         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
 622         msg.hdr.original_sc = io->io_hdr.original_sc;
 623         msg.hdr.serializing_sc = io;
 624         msg.hdr.nexus = io->io_hdr.nexus;
 625         msg.hdr.status = io->io_hdr.status;
 626         msg.dt.flags = io->io_hdr.flags;
 627
 628         /*
 629          * We convert everything into a S/G list here.  We can't
 630          * pass by reference, only by value between controllers.
 631          * So we can't pass a pointer to the S/G list, only as many
 632          * S/G entries as we can fit in here.  If it's possible for
 633          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
 634          * then we need to break this up into multiple transfers.
 635          */
 636         if (io->scsiio.kern_sg_entries == 0) {
 637                 msg.dt.kern_sg_entries = 1;
 638 #if 0
 639                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 640                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 641                 } else {
 642                         /* XXX KDM use busdma here! */
 643                         msg.dt.sg_list[0].addr =
 644                             (void *)vtophys(io->scsiio.kern_data_ptr);
 645                 }
 646 #else
 647                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
 648                     ("HA does not support BUS_ADDR"));
 649                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 650 #endif
 651                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
 652                 do_sg_copy = 0;
 653         } else {
 654                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
 655                 do_sg_copy = 1;
 656         }
 657
 658         msg.dt.kern_data_len = io->scsiio.kern_data_len;
 659         msg.dt.kern_total_len = io->scsiio.kern_total_len;
 660         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
 661         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
 662         msg.dt.sg_sequence = 0;
 663
 664         /*
 665          * Loop until we've sent all of the S/G entries.  On the
 666          * other end, we'll recompose these S/G entries into one
 667          * contiguous list before processing.
 668          */
 669         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
 670             msg.dt.sg_sequence++) {
 671                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
 672                     sizeof(msg.dt.sg_list[0])),
 673                     msg.dt.kern_sg_entries - sg_entries_sent);
 674                 if (do_sg_copy != 0) {
 675                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
 676                         for (i = sg_entries_sent, j = 0;
 677                              i < msg.dt.cur_sg_entries; i++, j++) {
 678 #if 0
 679                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 680                                         msg.dt.sg_list[j].addr = sgl[i].addr;
 681                                 } else {
 682                                         /* XXX KDM use busdma here! */
 683                                         msg.dt.sg_list[j].addr =
 684                                             (void *)vtophys(sgl[i].addr);
 685                                 }
 686 #else
 687                                 KASSERT((io->io_hdr.flags &
 688                                     CTL_FLAG_BUS_ADDR) == 0,
 689                                     ("HA does not support BUS_ADDR"));
 690                                 msg.dt.sg_list[j].addr = sgl[i].addr;
 691 #endif
 692                                 msg.dt.sg_list[j].len = sgl[i].len;
 693                         }
 694                 }
 695
 696                 sg_entries_sent += msg.dt.cur_sg_entries;
 697                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
 698                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 699                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
 700                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
 701                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
 702                         io->io_hdr.port_status = 31341;
 703                         io->scsiio.be_move_done(io);
 704                         return;
 705                 }
 706                 msg.dt.sent_sg_entries = sg_entries_sent;
 707         }
 708
 709         /*
 710          * Officially handover the request from us to peer.
 711          * If failover has just happened, then we must return error.
 712          * If failover happen just after, then it is not our problem.
 713          */
 714         if (lun)
 715                 mtx_lock(&lun->lun_lock);
 716         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
 717                 if (lun)
 718                         mtx_unlock(&lun->lun_lock);
 719                 io->io_hdr.port_status = 31342;
 720                 io->scsiio.be_move_done(io);
 721                 return;
 722         }
 723         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
 724         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
 725         if (lun)
 726                 mtx_unlock(&lun->lun_lock);
 727 }
 728
 729 static void
 730 ctl_ha_done(union ctl_io *io)
 731 {
 732         union ctl_ha_msg msg;
 733
 734         if (io->io_hdr.io_type == CTL_IO_SCSI) {
 735                 memset(&msg, 0, sizeof(msg));
 736                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
 737                 msg.hdr.original_sc = io->io_hdr.original_sc;
 738                 msg.hdr.nexus = io->io_hdr.nexus;
 739                 msg.hdr.status = io->io_hdr.status;
 740                 msg.scsi.scsi_status = io->scsiio.scsi_status;
 741                 msg.scsi.tag_num = io->scsiio.tag_num;
 742                 msg.scsi.tag_type = io->scsiio.tag_type;
 743                 msg.scsi.sense_len = io->scsiio.sense_len;
 744                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
 745                     io->scsiio.sense_len);
 746                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 747                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
 748                     msg.scsi.sense_len, M_WAITOK);
 749         }
 750         ctl_free_io(io);
 751 }
 752
 753 static void
 754 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 755                             union ctl_ha_msg *msg_info)
 756 {
 757         struct ctl_scsiio *ctsio;
 758
 759         if (msg_info->hdr.original_sc == NULL) {
 760                 printf("%s: original_sc == NULL!\n", __func__);
 761                 /* XXX KDM now what? */
 762                 return;
 763         }
 764
 765         ctsio = &msg_info->hdr.original_sc->scsiio;
 766         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 767         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 768         ctsio->io_hdr.status = msg_info->hdr.status;
 769         ctsio->scsi_status = msg_info->scsi.scsi_status;
 770         ctsio->sense_len = msg_info->scsi.sense_len;
 771         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 772                msg_info->scsi.sense_len);
 773         ctl_enqueue_isc((union ctl_io *)ctsio);
 774 }
 775
 776 static void
 777 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 778                                 union ctl_ha_msg *msg_info)
 779 {
 780         struct ctl_scsiio *ctsio;
 781
 782         if (msg_info->hdr.serializing_sc == NULL) {
 783                 printf("%s: serializing_sc == NULL!\n", __func__);
 784                 /* XXX KDM now what? */
 785                 return;
 786         }
 787
 788         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 789         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 790         ctl_enqueue_isc((union ctl_io *)ctsio);
 791 }
 792
 793 void
 794 ctl_isc_announce_lun(struct ctl_lun *lun)
 795 {
 796         struct ctl_softc *softc = lun->ctl_softc;
 797         union ctl_ha_msg *msg;
 798         struct ctl_ha_msg_lun_pr_key pr_key;
 799         int i, k;
 800
 801         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 802                 return;
 803         mtx_lock(&lun->lun_lock);
 804         i = sizeof(msg->lun);
 805         if (lun->lun_devid)
 806                 i += lun->lun_devid->len;
 807         i += sizeof(pr_key) * lun->pr_key_count;
 808 alloc:
 809         mtx_unlock(&lun->lun_lock);
 810         msg = malloc(i, M_CTL, M_WAITOK);
 811         mtx_lock(&lun->lun_lock);
 812         k = sizeof(msg->lun);
 813         if (lun->lun_devid)
 814                 k += lun->lun_devid->len;
 815         k += sizeof(pr_key) * lun->pr_key_count;
 816         if (i < k) {
 817                 free(msg, M_CTL);
 818                 i = k;
 819                 goto alloc;
 820         }
 821         bzero(&msg->lun, sizeof(msg->lun));
 822         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 823         msg->hdr.nexus.targ_lun = lun->lun;
 824         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 825         msg->lun.flags = lun->flags;
 826         msg->lun.pr_generation = lun->pr_generation;
 827         msg->lun.pr_res_idx = lun->pr_res_idx;
 828         msg->lun.pr_res_type = lun->pr_res_type;
 829         msg->lun.pr_key_count = lun->pr_key_count;
 830         i = 0;
 831         if (lun->lun_devid) {
 832                 msg->lun.lun_devid_len = lun->lun_devid->len;
 833                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 834                     msg->lun.lun_devid_len);
 835                 i += msg->lun.lun_devid_len;
 836         }
 837         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 838                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 839                         continue;
 840                 pr_key.pr_iid = k;
 841                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 842                 i += sizeof(pr_key);
 843         }
 844         mtx_unlock(&lun->lun_lock);
 845         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 846             M_WAITOK);
 847         free(msg, M_CTL);
 848
 849         if (lun->flags & CTL_LUN_PRIMARY_SC) {
 850                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 851                         ctl_isc_announce_mode(lun, -1,
 852                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
 853                             lun->mode_pages.index[i].subpage);
 854                 }
 855         }
 856 }
 857
 858 void
 859 ctl_isc_announce_port(struct ctl_port *port)
 860 {
 861         struct ctl_softc *softc = port->ctl_softc;
 862         union ctl_ha_msg *msg;
 863         int i;
 864
 865         if (port->targ_port < softc->port_min ||
 866             port->targ_port >= softc->port_max ||
 867             softc->ha_link != CTL_HA_LINK_ONLINE)
 868                 return;
 869         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 870         if (port->lun_map)
 871                 i += port->lun_map_size * sizeof(uint32_t);
 872         if (port->port_devid)
 873                 i += port->port_devid->len;
 874         if (port->target_devid)
 875                 i += port->target_devid->len;
 876         if (port->init_devid)
 877                 i += port->init_devid->len;
 878         msg = malloc(i, M_CTL, M_WAITOK);
 879         bzero(&msg->port, sizeof(msg->port));
 880         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 881         msg->hdr.nexus.targ_port = port->targ_port;
 882         msg->port.port_type = port->port_type;
 883         msg->port.physical_port = port->physical_port;
 884         msg->port.virtual_port = port->virtual_port;
 885         msg->port.status = port->status;
 886         i = 0;
 887         msg->port.name_len = sprintf(&msg->port.data[i],
 888             "%d:%s", softc->ha_id, port->port_name) + 1;
 889         i += msg->port.name_len;
 890         if (port->lun_map) {
 891                 msg->port.lun_map_len = port->lun_map_size * sizeof(uint32_t);
 892                 memcpy(&msg->port.data[i], port->lun_map,
 893                     msg->port.lun_map_len);
 894                 i += msg->port.lun_map_len;
 895         }
 896         if (port->port_devid) {
 897                 msg->port.port_devid_len = port->port_devid->len;
 898                 memcpy(&msg->port.data[i], port->port_devid->data,
 899                     msg->port.port_devid_len);
 900                 i += msg->port.port_devid_len;
 901         }
 902         if (port->target_devid) {
 903                 msg->port.target_devid_len = port->target_devid->len;
 904                 memcpy(&msg->port.data[i], port->target_devid->data,
 905                     msg->port.target_devid_len);
 906                 i += msg->port.target_devid_len;
 907         }
 908         if (port->init_devid) {
 909                 msg->port.init_devid_len = port->init_devid->len;
 910                 memcpy(&msg->port.data[i], port->init_devid->data,
 911                     msg->port.init_devid_len);
 912                 i += msg->port.init_devid_len;
 913         }
 914         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 915             M_WAITOK);
 916         free(msg, M_CTL);
 917 }
 918
 919 void
 920 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 921 {
 922         struct ctl_softc *softc = port->ctl_softc;
 923         union ctl_ha_msg *msg;
 924         int i, l;
 925
 926         if (port->targ_port < softc->port_min ||
 927             port->targ_port >= softc->port_max ||
 928             softc->ha_link != CTL_HA_LINK_ONLINE)
 929                 return;
 930         mtx_lock(&softc->ctl_lock);
 931         i = sizeof(msg->iid);
 932         l = 0;
 933         if (port->wwpn_iid[iid].name)
 934                 l = strlen(port->wwpn_iid[iid].name) + 1;
 935         i += l;
 936         msg = malloc(i, M_CTL, M_NOWAIT);
 937         if (msg == NULL) {
 938                 mtx_unlock(&softc->ctl_lock);
 939                 return;
 940         }
 941         bzero(&msg->iid, sizeof(msg->iid));
 942         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 943         msg->hdr.nexus.targ_port = port->targ_port;
 944         msg->hdr.nexus.initid = iid;
 945         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 946         msg->iid.name_len = l;
 947         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 948         if (port->wwpn_iid[iid].name)
 949                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 950         mtx_unlock(&softc->ctl_lock);
 951         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 952         free(msg, M_CTL);
 953 }
 954
 955 void
 956 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
 957     uint8_t page, uint8_t subpage)
 958 {
 959         struct ctl_softc *softc = lun->ctl_softc;
 960         union ctl_ha_msg msg;
 961         u_int i;
 962
 963         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 964                 return;
 965         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 966                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
 967                     page && lun->mode_pages.index[i].subpage == subpage)
 968                         break;
 969         }
 970         if (i == CTL_NUM_MODE_PAGES)
 971                 return;
 972
 973         /* Don't try to replicate pages not present on this device. */
 974         if (lun->mode_pages.index[i].page_data == NULL)
 975                 return;
 976
 977         bzero(&msg.mode, sizeof(msg.mode));
 978         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
 979         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
 980         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
 981         msg.hdr.nexus.targ_lun = lun->lun;
 982         msg.hdr.nexus.targ_mapped_lun = lun->lun;
 983         msg.mode.page_code = page;
 984         msg.mode.subpage = subpage;
 985         msg.mode.page_len = lun->mode_pages.index[i].page_len;
 986         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
 987             msg.mode.page_len);
 988         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
 989             M_WAITOK);
 990 }
 991
 992 static void
 993 ctl_isc_ha_link_up(struct ctl_softc *softc)
 994 {
 995         struct ctl_port *port;
 996         struct ctl_lun *lun;
 997         union ctl_ha_msg msg;
 998         int i;
 999
1000         /* Announce this node parameters to peer for validation. */
1001         msg.login.msg_type = CTL_MSG_LOGIN;
1002         msg.login.version = CTL_HA_VERSION;
1003         msg.login.ha_mode = softc->ha_mode;
1004         msg.login.ha_id = softc->ha_id;
1005         msg.login.max_luns = CTL_MAX_LUNS;
1006         msg.login.max_ports = CTL_MAX_PORTS;
1007         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
1008         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
1009             M_WAITOK);
1010
1011         STAILQ_FOREACH(port, &softc->port_list, links) {
1012                 ctl_isc_announce_port(port);
1013                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1014                         if (port->wwpn_iid[i].in_use)
1015                                 ctl_isc_announce_iid(port, i);
1016                 }
1017         }
1018         STAILQ_FOREACH(lun, &softc->lun_list, links)
1019                 ctl_isc_announce_lun(lun);
1020 }
1021
1022 static void
1023 ctl_isc_ha_link_down(struct ctl_softc *softc)
1024 {
1025         struct ctl_port *port;
1026         struct ctl_lun *lun;
1027         union ctl_io *io;
1028         int i;
1029
1030         mtx_lock(&softc->ctl_lock);
1031         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1032                 mtx_lock(&lun->lun_lock);
1033                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
1034                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1035                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1036                 }
1037                 mtx_unlock(&lun->lun_lock);
1038
1039                 mtx_unlock(&softc->ctl_lock);
1040                 io = ctl_alloc_io(softc->othersc_pool);
1041                 mtx_lock(&softc->ctl_lock);
1042                 ctl_zero_io(io);
1043                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1044                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1045                 ctl_enqueue_isc(io);
1046         }
1047
1048         STAILQ_FOREACH(port, &softc->port_list, links) {
1049                 if (port->targ_port >= softc->port_min &&
1050                     port->targ_port < softc->port_max)
1051                         continue;
1052                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1053                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1054                         port->wwpn_iid[i].in_use = 0;
1055                         free(port->wwpn_iid[i].name, M_CTL);
1056                         port->wwpn_iid[i].name = NULL;
1057                 }
1058         }
1059         mtx_unlock(&softc->ctl_lock);
1060 }
1061
1062 static void
1063 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1064 {
1065         struct ctl_lun *lun;
1066         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
1067
1068         mtx_lock(&softc->ctl_lock);
1069         if (msg->hdr.nexus.targ_mapped_lun >= CTL_MAX_LUNS ||
1070             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) == NULL) {
1071                 mtx_unlock(&softc->ctl_lock);
1072                 return;
1073         }
1074         mtx_lock(&lun->lun_lock);
1075         mtx_unlock(&softc->ctl_lock);
1076         if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES && msg->ua.ua_set)
1077                 memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1078         if (msg->ua.ua_all) {
1079                 if (msg->ua.ua_set)
1080                         ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1081                 else
1082                         ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1083         } else {
1084                 if (msg->ua.ua_set)
1085                         ctl_est_ua(lun, iid, msg->ua.ua_type);
1086                 else
1087                         ctl_clr_ua(lun, iid, msg->ua.ua_type);
1088         }
1089         mtx_unlock(&lun->lun_lock);
1090 }
1091
1092 static void
1093 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1094 {
1095         struct ctl_lun *lun;
1096         struct ctl_ha_msg_lun_pr_key pr_key;
1097         int i, k;
1098         ctl_lun_flags oflags;
1099         uint32_t targ_lun;
1100
1101         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1102         mtx_lock(&softc->ctl_lock);
1103         if (targ_lun >= CTL_MAX_LUNS ||
1104             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1105                 mtx_unlock(&softc->ctl_lock);
1106                 return;
1107         }
1108         mtx_lock(&lun->lun_lock);
1109         mtx_unlock(&softc->ctl_lock);
1110         if (lun->flags & CTL_LUN_DISABLED) {
1111                 mtx_unlock(&lun->lun_lock);
1112                 return;
1113         }
1114         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1115         if (msg->lun.lun_devid_len != i || (i > 0 &&
1116             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1117                 mtx_unlock(&lun->lun_lock);
1118                 printf("%s: Received conflicting HA LUN %d\n",
1119                     __func__, targ_lun);
1120                 return;
1121         } else {
1122                 /* Record whether peer is primary. */
1123                 oflags = lun->flags;
1124                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1125                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1126                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1127                 else
1128                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1129                 if (oflags != lun->flags)
1130                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1131
1132                 /* If peer is primary and we are not -- use data */
1133                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1134                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1135                         lun->pr_generation = msg->lun.pr_generation;
1136                         lun->pr_res_idx = msg->lun.pr_res_idx;
1137                         lun->pr_res_type = msg->lun.pr_res_type;
1138                         lun->pr_key_count = msg->lun.pr_key_count;
1139                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1140                                 ctl_clr_prkey(lun, k);
1141                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1142                                 memcpy(&pr_key, &msg->lun.data[i],
1143                                     sizeof(pr_key));
1144                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1145                                 ctl_set_prkey(lun, pr_key.pr_iid,
1146                                     pr_key.pr_key);
1147                                 i += sizeof(pr_key);
1148                         }
1149                 }
1150
1151                 mtx_unlock(&lun->lun_lock);
1152                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1153                     __func__, targ_lun,
1154                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1155                     "primary" : "secondary"));
1156
1157                 /* If we are primary but peer doesn't know -- notify */
1158                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1159                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1160                         ctl_isc_announce_lun(lun);
1161         }
1162 }
1163
1164 static void
1165 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1166 {
1167         struct ctl_port *port;
1168         struct ctl_lun *lun;
1169         int i, new;
1170
1171         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1172         if (port == NULL) {
1173                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1174                     msg->hdr.nexus.targ_port));
1175                 new = 1;
1176                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1177                 port->frontend = &ha_frontend;
1178                 port->targ_port = msg->hdr.nexus.targ_port;
1179                 port->fe_datamove = ctl_ha_datamove;
1180                 port->fe_done = ctl_ha_done;
1181         } else if (port->frontend == &ha_frontend) {
1182                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1183                     msg->hdr.nexus.targ_port));
1184                 new = 0;
1185         } else {
1186                 printf("%s: Received conflicting HA port %d\n",
1187                     __func__, msg->hdr.nexus.targ_port);
1188                 return;
1189         }
1190         port->port_type = msg->port.port_type;
1191         port->physical_port = msg->port.physical_port;
1192         port->virtual_port = msg->port.virtual_port;
1193         port->status = msg->port.status;
1194         i = 0;
1195         free(port->port_name, M_CTL);
1196         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1197             M_CTL);
1198         i += msg->port.name_len;
1199         if (msg->port.lun_map_len != 0) {
1200                 if (port->lun_map == NULL ||
1201                     port->lun_map_size * sizeof(uint32_t) <
1202                     msg->port.lun_map_len) {
1203                         port->lun_map_size = 0;
1204                         free(port->lun_map, M_CTL);
1205                         port->lun_map = malloc(msg->port.lun_map_len,
1206                             M_CTL, M_WAITOK);
1207                 }
1208                 memcpy(port->lun_map, &msg->port.data[i], msg->port.lun_map_len);
1209                 port->lun_map_size = msg->port.lun_map_len / sizeof(uint32_t);
1210                 i += msg->port.lun_map_len;
1211         } else {
1212                 port->lun_map_size = 0;
1213                 free(port->lun_map, M_CTL);
1214                 port->lun_map = NULL;
1215         }
1216         if (msg->port.port_devid_len != 0) {
1217                 if (port->port_devid == NULL ||
1218                     port->port_devid->len < msg->port.port_devid_len) {
1219                         free(port->port_devid, M_CTL);
1220                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1221                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1222                 }
1223                 memcpy(port->port_devid->data, &msg->port.data[i],
1224                     msg->port.port_devid_len);
1225                 port->port_devid->len = msg->port.port_devid_len;
1226                 i += msg->port.port_devid_len;
1227         } else {
1228                 free(port->port_devid, M_CTL);
1229                 port->port_devid = NULL;
1230         }
1231         if (msg->port.target_devid_len != 0) {
1232                 if (port->target_devid == NULL ||
1233                     port->target_devid->len < msg->port.target_devid_len) {
1234                         free(port->target_devid, M_CTL);
1235                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1236                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1237                 }
1238                 memcpy(port->target_devid->data, &msg->port.data[i],
1239                     msg->port.target_devid_len);
1240                 port->target_devid->len = msg->port.target_devid_len;
1241                 i += msg->port.target_devid_len;
1242         } else {
1243                 free(port->target_devid, M_CTL);
1244                 port->target_devid = NULL;
1245         }
1246         if (msg->port.init_devid_len != 0) {
1247                 if (port->init_devid == NULL ||
1248                     port->init_devid->len < msg->port.init_devid_len) {
1249                         free(port->init_devid, M_CTL);
1250                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1251                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1252                 }
1253                 memcpy(port->init_devid->data, &msg->port.data[i],
1254                     msg->port.init_devid_len);
1255                 port->init_devid->len = msg->port.init_devid_len;
1256                 i += msg->port.init_devid_len;
1257         } else {
1258                 free(port->init_devid, M_CTL);
1259                 port->init_devid = NULL;
1260         }
1261         if (new) {
1262                 if (ctl_port_register(port) != 0) {
1263                         printf("%s: ctl_port_register() failed with error\n",
1264                             __func__);
1265                 }
1266         }
1267         mtx_lock(&softc->ctl_lock);
1268         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1269                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
1270                         continue;
1271                 mtx_lock(&lun->lun_lock);
1272                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1273                 mtx_unlock(&lun->lun_lock);
1274         }
1275         mtx_unlock(&softc->ctl_lock);
1276 }
1277
1278 static void
1279 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1280 {
1281         struct ctl_port *port;
1282         int iid;
1283
1284         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1285         if (port == NULL) {
1286                 printf("%s: Received IID for unknown port %d\n",
1287                     __func__, msg->hdr.nexus.targ_port);
1288                 return;
1289         }
1290         iid = msg->hdr.nexus.initid;
1291         if (port->wwpn_iid[iid].in_use != 0 &&
1292             msg->iid.in_use == 0)
1293                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
1294         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1295         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1296         free(port->wwpn_iid[iid].name, M_CTL);
1297         if (msg->iid.name_len) {
1298                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1299                     msg->iid.name_len, M_CTL);
1300         } else
1301                 port->wwpn_iid[iid].name = NULL;
1302 }
1303
1304 static void
1305 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1306 {
1307
1308         if (msg->login.version != CTL_HA_VERSION) {
1309                 printf("CTL HA peers have different versions %d != %d\n",
1310                     msg->login.version, CTL_HA_VERSION);
1311                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1312                 return;
1313         }
1314         if (msg->login.ha_mode != softc->ha_mode) {
1315                 printf("CTL HA peers have different ha_mode %d != %d\n",
1316                     msg->login.ha_mode, softc->ha_mode);
1317                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1318                 return;
1319         }
1320         if (msg->login.ha_id == softc->ha_id) {
1321                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1322                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1323                 return;
1324         }
1325         if (msg->login.max_luns != CTL_MAX_LUNS ||
1326             msg->login.max_ports != CTL_MAX_PORTS ||
1327             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1328                 printf("CTL HA peers have different limits\n");
1329                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1330                 return;
1331         }
1332 }
1333
1334 static void
1335 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1336 {
1337         struct ctl_lun *lun;
1338         u_int i;
1339         uint32_t initidx, targ_lun;
1340
1341         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1342         mtx_lock(&softc->ctl_lock);
1343         if (targ_lun >= CTL_MAX_LUNS ||
1344             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1345                 mtx_unlock(&softc->ctl_lock);
1346                 return;
1347         }
1348         mtx_lock(&lun->lun_lock);
1349         mtx_unlock(&softc->ctl_lock);
1350         if (lun->flags & CTL_LUN_DISABLED) {
1351                 mtx_unlock(&lun->lun_lock);
1352                 return;
1353         }
1354         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1355                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1356                     msg->mode.page_code &&
1357                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1358                         break;
1359         }
1360         if (i == CTL_NUM_MODE_PAGES) {
1361                 mtx_unlock(&lun->lun_lock);
1362                 return;
1363         }
1364         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1365             lun->mode_pages.index[i].page_len);
1366         initidx = ctl_get_initindex(&msg->hdr.nexus);
1367         if (initidx != -1)
1368                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1369         mtx_unlock(&lun->lun_lock);
1370 }
1371
1372 /*
1373  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1374  * subsystem come in here.
1375  */
1376 static void
1377 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1378 {
1379         struct ctl_softc *softc = control_softc;
1380         union ctl_io *io;
1381         struct ctl_prio *presio;
1382         ctl_ha_status isc_status;
1383
1384         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1385         if (event == CTL_HA_EVT_MSG_RECV) {
1386                 union ctl_ha_msg *msg, msgbuf;
1387
1388                 if (param > sizeof(msgbuf))
1389                         msg = malloc(param, M_CTL, M_WAITOK);
1390                 else
1391                         msg = &msgbuf;
1392                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1393                     M_WAITOK);
1394                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1395                         printf("%s: Error receiving message: %d\n",
1396                             __func__, isc_status);
1397                         if (msg != &msgbuf)
1398                                 free(msg, M_CTL);
1399                         return;
1400                 }
1401
1402                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1403                 switch (msg->hdr.msg_type) {
1404                 case CTL_MSG_SERIALIZE:
1405                         io = ctl_alloc_io(softc->othersc_pool);
1406                         ctl_zero_io(io);
1407                         // populate ctsio from msg
1408                         io->io_hdr.io_type = CTL_IO_SCSI;
1409                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1410                         io->io_hdr.original_sc = msg->hdr.original_sc;
1411                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1412                                             CTL_FLAG_IO_ACTIVE;
1413                         /*
1414                          * If we're in serialization-only mode, we don't
1415                          * want to go through full done processing.  Thus
1416                          * the COPY flag.
1417                          *
1418                          * XXX KDM add another flag that is more specific.
1419                          */
1420                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1421                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1422                         io->io_hdr.nexus = msg->hdr.nexus;
1423 #if 0
1424                         printf("port %u, iid %u, lun %u\n",
1425                                io->io_hdr.nexus.targ_port,
1426                                io->io_hdr.nexus.initid,
1427                                io->io_hdr.nexus.targ_lun);
1428 #endif
1429                         io->scsiio.tag_num = msg->scsi.tag_num;
1430                         io->scsiio.tag_type = msg->scsi.tag_type;
1431 #ifdef CTL_TIME_IO
1432                         io->io_hdr.start_time = time_uptime;
1433                         getbinuptime(&io->io_hdr.start_bt);
1434 #endif /* CTL_TIME_IO */
1435                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1436                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1437                                CTL_MAX_CDBLEN);
1438                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1439                                 const struct ctl_cmd_entry *entry;
1440
1441                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1442                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1443                                 io->io_hdr.flags |=
1444                                         entry->flags & CTL_FLAG_DATA_MASK;
1445                         }
1446                         ctl_enqueue_isc(io);
1447                         break;
1448
1449                 /* Performed on the Originating SC, XFER mode only */
1450                 case CTL_MSG_DATAMOVE: {
1451                         struct ctl_sg_entry *sgl;
1452                         int i, j;
1453
1454                         io = msg->hdr.original_sc;
1455                         if (io == NULL) {
1456                                 printf("%s: original_sc == NULL!\n", __func__);
1457                                 /* XXX KDM do something here */
1458                                 break;
1459                         }
1460                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1461                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1462                         /*
1463                          * Keep track of this, we need to send it back over
1464                          * when the datamove is complete.
1465                          */
1466                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1467                         if (msg->hdr.status == CTL_SUCCESS)
1468                                 io->io_hdr.status = msg->hdr.status;
1469
1470                         if (msg->dt.sg_sequence == 0) {
1471 #ifdef CTL_TIME_IO
1472                                 getbinuptime(&io->io_hdr.dma_start_bt);
1473 #endif
1474                                 i = msg->dt.kern_sg_entries +
1475                                     msg->dt.kern_data_len /
1476                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1477                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1478                                     M_WAITOK | M_ZERO);
1479                                 io->io_hdr.remote_sglist = sgl;
1480                                 io->io_hdr.local_sglist =
1481                                     &sgl[msg->dt.kern_sg_entries];
1482
1483                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1484
1485                                 io->scsiio.kern_sg_entries =
1486                                         msg->dt.kern_sg_entries;
1487                                 io->scsiio.rem_sg_entries =
1488                                         msg->dt.kern_sg_entries;
1489                                 io->scsiio.kern_data_len =
1490                                         msg->dt.kern_data_len;
1491                                 io->scsiio.kern_total_len =
1492                                         msg->dt.kern_total_len;
1493                                 io->scsiio.kern_data_resid =
1494                                         msg->dt.kern_data_resid;
1495                                 io->scsiio.kern_rel_offset =
1496                                         msg->dt.kern_rel_offset;
1497                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1498                                 io->io_hdr.flags |= msg->dt.flags &
1499                                     CTL_FLAG_BUS_ADDR;
1500                         } else
1501                                 sgl = (struct ctl_sg_entry *)
1502                                         io->scsiio.kern_data_ptr;
1503
1504                         for (i = msg->dt.sent_sg_entries, j = 0;
1505                              i < (msg->dt.sent_sg_entries +
1506                              msg->dt.cur_sg_entries); i++, j++) {
1507                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1508                                 sgl[i].len = msg->dt.sg_list[j].len;
1509
1510 #if 0
1511                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1512                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1513 #endif
1514                         }
1515
1516                         /*
1517                          * If this is the last piece of the I/O, we've got
1518                          * the full S/G list.  Queue processing in the thread.
1519                          * Otherwise wait for the next piece.
1520                          */
1521                         if (msg->dt.sg_last != 0)
1522                                 ctl_enqueue_isc(io);
1523                         break;
1524                 }
1525                 /* Performed on the Serializing (primary) SC, XFER mode only */
1526                 case CTL_MSG_DATAMOVE_DONE: {
1527                         if (msg->hdr.serializing_sc == NULL) {
1528                                 printf("%s: serializing_sc == NULL!\n",
1529                                        __func__);
1530                                 /* XXX KDM now what? */
1531                                 break;
1532                         }
1533                         /*
1534                          * We grab the sense information here in case
1535                          * there was a failure, so we can return status
1536                          * back to the initiator.
1537                          */
1538                         io = msg->hdr.serializing_sc;
1539                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1540                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1541                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1542                         io->io_hdr.port_status = msg->scsi.port_status;
1543                         io->scsiio.kern_data_resid = msg->scsi.kern_data_resid;
1544                         if (msg->hdr.status != CTL_STATUS_NONE) {
1545                                 io->io_hdr.status = msg->hdr.status;
1546                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1547                                 io->scsiio.sense_len = msg->scsi.sense_len;
1548                                 memcpy(&io->scsiio.sense_data,
1549                                     &msg->scsi.sense_data,
1550                                     msg->scsi.sense_len);
1551                                 if (msg->hdr.status == CTL_SUCCESS)
1552                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1553                         }
1554                         ctl_enqueue_isc(io);
1555                         break;
1556                 }
1557
1558                 /* Preformed on Originating SC, SER_ONLY mode */
1559                 case CTL_MSG_R2R:
1560                         io = msg->hdr.original_sc;
1561                         if (io == NULL) {
1562                                 printf("%s: original_sc == NULL!\n",
1563                                     __func__);
1564                                 break;
1565                         }
1566                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1567                         io->io_hdr.msg_type = CTL_MSG_R2R;
1568                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1569                         ctl_enqueue_isc(io);
1570                         break;
1571
1572                 /*
1573                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1574                  * mode.
1575                  * Performed on the Originating (i.e. secondary) SC in XFER
1576                  * mode
1577                  */
1578                 case CTL_MSG_FINISH_IO:
1579                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1580                                 ctl_isc_handler_finish_xfer(softc, msg);
1581                         else
1582                                 ctl_isc_handler_finish_ser_only(softc, msg);
1583                         break;
1584
1585                 /* Preformed on Originating SC */
1586                 case CTL_MSG_BAD_JUJU:
1587                         io = msg->hdr.original_sc;
1588                         if (io == NULL) {
1589                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1590                                        __func__);
1591                                 break;
1592                         }
1593                         ctl_copy_sense_data(msg, io);
1594                         /*
1595                          * IO should have already been cleaned up on other
1596                          * SC so clear this flag so we won't send a message
1597                          * back to finish the IO there.
1598                          */
1599                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1600                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1601
1602                         /* io = msg->hdr.serializing_sc; */
1603                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1604                         ctl_enqueue_isc(io);
1605                         break;
1606
1607                 /* Handle resets sent from the other side */
1608                 case CTL_MSG_MANAGE_TASKS: {
1609                         struct ctl_taskio *taskio;
1610                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1611                             softc->othersc_pool);
1612                         ctl_zero_io((union ctl_io *)taskio);
1613                         taskio->io_hdr.io_type = CTL_IO_TASK;
1614                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1615                         taskio->io_hdr.nexus = msg->hdr.nexus;
1616                         taskio->task_action = msg->task.task_action;
1617                         taskio->tag_num = msg->task.tag_num;
1618                         taskio->tag_type = msg->task.tag_type;
1619 #ifdef CTL_TIME_IO
1620                         taskio->io_hdr.start_time = time_uptime;
1621                         getbinuptime(&taskio->io_hdr.start_bt);
1622 #endif /* CTL_TIME_IO */
1623                         ctl_run_task((union ctl_io *)taskio);
1624                         break;
1625                 }
1626                 /* Persistent Reserve action which needs attention */
1627                 case CTL_MSG_PERS_ACTION:
1628                         presio = (struct ctl_prio *)ctl_alloc_io(
1629                             softc->othersc_pool);
1630                         ctl_zero_io((union ctl_io *)presio);
1631                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1632                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1633                         presio->io_hdr.nexus = msg->hdr.nexus;
1634                         presio->pr_msg = msg->pr;
1635                         ctl_enqueue_isc((union ctl_io *)presio);
1636                         break;
1637                 case CTL_MSG_UA:
1638                         ctl_isc_ua(softc, msg, param);
1639                         break;
1640                 case CTL_MSG_PORT_SYNC:
1641                         ctl_isc_port_sync(softc, msg, param);
1642                         break;
1643                 case CTL_MSG_LUN_SYNC:
1644                         ctl_isc_lun_sync(softc, msg, param);
1645                         break;
1646                 case CTL_MSG_IID_SYNC:
1647                         ctl_isc_iid_sync(softc, msg, param);
1648                         break;
1649                 case CTL_MSG_LOGIN:
1650                         ctl_isc_login(softc, msg, param);
1651                         break;
1652                 case CTL_MSG_MODE_SYNC:
1653                         ctl_isc_mode_sync(softc, msg, param);
1654                         break;
1655                 default:
1656                         printf("Received HA message of unknown type %d\n",
1657                             msg->hdr.msg_type);
1658                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1659                         break;
1660                 }
1661                 if (msg != &msgbuf)
1662                         free(msg, M_CTL);
1663         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1664                 printf("CTL: HA link status changed from %d to %d\n",
1665                     softc->ha_link, param);
1666                 if (param == softc->ha_link)
1667                         return;
1668                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1669                         softc->ha_link = param;
1670                         ctl_isc_ha_link_down(softc);
1671                 } else {
1672                         softc->ha_link = param;
1673                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1674                                 ctl_isc_ha_link_up(softc);
1675                 }
1676                 return;
1677         } else {
1678                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1679                 return;
1680         }
1681 }
1682
1683 static void
1684 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1685 {
1686
1687         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1688             src->scsi.sense_len);
1689         dest->scsiio.scsi_status = src->scsi.scsi_status;
1690         dest->scsiio.sense_len = src->scsi.sense_len;
1691         dest->io_hdr.status = src->hdr.status;
1692 }
1693
1694 static void
1695 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1696 {
1697
1698         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1699             src->scsiio.sense_len);
1700         dest->scsi.scsi_status = src->scsiio.scsi_status;
1701         dest->scsi.sense_len = src->scsiio.sense_len;
1702         dest->hdr.status = src->io_hdr.status;
1703 }
1704
1705 void
1706 ctl_est_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_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1722 {
1723         int i;
1724
1725         mtx_assert(&lun->lun_lock, MA_OWNED);
1726         if (lun->pending_ua[port] == NULL)
1727                 return;
1728         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1729                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1730                         continue;
1731                 lun->pending_ua[port][i] |= ua;
1732         }
1733 }
1734
1735 void
1736 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1737 {
1738         struct ctl_softc *softc = lun->ctl_softc;
1739         int i;
1740
1741         mtx_assert(&lun->lun_lock, MA_OWNED);
1742         for (i = softc->port_min; i < softc->port_max; i++)
1743                 ctl_est_ua_port(lun, i, except, ua);
1744 }
1745
1746 void
1747 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1748 {
1749         struct ctl_softc *softc = lun->ctl_softc;
1750         ctl_ua_type *pu;
1751
1752         if (initidx < softc->init_min || initidx >= softc->init_max)
1753                 return;
1754         mtx_assert(&lun->lun_lock, MA_OWNED);
1755         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1756         if (pu == NULL)
1757                 return;
1758         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1759 }
1760
1761 void
1762 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1763 {
1764         struct ctl_softc *softc = lun->ctl_softc;
1765         int i, j;
1766
1767         mtx_assert(&lun->lun_lock, MA_OWNED);
1768         for (i = softc->port_min; i < softc->port_max; i++) {
1769                 if (lun->pending_ua[i] == NULL)
1770                         continue;
1771                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1772                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1773                                 continue;
1774                         lun->pending_ua[i][j] &= ~ua;
1775                 }
1776         }
1777 }
1778
1779 void
1780 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1781     ctl_ua_type ua_type)
1782 {
1783         struct ctl_lun *lun;
1784
1785         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1786         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1787                 mtx_lock(&lun->lun_lock);
1788                 ctl_clr_ua(lun, initidx, ua_type);
1789                 mtx_unlock(&lun->lun_lock);
1790         }
1791 }
1792
1793 static int
1794 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1795 {
1796         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1797         struct ctl_lun *lun;
1798         struct ctl_lun_req ireq;
1799         int error, value;
1800
1801         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1802         error = sysctl_handle_int(oidp, &value, 0, req);
1803         if ((error != 0) || (req->newptr == NULL))
1804                 return (error);
1805
1806         mtx_lock(&softc->ctl_lock);
1807         if (value == 0)
1808                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1809         else
1810                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1811         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1812                 mtx_unlock(&softc->ctl_lock);
1813                 bzero(&ireq, sizeof(ireq));
1814                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1815                 ireq.reqdata.modify.lun_id = lun->lun;
1816                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1817                     curthread);
1818                 if (ireq.status != CTL_LUN_OK) {
1819                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1820                             __func__, ireq.status, ireq.error_str);
1821                 }
1822                 mtx_lock(&softc->ctl_lock);
1823         }
1824         mtx_unlock(&softc->ctl_lock);
1825         return (0);
1826 }
1827
1828 static int
1829 ctl_init(void)
1830 {
1831         struct make_dev_args args;
1832         struct ctl_softc *softc;
1833         int i, error;
1834
1835         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1836                                M_WAITOK | M_ZERO);
1837
1838         make_dev_args_init(&args);
1839         args.mda_devsw = &ctl_cdevsw;
1840         args.mda_uid = UID_ROOT;
1841         args.mda_gid = GID_OPERATOR;
1842         args.mda_mode = 0600;
1843         args.mda_si_drv1 = softc;
1844         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1845         if (error != 0) {
1846                 free(softc, M_DEVBUF);
1847                 control_softc = NULL;
1848                 return (error);
1849         }
1850
1851         sysctl_ctx_init(&softc->sysctl_ctx);
1852         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1853                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1854                 CTLFLAG_RD, 0, "CAM Target Layer");
1855
1856         if (softc->sysctl_tree == NULL) {
1857                 printf("%s: unable to allocate sysctl tree\n", __func__);
1858                 destroy_dev(softc->dev);
1859                 free(softc, M_DEVBUF);
1860                 control_softc = NULL;
1861                 return (ENOMEM);
1862         }
1863
1864         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1865         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1866             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1867         softc->flags = 0;
1868
1869         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1870             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1871             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1872
1873         /*
1874          * In Copan's HA scheme, the "master" and "slave" roles are
1875          * figured out through the slot the controller is in.  Although it
1876          * is an active/active system, someone has to be in charge.
1877          */
1878         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1879             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1880             "HA head ID (0 - no HA)");
1881         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1882                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1883                 softc->is_single = 1;
1884                 softc->port_cnt = CTL_MAX_PORTS;
1885                 softc->port_min = 0;
1886         } else {
1887                 softc->port_cnt = CTL_MAX_PORTS / NUM_HA_SHELVES;
1888                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1889         }
1890         softc->port_max = softc->port_min + softc->port_cnt;
1891         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1892         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1893
1894         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1895             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1896             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1897
1898         STAILQ_INIT(&softc->lun_list);
1899         STAILQ_INIT(&softc->pending_lun_queue);
1900         STAILQ_INIT(&softc->fe_list);
1901         STAILQ_INIT(&softc->port_list);
1902         STAILQ_INIT(&softc->be_list);
1903         ctl_tpc_init(softc);
1904
1905         if (worker_threads <= 0)
1906                 worker_threads = max(1, mp_ncpus / 4);
1907         if (worker_threads > CTL_MAX_THREADS)
1908                 worker_threads = CTL_MAX_THREADS;
1909
1910         for (i = 0; i < worker_threads; i++) {
1911                 struct ctl_thread *thr = &softc->threads[i];
1912
1913                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1914                 thr->ctl_softc = softc;
1915                 STAILQ_INIT(&thr->incoming_queue);
1916                 STAILQ_INIT(&thr->rtr_queue);
1917                 STAILQ_INIT(&thr->done_queue);
1918                 STAILQ_INIT(&thr->isc_queue);
1919
1920                 error = kproc_kthread_add(ctl_work_thread, thr,
1921                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1922                 if (error != 0) {
1923                         printf("error creating CTL work thread!\n");
1924                         return (error);
1925                 }
1926         }
1927         error = kproc_kthread_add(ctl_lun_thread, softc,
1928             &softc->ctl_proc, &softc->lun_thread, 0, 0, "ctl", "lun");
1929         if (error != 0) {
1930                 printf("error creating CTL lun thread!\n");
1931                 return (error);
1932         }
1933         error = kproc_kthread_add(ctl_thresh_thread, softc,
1934             &softc->ctl_proc, &softc->thresh_thread, 0, 0, "ctl", "thresh");
1935         if (error != 0) {
1936                 printf("error creating CTL threshold thread!\n");
1937                 return (error);
1938         }
1939
1940         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1941             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1942             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1943
1944         if (softc->is_single == 0) {
1945                 if (ctl_frontend_register(&ha_frontend) != 0)
1946                         softc->is_single = 1;
1947         }
1948         return (0);
1949 }
1950
1951 static int
1952 ctl_shutdown(void)
1953 {
1954         struct ctl_softc *softc = control_softc;
1955         int i;
1956
1957         if (softc->is_single == 0)
1958                 ctl_frontend_deregister(&ha_frontend);
1959
1960         destroy_dev(softc->dev);
1961
1962         /* Shutdown CTL threads. */
1963         softc->shutdown = 1;
1964         for (i = 0; i < worker_threads; i++) {
1965                 struct ctl_thread *thr = &softc->threads[i];
1966                 while (thr->thread != NULL) {
1967                         wakeup(thr);
1968                         if (thr->thread != NULL)
1969                                 pause("CTL thr shutdown", 1);
1970                 }
1971                 mtx_destroy(&thr->queue_lock);
1972         }
1973         while (softc->lun_thread != NULL) {
1974                 wakeup(&softc->pending_lun_queue);
1975                 if (softc->lun_thread != NULL)
1976                         pause("CTL thr shutdown", 1);
1977         }
1978         while (softc->thresh_thread != NULL) {
1979                 wakeup(softc->thresh_thread);
1980                 if (softc->thresh_thread != NULL)
1981                         pause("CTL thr shutdown", 1);
1982         }
1983
1984         ctl_tpc_shutdown(softc);
1985         uma_zdestroy(softc->io_zone);
1986         mtx_destroy(&softc->ctl_lock);
1987
1988         sysctl_ctx_free(&softc->sysctl_ctx);
1989
1990         free(softc, M_DEVBUF);
1991         control_softc = NULL;
1992         return (0);
1993 }
1994
1995 static int
1996 ctl_module_event_handler(module_t mod, int what, void *arg)
1997 {
1998
1999         switch (what) {
2000         case MOD_LOAD:
2001                 return (ctl_init());
2002         case MOD_UNLOAD:
2003                 return (ctl_shutdown());
2004         default:
2005                 return (EOPNOTSUPP);
2006         }
2007 }
2008
2009 /*
2010  * XXX KDM should we do some access checks here?  Bump a reference count to
2011  * prevent a CTL module from being unloaded while someone has it open?
2012  */
2013 static int
2014 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
2015 {
2016         return (0);
2017 }
2018
2019 static int
2020 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2021 {
2022         return (0);
2023 }
2024
2025 /*
2026  * Remove an initiator by port number and initiator ID.
2027  * Returns 0 for success, -1 for failure.
2028  */
2029 int
2030 ctl_remove_initiator(struct ctl_port *port, int iid)
2031 {
2032         struct ctl_softc *softc = port->ctl_softc;
2033         int last;
2034
2035         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2036
2037         if (iid > CTL_MAX_INIT_PER_PORT) {
2038                 printf("%s: initiator ID %u > maximun %u!\n",
2039                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2040                 return (-1);
2041         }
2042
2043         mtx_lock(&softc->ctl_lock);
2044         last = (--port->wwpn_iid[iid].in_use == 0);
2045         port->wwpn_iid[iid].last_use = time_uptime;
2046         mtx_unlock(&softc->ctl_lock);
2047         if (last)
2048                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
2049         ctl_isc_announce_iid(port, iid);
2050
2051         return (0);
2052 }
2053
2054 /*
2055  * Add an initiator to the initiator map.
2056  * Returns iid for success, < 0 for failure.
2057  */
2058 int
2059 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2060 {
2061         struct ctl_softc *softc = port->ctl_softc;
2062         time_t best_time;
2063         int i, best;
2064
2065         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2066
2067         if (iid >= CTL_MAX_INIT_PER_PORT) {
2068                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2069                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2070                 free(name, M_CTL);
2071                 return (-1);
2072         }
2073
2074         mtx_lock(&softc->ctl_lock);
2075
2076         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2077                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2078                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2079                                 iid = i;
2080                                 break;
2081                         }
2082                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2083                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2084                                 iid = i;
2085                                 break;
2086                         }
2087                 }
2088         }
2089
2090         if (iid < 0) {
2091                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2092                         if (port->wwpn_iid[i].in_use == 0 &&
2093                             port->wwpn_iid[i].wwpn == 0 &&
2094                             port->wwpn_iid[i].name == NULL) {
2095                                 iid = i;
2096                                 break;
2097                         }
2098                 }
2099         }
2100
2101         if (iid < 0) {
2102                 best = -1;
2103                 best_time = INT32_MAX;
2104                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2105                         if (port->wwpn_iid[i].in_use == 0) {
2106                                 if (port->wwpn_iid[i].last_use < best_time) {
2107                                         best = i;
2108                                         best_time = port->wwpn_iid[i].last_use;
2109                                 }
2110                         }
2111                 }
2112                 iid = best;
2113         }
2114
2115         if (iid < 0) {
2116                 mtx_unlock(&softc->ctl_lock);
2117                 free(name, M_CTL);
2118                 return (-2);
2119         }
2120
2121         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2122                 /*
2123                  * This is not an error yet.
2124                  */
2125                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2126 #if 0
2127                         printf("%s: port %d iid %u WWPN %#jx arrived"
2128                             " again\n", __func__, port->targ_port,
2129                             iid, (uintmax_t)wwpn);
2130 #endif
2131                         goto take;
2132                 }
2133                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2134                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2135 #if 0
2136                         printf("%s: port %d iid %u name '%s' arrived"
2137                             " again\n", __func__, port->targ_port,
2138                             iid, name);
2139 #endif
2140                         goto take;
2141                 }
2142
2143                 /*
2144                  * This is an error, but what do we do about it?  The
2145                  * driver is telling us we have a new WWPN for this
2146                  * initiator ID, so we pretty much need to use it.
2147                  */
2148                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2149                     " but WWPN %#jx '%s' is still at that address\n",
2150                     __func__, port->targ_port, iid, wwpn, name,
2151                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2152                     port->wwpn_iid[iid].name);
2153         }
2154 take:
2155         free(port->wwpn_iid[iid].name, M_CTL);
2156         port->wwpn_iid[iid].name = name;
2157         port->wwpn_iid[iid].wwpn = wwpn;
2158         port->wwpn_iid[iid].in_use++;
2159         mtx_unlock(&softc->ctl_lock);
2160         ctl_isc_announce_iid(port, iid);
2161
2162         return (iid);
2163 }
2164
2165 static int
2166 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2167 {
2168         int len;
2169
2170         switch (port->port_type) {
2171         case CTL_PORT_FC:
2172         {
2173                 struct scsi_transportid_fcp *id =
2174                     (struct scsi_transportid_fcp *)buf;
2175                 if (port->wwpn_iid[iid].wwpn == 0)
2176                         return (0);
2177                 memset(id, 0, sizeof(*id));
2178                 id->format_protocol = SCSI_PROTO_FC;
2179                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2180                 return (sizeof(*id));
2181         }
2182         case CTL_PORT_ISCSI:
2183         {
2184                 struct scsi_transportid_iscsi_port *id =
2185                     (struct scsi_transportid_iscsi_port *)buf;
2186                 if (port->wwpn_iid[iid].name == NULL)
2187                         return (0);
2188                 memset(id, 0, 256);
2189                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2190                     SCSI_PROTO_ISCSI;
2191                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2192                 len = roundup2(min(len, 252), 4);
2193                 scsi_ulto2b(len, id->additional_length);
2194                 return (sizeof(*id) + len);
2195         }
2196         case CTL_PORT_SAS:
2197         {
2198                 struct scsi_transportid_sas *id =
2199                     (struct scsi_transportid_sas *)buf;
2200                 if (port->wwpn_iid[iid].wwpn == 0)
2201                         return (0);
2202                 memset(id, 0, sizeof(*id));
2203                 id->format_protocol = SCSI_PROTO_SAS;
2204                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2205                 return (sizeof(*id));
2206         }
2207         default:
2208         {
2209                 struct scsi_transportid_spi *id =
2210                     (struct scsi_transportid_spi *)buf;
2211                 memset(id, 0, sizeof(*id));
2212                 id->format_protocol = SCSI_PROTO_SPI;
2213                 scsi_ulto2b(iid, id->scsi_addr);
2214                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2215                 return (sizeof(*id));
2216         }
2217         }
2218 }
2219
2220 /*
2221  * Serialize a command that went down the "wrong" side, and so was sent to
2222  * this controller for execution.  The logic is a little different than the
2223  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2224  * sent back to the other side, but in the success case, we execute the
2225  * command on this side (XFER mode) or tell the other side to execute it
2226  * (SER_ONLY mode).
2227  */
2228 static void
2229 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2230 {
2231         struct ctl_softc *softc = CTL_SOFTC(ctsio);
2232         struct ctl_port *port = CTL_PORT(ctsio);
2233         union ctl_ha_msg msg_info;
2234         struct ctl_lun *lun;
2235         const struct ctl_cmd_entry *entry;
2236         uint32_t targ_lun;
2237
2238         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2239
2240         /* Make sure that we know about this port. */
2241         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2242                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2243                                          /*retry_count*/ 1);
2244                 goto badjuju;
2245         }
2246
2247         /* Make sure that we know about this LUN. */
2248         mtx_lock(&softc->ctl_lock);
2249         if (targ_lun >= CTL_MAX_LUNS ||
2250             (lun = softc->ctl_luns[targ_lun]) == NULL) {
2251                 mtx_unlock(&softc->ctl_lock);
2252
2253                 /*
2254                  * The other node would not send this request to us unless
2255                  * received announce that we are primary node for this LUN.
2256                  * If this LUN does not exist now, it is probably result of
2257                  * a race, so respond to initiator in the most opaque way.
2258                  */
2259                 ctl_set_busy(ctsio);
2260                 goto badjuju;
2261         }
2262         mtx_lock(&lun->lun_lock);
2263         mtx_unlock(&softc->ctl_lock);
2264
2265         /*
2266          * If the LUN is invalid, pretend that it doesn't exist.
2267          * It will go away as soon as all pending I/Os completed.
2268          */
2269         if (lun->flags & CTL_LUN_DISABLED) {
2270                 mtx_unlock(&lun->lun_lock);
2271                 ctl_set_busy(ctsio);
2272                 goto badjuju;
2273         }
2274
2275         entry = ctl_get_cmd_entry(ctsio, NULL);
2276         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2277                 mtx_unlock(&lun->lun_lock);
2278                 goto badjuju;
2279         }
2280
2281         CTL_LUN(ctsio) = lun;
2282         CTL_BACKEND_LUN(ctsio) = lun->be_lun;
2283
2284         /*
2285          * Every I/O goes into the OOA queue for a
2286          * particular LUN, and stays there until completion.
2287          */
2288 #ifdef CTL_TIME_IO
2289         if (TAILQ_EMPTY(&lun->ooa_queue))
2290                 lun->idle_time += getsbinuptime() - lun->last_busy;
2291 #endif
2292         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2293
2294         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2295                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2296                  ooa_links))) {
2297         case CTL_ACTION_BLOCK:
2298                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2299                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2300                                   blocked_links);
2301                 mtx_unlock(&lun->lun_lock);
2302                 break;
2303         case CTL_ACTION_PASS:
2304         case CTL_ACTION_SKIP:
2305                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2306                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2307                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2308                         mtx_unlock(&lun->lun_lock);
2309                 } else {
2310                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2311                         mtx_unlock(&lun->lun_lock);
2312
2313                         /* send msg back to other side */
2314                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2315                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2316                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2317                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2318                             sizeof(msg_info.hdr), M_WAITOK);
2319                 }
2320                 break;
2321         case CTL_ACTION_OVERLAP:
2322                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2323                 mtx_unlock(&lun->lun_lock);
2324                 ctl_set_overlapped_cmd(ctsio);
2325                 goto badjuju;
2326         case CTL_ACTION_OVERLAP_TAG:
2327                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2328                 mtx_unlock(&lun->lun_lock);
2329                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2330                 goto badjuju;
2331         case CTL_ACTION_ERROR:
2332         default:
2333                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2334                 mtx_unlock(&lun->lun_lock);
2335
2336                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2337                                          /*retry_count*/ 0);
2338 badjuju:
2339                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2340                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2341                 msg_info.hdr.serializing_sc = NULL;
2342                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2343                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2344                     sizeof(msg_info.scsi), M_WAITOK);
2345                 ctl_free_io((union ctl_io *)ctsio);
2346                 break;
2347         }
2348 }
2349
2350 /*
2351  * Returns 0 for success, errno for failure.
2352  */
2353 static void
2354 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2355                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2356 {
2357         union ctl_io *io;
2358
2359         mtx_lock(&lun->lun_lock);
2360         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2361              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2362              ooa_links)) {
2363                 struct ctl_ooa_entry *entry;
2364
2365                 /*
2366                  * If we've got more than we can fit, just count the
2367                  * remaining entries.
2368                  */
2369                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2370                         continue;
2371
2372                 entry = &kern_entries[*cur_fill_num];
2373
2374                 entry->tag_num = io->scsiio.tag_num;
2375                 entry->lun_num = lun->lun;
2376 #ifdef CTL_TIME_IO
2377                 entry->start_bt = io->io_hdr.start_bt;
2378 #endif
2379                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2380                 entry->cdb_len = io->scsiio.cdb_len;
2381                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2382                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2383
2384                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2385                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2386
2387                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2388                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2389
2390                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2391                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2392
2393                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2394                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2395         }
2396         mtx_unlock(&lun->lun_lock);
2397 }
2398
2399 static void *
2400 ctl_copyin_alloc(void *user_addr, unsigned int len, char *error_str,
2401                  size_t error_str_len)
2402 {
2403         void *kptr;
2404
2405         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2406
2407         if (copyin(user_addr, kptr, len) != 0) {
2408                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2409                          "from user address %p to kernel address %p", len,
2410                          user_addr, kptr);
2411                 free(kptr, M_CTL);
2412                 return (NULL);
2413         }
2414
2415         return (kptr);
2416 }
2417
2418 static void
2419 ctl_free_args(int num_args, struct ctl_be_arg *args)
2420 {
2421         int i;
2422
2423         if (args == NULL)
2424                 return;
2425
2426         for (i = 0; i < num_args; i++) {
2427                 free(args[i].kname, M_CTL);
2428                 free(args[i].kvalue, M_CTL);
2429         }
2430
2431         free(args, M_CTL);
2432 }
2433
2434 static struct ctl_be_arg *
2435 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2436                 char *error_str, size_t error_str_len)
2437 {
2438         struct ctl_be_arg *args;
2439         int i;
2440
2441         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2442                                 error_str, error_str_len);
2443
2444         if (args == NULL)
2445                 goto bailout;
2446
2447         for (i = 0; i < num_args; i++) {
2448                 args[i].kname = NULL;
2449                 args[i].kvalue = NULL;
2450         }
2451
2452         for (i = 0; i < num_args; i++) {
2453                 uint8_t *tmpptr;
2454
2455                 if (args[i].namelen == 0) {
2456                         snprintf(error_str, error_str_len, "Argument %d "
2457                                  "name length is zero", i);
2458                         goto bailout;
2459                 }
2460
2461                 args[i].kname = ctl_copyin_alloc(args[i].name,
2462                         args[i].namelen, error_str, error_str_len);
2463                 if (args[i].kname == NULL)
2464                         goto bailout;
2465
2466                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2467                         snprintf(error_str, error_str_len, "Argument %d "
2468                                  "name is not NUL-terminated", i);
2469                         goto bailout;
2470                 }
2471
2472                 if (args[i].flags & CTL_BEARG_RD) {
2473                         if (args[i].vallen == 0) {
2474                                 snprintf(error_str, error_str_len, "Argument %d "
2475                                          "value length is zero", i);
2476                                 goto bailout;
2477                         }
2478
2479                         tmpptr = ctl_copyin_alloc(args[i].value,
2480                                 args[i].vallen, error_str, error_str_len);
2481                         if (tmpptr == NULL)
2482                                 goto bailout;
2483
2484                         if ((args[i].flags & CTL_BEARG_ASCII)
2485                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2486                                 snprintf(error_str, error_str_len, "Argument "
2487                                     "%d value is not NUL-terminated", i);
2488                                 free(tmpptr, M_CTL);
2489                                 goto bailout;
2490                         }
2491                         args[i].kvalue = tmpptr;
2492                 } else {
2493                         args[i].kvalue = malloc(args[i].vallen,
2494                             M_CTL, M_WAITOK | M_ZERO);
2495                 }
2496         }
2497
2498         return (args);
2499 bailout:
2500
2501         ctl_free_args(num_args, args);
2502
2503         return (NULL);
2504 }
2505
2506 static void
2507 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2508 {
2509         int i;
2510
2511         for (i = 0; i < num_args; i++) {
2512                 if (args[i].flags & CTL_BEARG_WR)
2513                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2514         }
2515 }
2516
2517 /*
2518  * Escape characters that are illegal or not recommended in XML.
2519  */
2520 int
2521 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2522 {
2523         char *end = str + size;
2524         int retval;
2525
2526         retval = 0;
2527
2528         for (; *str && str < end; str++) {
2529                 switch (*str) {
2530                 case '&':
2531                         retval = sbuf_printf(sb, "&amp;");
2532                         break;
2533                 case '>':
2534                         retval = sbuf_printf(sb, "&gt;");
2535                         break;
2536                 case '<':
2537                         retval = sbuf_printf(sb, "&lt;");
2538                         break;
2539                 default:
2540                         retval = sbuf_putc(sb, *str);
2541                         break;
2542                 }
2543
2544                 if (retval != 0)
2545                         break;
2546
2547         }
2548
2549         return (retval);
2550 }
2551
2552 static void
2553 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2554 {
2555         struct scsi_vpd_id_descriptor *desc;
2556         int i;
2557
2558         if (id == NULL || id->len < 4)
2559                 return;
2560         desc = (struct scsi_vpd_id_descriptor *)id->data;
2561         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2562         case SVPD_ID_TYPE_T10:
2563                 sbuf_printf(sb, "t10.");
2564                 break;
2565         case SVPD_ID_TYPE_EUI64:
2566                 sbuf_printf(sb, "eui.");
2567                 break;
2568         case SVPD_ID_TYPE_NAA:
2569                 sbuf_printf(sb, "naa.");
2570                 break;
2571         case SVPD_ID_TYPE_SCSI_NAME:
2572                 break;
2573         }
2574         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2575         case SVPD_ID_CODESET_BINARY:
2576                 for (i = 0; i < desc->length; i++)
2577                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2578                 break;
2579         case SVPD_ID_CODESET_ASCII:
2580                 sbuf_printf(sb, "%.*s", (int)desc->length,
2581                     (char *)desc->identifier);
2582                 break;
2583         case SVPD_ID_CODESET_UTF8:
2584                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2585                 break;
2586         }
2587 }
2588
2589 static int
2590 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2591           struct thread *td)
2592 {
2593         struct ctl_softc *softc = dev->si_drv1;
2594         struct ctl_port *port;
2595         struct ctl_lun *lun;
2596         int retval;
2597
2598         retval = 0;
2599
2600         switch (cmd) {
2601         case CTL_IO:
2602                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2603                 break;
2604         case CTL_ENABLE_PORT:
2605         case CTL_DISABLE_PORT:
2606         case CTL_SET_PORT_WWNS: {
2607                 struct ctl_port *port;
2608                 struct ctl_port_entry *entry;
2609
2610                 entry = (struct ctl_port_entry *)addr;
2611
2612                 mtx_lock(&softc->ctl_lock);
2613                 STAILQ_FOREACH(port, &softc->port_list, links) {
2614                         int action, done;
2615
2616                         if (port->targ_port < softc->port_min ||
2617                             port->targ_port >= softc->port_max)
2618                                 continue;
2619
2620                         action = 0;
2621                         done = 0;
2622                         if ((entry->port_type == CTL_PORT_NONE)
2623                          && (entry->targ_port == port->targ_port)) {
2624                                 /*
2625                                  * If the user only wants to enable or
2626                                  * disable or set WWNs on a specific port,
2627                                  * do the operation and we're done.
2628                                  */
2629                                 action = 1;
2630                                 done = 1;
2631                         } else if (entry->port_type & port->port_type) {
2632                                 /*
2633                                  * Compare the user's type mask with the
2634                                  * particular frontend type to see if we
2635                                  * have a match.
2636                                  */
2637                                 action = 1;
2638                                 done = 0;
2639
2640                                 /*
2641                                  * Make sure the user isn't trying to set
2642                                  * WWNs on multiple ports at the same time.
2643                                  */
2644                                 if (cmd == CTL_SET_PORT_WWNS) {
2645                                         printf("%s: Can't set WWNs on "
2646                                                "multiple ports\n", __func__);
2647                                         retval = EINVAL;
2648                                         break;
2649                                 }
2650                         }
2651                         if (action == 0)
2652                                 continue;
2653
2654                         /*
2655                          * XXX KDM we have to drop the lock here, because
2656                          * the online/offline operations can potentially
2657                          * block.  We need to reference count the frontends
2658                          * so they can't go away,
2659                          */
2660                         if (cmd == CTL_ENABLE_PORT) {
2661                                 mtx_unlock(&softc->ctl_lock);
2662                                 ctl_port_online(port);
2663                                 mtx_lock(&softc->ctl_lock);
2664                         } else if (cmd == CTL_DISABLE_PORT) {
2665                                 mtx_unlock(&softc->ctl_lock);
2666                                 ctl_port_offline(port);
2667                                 mtx_lock(&softc->ctl_lock);
2668                         } else if (cmd == CTL_SET_PORT_WWNS) {
2669                                 ctl_port_set_wwns(port,
2670                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2671                                     1 : 0, entry->wwnn,
2672                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2673                                     1 : 0, entry->wwpn);
2674                         }
2675                         if (done != 0)
2676                                 break;
2677                 }
2678                 mtx_unlock(&softc->ctl_lock);
2679                 break;
2680         }
2681         case CTL_GET_OOA: {
2682                 struct ctl_ooa *ooa_hdr;
2683                 struct ctl_ooa_entry *entries;
2684                 uint32_t cur_fill_num;
2685
2686                 ooa_hdr = (struct ctl_ooa *)addr;
2687
2688                 if ((ooa_hdr->alloc_len == 0)
2689                  || (ooa_hdr->alloc_num == 0)) {
2690                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2691                                "must be non-zero\n", __func__,
2692                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2693                         retval = EINVAL;
2694                         break;
2695                 }
2696
2697                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2698                     sizeof(struct ctl_ooa_entry))) {
2699                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2700                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2701                                __func__, ooa_hdr->alloc_len,
2702                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2703                         retval = EINVAL;
2704                         break;
2705                 }
2706
2707                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2708                 if (entries == NULL) {
2709                         printf("%s: could not allocate %d bytes for OOA "
2710                                "dump\n", __func__, ooa_hdr->alloc_len);
2711                         retval = ENOMEM;
2712                         break;
2713                 }
2714
2715                 mtx_lock(&softc->ctl_lock);
2716                 if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 &&
2717                     (ooa_hdr->lun_num >= CTL_MAX_LUNS ||
2718                      softc->ctl_luns[ooa_hdr->lun_num] == NULL)) {
2719                         mtx_unlock(&softc->ctl_lock);
2720                         free(entries, M_CTL);
2721                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2722                                __func__, (uintmax_t)ooa_hdr->lun_num);
2723                         retval = EINVAL;
2724                         break;
2725                 }
2726
2727                 cur_fill_num = 0;
2728
2729                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2730                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2731                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2732                                     ooa_hdr, entries);
2733                         }
2734                 } else {
2735                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2736                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2737                             entries);
2738                 }
2739                 mtx_unlock(&softc->ctl_lock);
2740
2741                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2742                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2743                         sizeof(struct ctl_ooa_entry);
2744                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2745                 if (retval != 0) {
2746                         printf("%s: error copying out %d bytes for OOA dump\n",
2747                                __func__, ooa_hdr->fill_len);
2748                 }
2749
2750                 getbinuptime(&ooa_hdr->cur_bt);
2751
2752                 if (cur_fill_num > ooa_hdr->alloc_num) {
2753                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2754                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2755                 } else {
2756                         ooa_hdr->dropped_num = 0;
2757                         ooa_hdr->status = CTL_OOA_OK;
2758                 }
2759
2760                 free(entries, M_CTL);
2761                 break;
2762         }
2763         case CTL_DELAY_IO: {
2764                 struct ctl_io_delay_info *delay_info;
2765
2766                 delay_info = (struct ctl_io_delay_info *)addr;
2767
2768 #ifdef CTL_IO_DELAY
2769                 mtx_lock(&softc->ctl_lock);
2770                 if (delay_info->lun_id >= CTL_MAX_LUNS ||
2771                     (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) {
2772                         mtx_unlock(&softc->ctl_lock);
2773                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2774                         break;
2775                 }
2776                 mtx_lock(&lun->lun_lock);
2777                 mtx_unlock(&softc->ctl_lock);
2778                 delay_info->status = CTL_DELAY_STATUS_OK;
2779                 switch (delay_info->delay_type) {
2780                 case CTL_DELAY_TYPE_CONT:
2781                 case CTL_DELAY_TYPE_ONESHOT:
2782                         break;
2783                 default:
2784                         delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE;
2785                         break;
2786                 }
2787                 switch (delay_info->delay_loc) {
2788                 case CTL_DELAY_LOC_DATAMOVE:
2789                         lun->delay_info.datamove_type = delay_info->delay_type;
2790                         lun->delay_info.datamove_delay = delay_info->delay_secs;
2791                         break;
2792                 case CTL_DELAY_LOC_DONE:
2793                         lun->delay_info.done_type = delay_info->delay_type;
2794                         lun->delay_info.done_delay = delay_info->delay_secs;
2795                         break;
2796                 default:
2797                         delay_info->status = CTL_DELAY_STATUS_INVALID_LOC;
2798                         break;
2799                 }
2800                 mtx_unlock(&lun->lun_lock);
2801 #else
2802                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2803 #endif /* CTL_IO_DELAY */
2804                 break;
2805         }
2806 #ifdef CTL_LEGACY_STATS
2807         case CTL_GETSTATS: {
2808                 struct ctl_stats *stats = (struct ctl_stats *)addr;
2809                 int i;
2810
2811                 /*
2812                  * XXX KDM no locking here.  If the LUN list changes,
2813                  * things can blow up.
2814                  */
2815                 i = 0;
2816                 stats->status = CTL_SS_OK;
2817                 stats->fill_len = 0;
2818                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2819                         if (stats->fill_len + sizeof(lun->legacy_stats) >
2820                             stats->alloc_len) {
2821                                 stats->status = CTL_SS_NEED_MORE_SPACE;
2822                                 break;
2823                         }
2824                         retval = copyout(&lun->legacy_stats, &stats->lun_stats[i++],
2825                                          sizeof(lun->legacy_stats));
2826                         if (retval != 0)
2827                                 break;
2828                         stats->fill_len += sizeof(lun->legacy_stats);
2829                 }
2830                 stats->num_luns = softc->num_luns;
2831                 stats->flags = CTL_STATS_FLAG_NONE;
2832 #ifdef CTL_TIME_IO
2833                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
2834 #endif
2835                 getnanouptime(&stats->timestamp);
2836                 break;
2837         }
2838 #endif /* CTL_LEGACY_STATS */
2839         case CTL_ERROR_INJECT: {
2840                 struct ctl_error_desc *err_desc, *new_err_desc;
2841
2842                 err_desc = (struct ctl_error_desc *)addr;
2843
2844                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2845                                       M_WAITOK | M_ZERO);
2846                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2847
2848                 mtx_lock(&softc->ctl_lock);
2849                 if (err_desc->lun_id >= CTL_MAX_LUNS ||
2850                     (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) {
2851                         mtx_unlock(&softc->ctl_lock);
2852                         free(new_err_desc, M_CTL);
2853                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2854                                __func__, (uintmax_t)err_desc->lun_id);
2855                         retval = EINVAL;
2856                         break;
2857                 }
2858                 mtx_lock(&lun->lun_lock);
2859                 mtx_unlock(&softc->ctl_lock);
2860
2861                 /*
2862                  * We could do some checking here to verify the validity
2863                  * of the request, but given the complexity of error
2864                  * injection requests, the checking logic would be fairly
2865                  * complex.
2866                  *
2867                  * For now, if the request is invalid, it just won't get
2868                  * executed and might get deleted.
2869                  */
2870                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2871
2872                 /*
2873                  * XXX KDM check to make sure the serial number is unique,
2874                  * in case we somehow manage to wrap.  That shouldn't
2875                  * happen for a very long time, but it's the right thing to
2876                  * do.
2877                  */
2878                 new_err_desc->serial = lun->error_serial;
2879                 err_desc->serial = lun->error_serial;
2880                 lun->error_serial++;
2881
2882                 mtx_unlock(&lun->lun_lock);
2883                 break;
2884         }
2885         case CTL_ERROR_INJECT_DELETE: {
2886                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2887                 int delete_done;
2888
2889                 delete_desc = (struct ctl_error_desc *)addr;
2890                 delete_done = 0;
2891
2892                 mtx_lock(&softc->ctl_lock);
2893                 if (delete_desc->lun_id >= CTL_MAX_LUNS ||
2894                     (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) {
2895                         mtx_unlock(&softc->ctl_lock);
2896                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2897                                __func__, (uintmax_t)delete_desc->lun_id);
2898                         retval = EINVAL;
2899                         break;
2900                 }
2901                 mtx_lock(&lun->lun_lock);
2902                 mtx_unlock(&softc->ctl_lock);
2903                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2904                         if (desc->serial != delete_desc->serial)
2905                                 continue;
2906
2907                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2908                                       links);
2909                         free(desc, M_CTL);
2910                         delete_done = 1;
2911                 }
2912                 mtx_unlock(&lun->lun_lock);
2913                 if (delete_done == 0) {
2914                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2915                                "error serial %ju on LUN %u\n", __func__,
2916                                delete_desc->serial, delete_desc->lun_id);
2917                         retval = EINVAL;
2918                         break;
2919                 }
2920                 break;
2921         }
2922         case CTL_DUMP_STRUCTS: {
2923                 int j, k;
2924                 struct ctl_port *port;
2925                 struct ctl_frontend *fe;
2926
2927                 mtx_lock(&softc->ctl_lock);
2928                 printf("CTL Persistent Reservation information start:\n");
2929                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2930                         mtx_lock(&lun->lun_lock);
2931                         if ((lun->flags & CTL_LUN_DISABLED) != 0) {
2932                                 mtx_unlock(&lun->lun_lock);
2933                                 continue;
2934                         }
2935
2936                         for (j = 0; j < CTL_MAX_PORTS; j++) {
2937                                 if (lun->pr_keys[j] == NULL)
2938                                         continue;
2939                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2940                                         if (lun->pr_keys[j][k] == 0)
2941                                                 continue;
2942                                         printf("  LUN %ju port %d iid %d key "
2943                                                "%#jx\n", lun->lun, j, k,
2944                                                (uintmax_t)lun->pr_keys[j][k]);
2945                                 }
2946                         }
2947                         mtx_unlock(&lun->lun_lock);
2948                 }
2949                 printf("CTL Persistent Reservation information end\n");
2950                 printf("CTL Ports:\n");
2951                 STAILQ_FOREACH(port, &softc->port_list, links) {
2952                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2953                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2954                                port->frontend->name, port->port_type,
2955                                port->physical_port, port->virtual_port,
2956                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2957                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2958                                 if (port->wwpn_iid[j].in_use == 0 &&
2959                                     port->wwpn_iid[j].wwpn == 0 &&
2960                                     port->wwpn_iid[j].name == NULL)
2961                                         continue;
2962
2963                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2964                                     j, port->wwpn_iid[j].in_use,
2965                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2966                                     port->wwpn_iid[j].name);
2967                         }
2968                 }
2969                 printf("CTL Port information end\n");
2970                 mtx_unlock(&softc->ctl_lock);
2971                 /*
2972                  * XXX KDM calling this without a lock.  We'd likely want
2973                  * to drop the lock before calling the frontend's dump
2974                  * routine anyway.
2975                  */
2976                 printf("CTL Frontends:\n");
2977                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2978                         printf("  Frontend '%s'\n", fe->name);
2979                         if (fe->fe_dump != NULL)
2980                                 fe->fe_dump();
2981                 }
2982                 printf("CTL Frontend information end\n");
2983                 break;
2984         }
2985         case CTL_LUN_REQ: {
2986                 struct ctl_lun_req *lun_req;
2987                 struct ctl_backend_driver *backend;
2988
2989                 lun_req = (struct ctl_lun_req *)addr;
2990
2991                 backend = ctl_backend_find(lun_req->backend);
2992                 if (backend == NULL) {
2993                         lun_req->status = CTL_LUN_ERROR;
2994                         snprintf(lun_req->error_str,
2995                                  sizeof(lun_req->error_str),
2996                                  "Backend \"%s\" not found.",
2997                                  lun_req->backend);
2998                         break;
2999                 }
3000                 if (lun_req->num_be_args > 0) {
3001                         lun_req->kern_be_args = ctl_copyin_args(
3002                                 lun_req->num_be_args,
3003                                 lun_req->be_args,
3004                                 lun_req->error_str,
3005                                 sizeof(lun_req->error_str));
3006                         if (lun_req->kern_be_args == NULL) {
3007                                 lun_req->status = CTL_LUN_ERROR;
3008                                 break;
3009                         }
3010                 }
3011
3012                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3013
3014                 if (lun_req->num_be_args > 0) {
3015                         ctl_copyout_args(lun_req->num_be_args,
3016                                       lun_req->kern_be_args);
3017                         ctl_free_args(lun_req->num_be_args,
3018                                       lun_req->kern_be_args);
3019                 }
3020                 break;
3021         }
3022         case CTL_LUN_LIST: {
3023                 struct sbuf *sb;
3024                 struct ctl_lun_list *list;
3025                 struct ctl_option *opt;
3026
3027                 list = (struct ctl_lun_list *)addr;
3028
3029                 /*
3030                  * Allocate a fixed length sbuf here, based on the length
3031                  * of the user's buffer.  We could allocate an auto-extending
3032                  * buffer, and then tell the user how much larger our
3033                  * amount of data is than his buffer, but that presents
3034                  * some problems:
3035                  *
3036                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3037                  *     we can't hold a lock while calling them with an
3038                  *     auto-extending buffer.
3039                  *
3040                  * 2.  There is not currently a LUN reference counting
3041                  *     mechanism, outside of outstanding transactions on
3042                  *     the LUN's OOA queue.  So a LUN could go away on us
3043                  *     while we're getting the LUN number, backend-specific
3044                  *     information, etc.  Thus, given the way things
3045                  *     currently work, we need to hold the CTL lock while
3046                  *     grabbing LUN information.
3047                  *
3048                  * So, from the user's standpoint, the best thing to do is
3049                  * allocate what he thinks is a reasonable buffer length,
3050                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3051                  * double the buffer length and try again.  (And repeat
3052                  * that until he succeeds.)
3053                  */
3054                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3055                 if (sb == NULL) {
3056                         list->status = CTL_LUN_LIST_ERROR;
3057                         snprintf(list->error_str, sizeof(list->error_str),
3058                                  "Unable to allocate %d bytes for LUN list",
3059                                  list->alloc_len);
3060                         break;
3061                 }
3062
3063                 sbuf_printf(sb, "<ctllunlist>\n");
3064
3065                 mtx_lock(&softc->ctl_lock);
3066                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3067                         mtx_lock(&lun->lun_lock);
3068                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3069                                              (uintmax_t)lun->lun);
3070
3071                         /*
3072                          * Bail out as soon as we see that we've overfilled
3073                          * the buffer.
3074                          */
3075                         if (retval != 0)
3076                                 break;
3077
3078                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3079                                              "</backend_type>\n",
3080                                              (lun->backend == NULL) ?  "none" :
3081                                              lun->backend->name);
3082
3083                         if (retval != 0)
3084                                 break;
3085
3086                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3087                                              lun->be_lun->lun_type);
3088
3089                         if (retval != 0)
3090                                 break;
3091
3092                         if (lun->backend == NULL) {
3093                                 retval = sbuf_printf(sb, "</lun>\n");
3094                                 if (retval != 0)
3095                                         break;
3096                                 continue;
3097                         }
3098
3099                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3100                                              (lun->be_lun->maxlba > 0) ?
3101                                              lun->be_lun->maxlba + 1 : 0);
3102
3103                         if (retval != 0)
3104                                 break;
3105
3106                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3107                                              lun->be_lun->blocksize);
3108
3109                         if (retval != 0)
3110                                 break;
3111
3112                         retval = sbuf_printf(sb, "\t<serial_number>");
3113
3114                         if (retval != 0)
3115                                 break;
3116
3117                         retval = ctl_sbuf_printf_esc(sb,
3118                             lun->be_lun->serial_num,
3119                             sizeof(lun->be_lun->serial_num));
3120
3121                         if (retval != 0)
3122                                 break;
3123
3124                         retval = sbuf_printf(sb, "</serial_number>\n");
3125
3126                         if (retval != 0)
3127                                 break;
3128
3129                         retval = sbuf_printf(sb, "\t<device_id>");
3130
3131                         if (retval != 0)
3132                                 break;
3133
3134                         retval = ctl_sbuf_printf_esc(sb,
3135                             lun->be_lun->device_id,
3136                             sizeof(lun->be_lun->device_id));
3137
3138                         if (retval != 0)
3139                                 break;
3140
3141                         retval = sbuf_printf(sb, "</device_id>\n");
3142
3143                         if (retval != 0)
3144                                 break;
3145
3146                         if (lun->backend->lun_info != NULL) {
3147                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3148                                 if (retval != 0)
3149                                         break;
3150                         }
3151                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3152                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3153                                     opt->name, opt->value, opt->name);
3154                                 if (retval != 0)
3155                                         break;
3156                         }
3157
3158                         retval = sbuf_printf(sb, "</lun>\n");
3159
3160                         if (retval != 0)
3161                                 break;
3162                         mtx_unlock(&lun->lun_lock);
3163                 }
3164                 if (lun != NULL)
3165                         mtx_unlock(&lun->lun_lock);
3166                 mtx_unlock(&softc->ctl_lock);
3167
3168                 if ((retval != 0)
3169                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3170                         retval = 0;
3171                         sbuf_delete(sb);
3172                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3173                         snprintf(list->error_str, sizeof(list->error_str),
3174                                  "Out of space, %d bytes is too small",
3175                                  list->alloc_len);
3176                         break;
3177                 }
3178
3179                 sbuf_finish(sb);
3180
3181                 retval = copyout(sbuf_data(sb), list->lun_xml,
3182                                  sbuf_len(sb) + 1);
3183
3184                 list->fill_len = sbuf_len(sb) + 1;
3185                 list->status = CTL_LUN_LIST_OK;
3186                 sbuf_delete(sb);
3187                 break;
3188         }
3189         case CTL_ISCSI: {
3190                 struct ctl_iscsi *ci;
3191                 struct ctl_frontend *fe;
3192
3193                 ci = (struct ctl_iscsi *)addr;
3194
3195                 fe = ctl_frontend_find("iscsi");
3196                 if (fe == NULL) {
3197                         ci->status = CTL_ISCSI_ERROR;
3198                         snprintf(ci->error_str, sizeof(ci->error_str),
3199                             "Frontend \"iscsi\" not found.");
3200                         break;
3201                 }
3202
3203                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3204                 break;
3205         }
3206         case CTL_PORT_REQ: {
3207                 struct ctl_req *req;
3208                 struct ctl_frontend *fe;
3209
3210                 req = (struct ctl_req *)addr;
3211
3212                 fe = ctl_frontend_find(req->driver);
3213                 if (fe == NULL) {
3214                         req->status = CTL_LUN_ERROR;
3215                         snprintf(req->error_str, sizeof(req->error_str),
3216                             "Frontend \"%s\" not found.", req->driver);
3217                         break;
3218                 }
3219                 if (req->num_args > 0) {
3220                         req->kern_args = ctl_copyin_args(req->num_args,
3221                             req->args, req->error_str, sizeof(req->error_str));
3222                         if (req->kern_args == NULL) {
3223                                 req->status = CTL_LUN_ERROR;
3224                                 break;
3225                         }
3226                 }
3227
3228                 if (fe->ioctl)
3229                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3230                 else
3231                         retval = ENODEV;
3232
3233                 if (req->num_args > 0) {
3234                         ctl_copyout_args(req->num_args, req->kern_args);
3235                         ctl_free_args(req->num_args, req->kern_args);
3236                 }
3237                 break;
3238         }
3239         case CTL_PORT_LIST: {
3240                 struct sbuf *sb;
3241                 struct ctl_port *port;
3242                 struct ctl_lun_list *list;
3243                 struct ctl_option *opt;
3244                 int j;
3245                 uint32_t plun;
3246
3247                 list = (struct ctl_lun_list *)addr;
3248
3249                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3250                 if (sb == NULL) {
3251                         list->status = CTL_LUN_LIST_ERROR;
3252                         snprintf(list->error_str, sizeof(list->error_str),
3253                                  "Unable to allocate %d bytes for LUN list",
3254                                  list->alloc_len);
3255                         break;
3256                 }
3257
3258                 sbuf_printf(sb, "<ctlportlist>\n");
3259
3260                 mtx_lock(&softc->ctl_lock);
3261                 STAILQ_FOREACH(port, &softc->port_list, links) {
3262                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3263                                              (uintmax_t)port->targ_port);
3264
3265                         /*
3266                          * Bail out as soon as we see that we've overfilled
3267                          * the buffer.
3268                          */
3269                         if (retval != 0)
3270                                 break;
3271
3272                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3273                             "</frontend_type>\n", port->frontend->name);
3274                         if (retval != 0)
3275                                 break;
3276
3277                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3278                                              port->port_type);
3279                         if (retval != 0)
3280                                 break;
3281
3282                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3283                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3284                         if (retval != 0)
3285                                 break;
3286
3287                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3288                             port->port_name);
3289                         if (retval != 0)
3290                                 break;
3291
3292                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3293                             port->physical_port);
3294                         if (retval != 0)
3295                                 break;
3296
3297                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3298                             port->virtual_port);
3299                         if (retval != 0)
3300                                 break;
3301
3302                         if (port->target_devid != NULL) {
3303                                 sbuf_printf(sb, "\t<target>");
3304                                 ctl_id_sbuf(port->target_devid, sb);
3305                                 sbuf_printf(sb, "</target>\n");
3306                         }
3307
3308                         if (port->port_devid != NULL) {
3309                                 sbuf_printf(sb, "\t<port>");
3310                                 ctl_id_sbuf(port->port_devid, sb);
3311                                 sbuf_printf(sb, "</port>\n");
3312                         }
3313
3314                         if (port->port_info != NULL) {
3315                                 retval = port->port_info(port->onoff_arg, sb);
3316                                 if (retval != 0)
3317                                         break;
3318                         }
3319                         STAILQ_FOREACH(opt, &port->options, links) {
3320                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3321                                     opt->name, opt->value, opt->name);
3322                                 if (retval != 0)
3323                                         break;
3324                         }
3325
3326                         if (port->lun_map != NULL) {
3327                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3328                                 for (j = 0; j < port->lun_map_size; j++) {
3329                                         plun = ctl_lun_map_from_port(port, j);
3330                                         if (plun == UINT32_MAX)
3331                                                 continue;
3332                                         sbuf_printf(sb,
3333                                             "\t<lun id=\"%u\">%u</lun>\n",
3334                                             j, plun);
3335                                 }
3336                         }
3337
3338                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3339                                 if (port->wwpn_iid[j].in_use == 0 ||
3340                                     (port->wwpn_iid[j].wwpn == 0 &&
3341                                      port->wwpn_iid[j].name == NULL))
3342                                         continue;
3343
3344                                 if (port->wwpn_iid[j].name != NULL)
3345                                         retval = sbuf_printf(sb,
3346                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3347                                             j, port->wwpn_iid[j].name);
3348                                 else
3349                                         retval = sbuf_printf(sb,
3350                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3351                                             j, port->wwpn_iid[j].wwpn);
3352                                 if (retval != 0)
3353                                         break;
3354                         }
3355                         if (retval != 0)
3356                                 break;
3357
3358                         retval = sbuf_printf(sb, "</targ_port>\n");
3359                         if (retval != 0)
3360                                 break;
3361                 }
3362                 mtx_unlock(&softc->ctl_lock);
3363
3364                 if ((retval != 0)
3365                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3366                         retval = 0;
3367                         sbuf_delete(sb);
3368                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3369                         snprintf(list->error_str, sizeof(list->error_str),
3370                                  "Out of space, %d bytes is too small",
3371                                  list->alloc_len);
3372                         break;
3373                 }
3374
3375                 sbuf_finish(sb);
3376
3377                 retval = copyout(sbuf_data(sb), list->lun_xml,
3378                                  sbuf_len(sb) + 1);
3379
3380                 list->fill_len = sbuf_len(sb) + 1;
3381                 list->status = CTL_LUN_LIST_OK;
3382                 sbuf_delete(sb);
3383                 break;
3384         }
3385         case CTL_LUN_MAP: {
3386                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3387                 struct ctl_port *port;
3388
3389                 mtx_lock(&softc->ctl_lock);
3390                 if (lm->port < softc->port_min ||
3391                     lm->port >= softc->port_max ||
3392                     (port = softc->ctl_ports[lm->port]) == NULL) {
3393                         mtx_unlock(&softc->ctl_lock);
3394                         return (ENXIO);
3395                 }
3396                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3397                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3398                                 if (ctl_lun_map_to_port(port, lun->lun) ==
3399                                     UINT32_MAX)
3400                                         continue;
3401                                 mtx_lock(&lun->lun_lock);
3402                                 ctl_est_ua_port(lun, lm->port, -1,
3403                                     CTL_UA_LUN_CHANGE);
3404                                 mtx_unlock(&lun->lun_lock);
3405                         }
3406                 }
3407                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3408                 if (lm->plun != UINT32_MAX) {
3409                         if (lm->lun == UINT32_MAX)
3410                                 retval = ctl_lun_map_unset(port, lm->plun);
3411                         else if (lm->lun < CTL_MAX_LUNS &&
3412                             softc->ctl_luns[lm->lun] != NULL)
3413                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3414                         else
3415                                 return (ENXIO);
3416                 } else {
3417                         if (lm->lun == UINT32_MAX)
3418                                 retval = ctl_lun_map_deinit(port);
3419                         else
3420                                 retval = ctl_lun_map_init(port);
3421                 }
3422                 if (port->status & CTL_PORT_STATUS_ONLINE)
3423                         ctl_isc_announce_port(port);
3424                 break;
3425         }
3426         case CTL_GET_LUN_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(lun, &softc->lun_list, links) {
3438                         if (lun->lun < stats->first_item)
3439                                 continue;
3440                         if (stats->fill_len + sizeof(lun->stats) >
3441                             stats->alloc_len) {
3442                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3443                                 break;
3444                         }
3445                         retval = copyout(&lun->stats, &stats->stats[i++],
3446                                          sizeof(lun->stats));
3447                         if (retval != 0)
3448                                 break;
3449                         stats->fill_len += sizeof(lun->stats);
3450                 }
3451                 stats->num_items = softc->num_luns;
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         case CTL_GET_PORT_STATS: {
3460                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3461                 int i;
3462
3463                 /*
3464                  * XXX KDM no locking here.  If the LUN list changes,
3465                  * things can blow up.
3466                  */
3467                 i = 0;
3468                 stats->status = CTL_SS_OK;
3469                 stats->fill_len = 0;
3470                 STAILQ_FOREACH(port, &softc->port_list, links) {
3471                         if (port->targ_port < stats->first_item)
3472                                 continue;
3473                         if (stats->fill_len + sizeof(port->stats) >
3474                             stats->alloc_len) {
3475                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3476                                 break;
3477                         }
3478                         retval = copyout(&port->stats, &stats->stats[i++],
3479                                          sizeof(port->stats));
3480                         if (retval != 0)
3481                                 break;
3482                         stats->fill_len += sizeof(port->stats);
3483                 }
3484                 stats->num_items = softc->num_ports;
3485                 stats->flags = CTL_STATS_FLAG_NONE;
3486 #ifdef CTL_TIME_IO
3487                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3488 #endif
3489                 getnanouptime(&stats->timestamp);
3490                 break;
3491         }
3492         default: {
3493                 /* XXX KDM should we fix this? */
3494 #if 0
3495                 struct ctl_backend_driver *backend;
3496                 unsigned int type;
3497                 int found;
3498
3499                 found = 0;
3500
3501                 /*
3502                  * We encode the backend type as the ioctl type for backend
3503                  * ioctls.  So parse it out here, and then search for a
3504                  * backend of this type.
3505                  */
3506                 type = _IOC_TYPE(cmd);
3507
3508                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3509                         if (backend->type == type) {
3510                                 found = 1;
3511                                 break;
3512                         }
3513                 }
3514                 if (found == 0) {
3515                         printf("ctl: unknown ioctl command %#lx or backend "
3516                                "%d\n", cmd, type);
3517                         retval = EINVAL;
3518                         break;
3519                 }
3520                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3521 #endif
3522                 retval = ENOTTY;
3523                 break;
3524         }
3525         }
3526         return (retval);
3527 }
3528
3529 uint32_t
3530 ctl_get_initindex(struct ctl_nexus *nexus)
3531 {
3532         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3533 }
3534
3535 int
3536 ctl_lun_map_init(struct ctl_port *port)
3537 {
3538         struct ctl_softc *softc = port->ctl_softc;
3539         struct ctl_lun *lun;
3540         int size = ctl_lun_map_size;
3541         uint32_t i;
3542
3543         if (port->lun_map == NULL || port->lun_map_size < size) {
3544                 port->lun_map_size = 0;
3545                 free(port->lun_map, M_CTL);
3546                 port->lun_map = malloc(size * sizeof(uint32_t),
3547                     M_CTL, M_NOWAIT);
3548         }
3549         if (port->lun_map == NULL)
3550                 return (ENOMEM);
3551         for (i = 0; i < size; i++)
3552                 port->lun_map[i] = UINT32_MAX;
3553         port->lun_map_size = size;
3554         if (port->status & CTL_PORT_STATUS_ONLINE) {
3555                 if (port->lun_disable != NULL) {
3556                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3557                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3558                 }
3559                 ctl_isc_announce_port(port);
3560         }
3561         return (0);
3562 }
3563
3564 int
3565 ctl_lun_map_deinit(struct ctl_port *port)
3566 {
3567         struct ctl_softc *softc = port->ctl_softc;
3568         struct ctl_lun *lun;
3569
3570         if (port->lun_map == NULL)
3571                 return (0);
3572         port->lun_map_size = 0;
3573         free(port->lun_map, M_CTL);
3574         port->lun_map = NULL;
3575         if (port->status & CTL_PORT_STATUS_ONLINE) {
3576                 if (port->lun_enable != NULL) {
3577                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3578                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3579                 }
3580                 ctl_isc_announce_port(port);
3581         }
3582         return (0);
3583 }
3584
3585 int
3586 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3587 {
3588         int status;
3589         uint32_t old;
3590
3591         if (port->lun_map == NULL) {
3592                 status = ctl_lun_map_init(port);
3593                 if (status != 0)
3594                         return (status);
3595         }
3596         if (plun >= port->lun_map_size)
3597                 return (EINVAL);
3598         old = port->lun_map[plun];
3599         port->lun_map[plun] = glun;
3600         if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) {
3601                 if (port->lun_enable != NULL)
3602                         port->lun_enable(port->targ_lun_arg, plun);
3603                 ctl_isc_announce_port(port);
3604         }
3605         return (0);
3606 }
3607
3608 int
3609 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3610 {
3611         uint32_t old;
3612
3613         if (port->lun_map == NULL || plun >= port->lun_map_size)
3614                 return (0);
3615         old = port->lun_map[plun];
3616         port->lun_map[plun] = UINT32_MAX;
3617         if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) {
3618                 if (port->lun_disable != NULL)
3619                         port->lun_disable(port->targ_lun_arg, plun);
3620                 ctl_isc_announce_port(port);
3621         }
3622         return (0);
3623 }
3624
3625 uint32_t
3626 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3627 {
3628
3629         if (port == NULL)
3630                 return (UINT32_MAX);
3631         if (port->lun_map == NULL)
3632                 return (lun_id);
3633         if (lun_id > port->lun_map_size)
3634                 return (UINT32_MAX);
3635         return (port->lun_map[lun_id]);
3636 }
3637
3638 uint32_t
3639 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3640 {
3641         uint32_t i;
3642
3643         if (port == NULL)
3644                 return (UINT32_MAX);
3645         if (port->lun_map == NULL)
3646                 return (lun_id);
3647         for (i = 0; i < port->lun_map_size; i++) {
3648                 if (port->lun_map[i] == lun_id)
3649                         return (i);
3650         }
3651         return (UINT32_MAX);
3652 }
3653
3654 uint32_t
3655 ctl_decode_lun(uint64_t encoded)
3656 {
3657         uint8_t lun[8];
3658         uint32_t result = 0xffffffff;
3659
3660         be64enc(lun, encoded);
3661         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3662         case RPL_LUNDATA_ATYP_PERIPH:
3663                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3664                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3665                         result = lun[1];
3666                 break;
3667         case RPL_LUNDATA_ATYP_FLAT:
3668                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3669                     lun[6] == 0 && lun[7] == 0)
3670                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3671                 break;
3672         case RPL_LUNDATA_ATYP_EXTLUN:
3673                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3674                 case 0x02:
3675                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3676                         case 0x00:
3677                                 result = lun[1];
3678                                 break;
3679                         case 0x10:
3680                                 result = (lun[1] << 16) + (lun[2] << 8) +
3681                                     lun[3];
3682                                 break;
3683                         case 0x20:
3684                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3685                                         result = (lun[2] << 24) +
3686                                             (lun[3] << 16) + (lun[4] << 8) +
3687                                             lun[5];
3688                                 break;
3689                         }
3690                         break;
3691                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3692                         result = 0xffffffff;
3693                         break;
3694                 }
3695                 break;
3696         }
3697         return (result);
3698 }
3699
3700 uint64_t
3701 ctl_encode_lun(uint32_t decoded)
3702 {
3703         uint64_t l = decoded;
3704
3705         if (l <= 0xff)
3706                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3707         if (l <= 0x3fff)
3708                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3709         if (l <= 0xffffff)
3710                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3711                     (l << 32));
3712         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3713 }
3714
3715 int
3716 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3717 {
3718         int i;
3719
3720         for (i = first; i < last; i++) {
3721                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3722                         return (i);
3723         }
3724         return (-1);
3725 }
3726
3727 int
3728 ctl_set_mask(uint32_t *mask, uint32_t bit)
3729 {
3730         uint32_t chunk, piece;
3731
3732         chunk = bit >> 5;
3733         piece = bit % (sizeof(uint32_t) * 8);
3734
3735         if ((mask[chunk] & (1 << piece)) != 0)
3736                 return (-1);
3737         else
3738                 mask[chunk] |= (1 << piece);
3739
3740         return (0);
3741 }
3742
3743 int
3744 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3745 {
3746         uint32_t chunk, piece;
3747
3748         chunk = bit >> 5;
3749         piece = bit % (sizeof(uint32_t) * 8);
3750
3751         if ((mask[chunk] & (1 << piece)) == 0)
3752                 return (-1);
3753         else
3754                 mask[chunk] &= ~(1 << piece);
3755
3756         return (0);
3757 }
3758
3759 int
3760 ctl_is_set(uint32_t *mask, uint32_t bit)
3761 {
3762         uint32_t chunk, piece;
3763
3764         chunk = bit >> 5;
3765         piece = bit % (sizeof(uint32_t) * 8);
3766
3767         if ((mask[chunk] & (1 << piece)) == 0)
3768                 return (0);
3769         else
3770                 return (1);
3771 }
3772
3773 static uint64_t
3774 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3775 {
3776         uint64_t *t;
3777
3778         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3779         if (t == NULL)
3780                 return (0);
3781         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3782 }
3783
3784 static void
3785 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3786 {
3787         uint64_t *t;
3788
3789         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3790         if (t == NULL)
3791                 return;
3792         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3793 }
3794
3795 static void
3796 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3797 {
3798         uint64_t *p;
3799         u_int i;
3800
3801         i = residx/CTL_MAX_INIT_PER_PORT;
3802         if (lun->pr_keys[i] != NULL)
3803                 return;
3804         mtx_unlock(&lun->lun_lock);
3805         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3806             M_WAITOK | M_ZERO);
3807         mtx_lock(&lun->lun_lock);
3808         if (lun->pr_keys[i] == NULL)
3809                 lun->pr_keys[i] = p;
3810         else
3811                 free(p, M_CTL);
3812 }
3813
3814 static void
3815 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3816 {
3817         uint64_t *t;
3818
3819         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3820         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3821         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3822 }
3823
3824 /*
3825  * ctl_softc, pool_name, total_ctl_io are passed in.
3826  * npool is passed out.
3827  */
3828 int
3829 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3830                 uint32_t total_ctl_io, void **npool)
3831 {
3832         struct ctl_io_pool *pool;
3833
3834         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3835                                             M_NOWAIT | M_ZERO);
3836         if (pool == NULL)
3837                 return (ENOMEM);
3838
3839         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3840         pool->ctl_softc = ctl_softc;
3841 #ifdef IO_POOLS
3842         pool->zone = uma_zsecond_create(pool->name, NULL,
3843             NULL, NULL, NULL, ctl_softc->io_zone);
3844         /* uma_prealloc(pool->zone, total_ctl_io); */
3845 #else
3846         pool->zone = ctl_softc->io_zone;
3847 #endif
3848
3849         *npool = pool;
3850         return (0);
3851 }
3852
3853 void
3854 ctl_pool_free(struct ctl_io_pool *pool)
3855 {
3856
3857         if (pool == NULL)
3858                 return;
3859
3860 #ifdef IO_POOLS
3861         uma_zdestroy(pool->zone);
3862 #endif
3863         free(pool, M_CTL);
3864 }
3865
3866 union ctl_io *
3867 ctl_alloc_io(void *pool_ref)
3868 {
3869         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3870         union ctl_io *io;
3871
3872         io = uma_zalloc(pool->zone, M_WAITOK);
3873         if (io != NULL) {
3874                 io->io_hdr.pool = pool_ref;
3875                 CTL_SOFTC(io) = pool->ctl_softc;
3876         }
3877         return (io);
3878 }
3879
3880 union ctl_io *
3881 ctl_alloc_io_nowait(void *pool_ref)
3882 {
3883         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3884         union ctl_io *io;
3885
3886         io = uma_zalloc(pool->zone, M_NOWAIT);
3887         if (io != NULL) {
3888                 io->io_hdr.pool = pool_ref;
3889                 CTL_SOFTC(io) = pool->ctl_softc;
3890         }
3891         return (io);
3892 }
3893
3894 void
3895 ctl_free_io(union ctl_io *io)
3896 {
3897         struct ctl_io_pool *pool;
3898
3899         if (io == NULL)
3900                 return;
3901
3902         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3903         uma_zfree(pool->zone, io);
3904 }
3905
3906 void
3907 ctl_zero_io(union ctl_io *io)
3908 {
3909         struct ctl_io_pool *pool;
3910
3911         if (io == NULL)
3912                 return;
3913
3914         /*
3915          * May need to preserve linked list pointers at some point too.
3916          */
3917         pool = io->io_hdr.pool;
3918         memset(io, 0, sizeof(*io));
3919         io->io_hdr.pool = pool;
3920         CTL_SOFTC(io) = pool->ctl_softc;
3921 }
3922
3923 int
3924 ctl_expand_number(const char *buf, uint64_t *num)
3925 {
3926         char *endptr;
3927         uint64_t number;
3928         unsigned shift;
3929
3930         number = strtoq(buf, &endptr, 0);
3931
3932         switch (tolower((unsigned char)*endptr)) {
3933         case 'e':
3934                 shift = 60;
3935                 break;
3936         case 'p':
3937                 shift = 50;
3938                 break;
3939         case 't':
3940                 shift = 40;
3941                 break;
3942         case 'g':
3943                 shift = 30;
3944                 break;
3945         case 'm':
3946                 shift = 20;
3947                 break;
3948         case 'k':
3949                 shift = 10;
3950                 break;
3951         case 'b':
3952         case '\0': /* No unit. */
3953                 *num = number;
3954                 return (0);
3955         default:
3956                 /* Unrecognized unit. */
3957                 return (-1);
3958         }
3959
3960         if ((number << shift) >> shift != number) {
3961                 /* Overflow */
3962                 return (-1);
3963         }
3964         *num = number << shift;
3965         return (0);
3966 }
3967
3968
3969 /*
3970  * This routine could be used in the future to load default and/or saved
3971  * mode page parameters for a particuar lun.
3972  */
3973 static int
3974 ctl_init_page_index(struct ctl_lun *lun)
3975 {
3976         int i, page_code;
3977         struct ctl_page_index *page_index;
3978         const char *value;
3979         uint64_t ival;
3980
3981         memcpy(&lun->mode_pages.index, page_index_template,
3982                sizeof(page_index_template));
3983
3984         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3985
3986                 page_index = &lun->mode_pages.index[i];
3987                 if (lun->be_lun->lun_type == T_DIRECT &&
3988                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
3989                         continue;
3990                 if (lun->be_lun->lun_type == T_PROCESSOR &&
3991                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
3992                         continue;
3993                 if (lun->be_lun->lun_type == T_CDROM &&
3994                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
3995                         continue;
3996
3997                 page_code = page_index->page_code & SMPH_PC_MASK;
3998                 switch (page_code) {
3999                 case SMS_RW_ERROR_RECOVERY_PAGE: {
4000                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4001                             ("subpage %#x for page %#x is incorrect!",
4002                             page_index->subpage, page_code));
4003                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4004                                &rw_er_page_default,
4005                                sizeof(rw_er_page_default));
4006                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4007                                &rw_er_page_changeable,
4008                                sizeof(rw_er_page_changeable));
4009                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4010                                &rw_er_page_default,
4011                                sizeof(rw_er_page_default));
4012                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4013                                &rw_er_page_default,
4014                                sizeof(rw_er_page_default));
4015                         page_index->page_data =
4016                                 (uint8_t *)lun->mode_pages.rw_er_page;
4017                         break;
4018                 }
4019                 case SMS_FORMAT_DEVICE_PAGE: {
4020                         struct scsi_format_page *format_page;
4021
4022                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4023                             ("subpage %#x for page %#x is incorrect!",
4024                             page_index->subpage, page_code));
4025
4026                         /*
4027                          * Sectors per track are set above.  Bytes per
4028                          * sector need to be set here on a per-LUN basis.
4029                          */
4030                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4031                                &format_page_default,
4032                                sizeof(format_page_default));
4033                         memcpy(&lun->mode_pages.format_page[
4034                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4035                                sizeof(format_page_changeable));
4036                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4037                                &format_page_default,
4038                                sizeof(format_page_default));
4039                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4040                                &format_page_default,
4041                                sizeof(format_page_default));
4042
4043                         format_page = &lun->mode_pages.format_page[
4044                                 CTL_PAGE_CURRENT];
4045                         scsi_ulto2b(lun->be_lun->blocksize,
4046                                     format_page->bytes_per_sector);
4047
4048                         format_page = &lun->mode_pages.format_page[
4049                                 CTL_PAGE_DEFAULT];
4050                         scsi_ulto2b(lun->be_lun->blocksize,
4051                                     format_page->bytes_per_sector);
4052
4053                         format_page = &lun->mode_pages.format_page[
4054                                 CTL_PAGE_SAVED];
4055                         scsi_ulto2b(lun->be_lun->blocksize,
4056                                     format_page->bytes_per_sector);
4057
4058                         page_index->page_data =
4059                                 (uint8_t *)lun->mode_pages.format_page;
4060                         break;
4061                 }
4062                 case SMS_RIGID_DISK_PAGE: {
4063                         struct scsi_rigid_disk_page *rigid_disk_page;
4064                         uint32_t sectors_per_cylinder;
4065                         uint64_t cylinders;
4066 #ifndef __XSCALE__
4067                         int shift;
4068 #endif /* !__XSCALE__ */
4069
4070                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4071                             ("subpage %#x for page %#x is incorrect!",
4072                             page_index->subpage, page_code));
4073
4074                         /*
4075                          * Rotation rate and sectors per track are set
4076                          * above.  We calculate the cylinders here based on
4077                          * capacity.  Due to the number of heads and
4078                          * sectors per track we're using, smaller arrays
4079                          * may turn out to have 0 cylinders.  Linux and
4080                          * FreeBSD don't pay attention to these mode pages
4081                          * to figure out capacity, but Solaris does.  It
4082                          * seems to deal with 0 cylinders just fine, and
4083                          * works out a fake geometry based on the capacity.
4084                          */
4085                         memcpy(&lun->mode_pages.rigid_disk_page[
4086                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4087                                sizeof(rigid_disk_page_default));
4088                         memcpy(&lun->mode_pages.rigid_disk_page[
4089                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4090                                sizeof(rigid_disk_page_changeable));
4091
4092                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4093                                 CTL_DEFAULT_HEADS;
4094
4095                         /*
4096                          * The divide method here will be more accurate,
4097                          * probably, but results in floating point being
4098                          * used in the kernel on i386 (__udivdi3()).  On the
4099                          * XScale, though, __udivdi3() is implemented in
4100                          * software.
4101                          *
4102                          * The shift method for cylinder calculation is
4103                          * accurate if sectors_per_cylinder is a power of
4104                          * 2.  Otherwise it might be slightly off -- you
4105                          * might have a bit of a truncation problem.
4106                          */
4107 #ifdef  __XSCALE__
4108                         cylinders = (lun->be_lun->maxlba + 1) /
4109                                 sectors_per_cylinder;
4110 #else
4111                         for (shift = 31; shift > 0; shift--) {
4112                                 if (sectors_per_cylinder & (1 << shift))
4113                                         break;
4114                         }
4115                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4116 #endif
4117
4118                         /*
4119                          * We've basically got 3 bytes, or 24 bits for the
4120                          * cylinder size in the mode page.  If we're over,
4121                          * just round down to 2^24.
4122                          */
4123                         if (cylinders > 0xffffff)
4124                                 cylinders = 0xffffff;
4125
4126                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4127                                 CTL_PAGE_DEFAULT];
4128                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4129
4130                         if ((value = ctl_get_opt(&lun->be_lun->options,
4131                             "rpm")) != NULL) {
4132                                 scsi_ulto2b(strtol(value, NULL, 0),
4133                                      rigid_disk_page->rotation_rate);
4134                         }
4135
4136                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4137                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4138                                sizeof(rigid_disk_page_default));
4139                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4140                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4141                                sizeof(rigid_disk_page_default));
4142
4143                         page_index->page_data =
4144                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4145                         break;
4146                 }
4147                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4148                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4149                             ("subpage %#x for page %#x is incorrect!",
4150                             page_index->subpage, page_code));
4151                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4152                                &verify_er_page_default,
4153                                sizeof(verify_er_page_default));
4154                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4155                                &verify_er_page_changeable,
4156                                sizeof(verify_er_page_changeable));
4157                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4158                                &verify_er_page_default,
4159                                sizeof(verify_er_page_default));
4160                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4161                                &verify_er_page_default,
4162                                sizeof(verify_er_page_default));
4163                         page_index->page_data =
4164                                 (uint8_t *)lun->mode_pages.verify_er_page;
4165                         break;
4166                 }
4167                 case SMS_CACHING_PAGE: {
4168                         struct scsi_caching_page *caching_page;
4169
4170                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4171                             ("subpage %#x for page %#x is incorrect!",
4172                             page_index->subpage, page_code));
4173                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4174                                &caching_page_default,
4175                                sizeof(caching_page_default));
4176                         memcpy(&lun->mode_pages.caching_page[
4177                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4178                                sizeof(caching_page_changeable));
4179                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4180                                &caching_page_default,
4181                                sizeof(caching_page_default));
4182                         caching_page = &lun->mode_pages.caching_page[
4183                             CTL_PAGE_SAVED];
4184                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
4185                         if (value != NULL && strcmp(value, "off") == 0)
4186                                 caching_page->flags1 &= ~SCP_WCE;
4187                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4188                         if (value != NULL && strcmp(value, "off") == 0)
4189                                 caching_page->flags1 |= SCP_RCD;
4190                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4191                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4192                                sizeof(caching_page_default));
4193                         page_index->page_data =
4194                                 (uint8_t *)lun->mode_pages.caching_page;
4195                         break;
4196                 }
4197                 case SMS_CONTROL_MODE_PAGE: {
4198                         switch (page_index->subpage) {
4199                         case SMS_SUBPAGE_PAGE_0: {
4200                                 struct scsi_control_page *control_page;
4201
4202                                 memcpy(&lun->mode_pages.control_page[
4203                                     CTL_PAGE_DEFAULT],
4204                                        &control_page_default,
4205                                        sizeof(control_page_default));
4206                                 memcpy(&lun->mode_pages.control_page[
4207                                     CTL_PAGE_CHANGEABLE],
4208                                        &control_page_changeable,
4209                                        sizeof(control_page_changeable));
4210                                 memcpy(&lun->mode_pages.control_page[
4211                                     CTL_PAGE_SAVED],
4212                                        &control_page_default,
4213                                        sizeof(control_page_default));
4214                                 control_page = &lun->mode_pages.control_page[
4215                                     CTL_PAGE_SAVED];
4216                                 value = ctl_get_opt(&lun->be_lun->options,
4217                                     "reordering");
4218                                 if (value != NULL &&
4219                                     strcmp(value, "unrestricted") == 0) {
4220                                         control_page->queue_flags &=
4221                                             ~SCP_QUEUE_ALG_MASK;
4222                                         control_page->queue_flags |=
4223                                             SCP_QUEUE_ALG_UNRESTRICTED;
4224                                 }
4225                                 memcpy(&lun->mode_pages.control_page[
4226                                     CTL_PAGE_CURRENT],
4227                                        &lun->mode_pages.control_page[
4228                                     CTL_PAGE_SAVED],
4229                                        sizeof(control_page_default));
4230                                 page_index->page_data =
4231                                     (uint8_t *)lun->mode_pages.control_page;
4232                                 break;
4233                         }
4234                         case 0x01:
4235                                 memcpy(&lun->mode_pages.control_ext_page[
4236                                     CTL_PAGE_DEFAULT],
4237                                        &control_ext_page_default,
4238                                        sizeof(control_ext_page_default));
4239                                 memcpy(&lun->mode_pages.control_ext_page[
4240                                     CTL_PAGE_CHANGEABLE],
4241                                        &control_ext_page_changeable,
4242                                        sizeof(control_ext_page_changeable));
4243                                 memcpy(&lun->mode_pages.control_ext_page[
4244                                     CTL_PAGE_SAVED],
4245                                        &control_ext_page_default,
4246                                        sizeof(control_ext_page_default));
4247                                 memcpy(&lun->mode_pages.control_ext_page[
4248                                     CTL_PAGE_CURRENT],
4249                                        &lun->mode_pages.control_ext_page[
4250                                     CTL_PAGE_SAVED],
4251                                        sizeof(control_ext_page_default));
4252                                 page_index->page_data =
4253                                     (uint8_t *)lun->mode_pages.control_ext_page;
4254                                 break;
4255                         default:
4256                                 panic("subpage %#x for page %#x is incorrect!",
4257                                       page_index->subpage, page_code);
4258                         }
4259                         break;
4260                 }
4261                 case SMS_INFO_EXCEPTIONS_PAGE: {
4262                         switch (page_index->subpage) {
4263                         case SMS_SUBPAGE_PAGE_0:
4264                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4265                                        &ie_page_default,
4266                                        sizeof(ie_page_default));
4267                                 memcpy(&lun->mode_pages.ie_page[
4268                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4269                                        sizeof(ie_page_changeable));
4270                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4271                                        &ie_page_default,
4272                                        sizeof(ie_page_default));
4273                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4274                                        &ie_page_default,
4275                                        sizeof(ie_page_default));
4276                                 page_index->page_data =
4277                                         (uint8_t *)lun->mode_pages.ie_page;
4278                                 break;
4279                         case 0x02: {
4280                                 struct ctl_logical_block_provisioning_page *page;
4281
4282                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4283                                        &lbp_page_default,
4284                                        sizeof(lbp_page_default));
4285                                 memcpy(&lun->mode_pages.lbp_page[
4286                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4287                                        sizeof(lbp_page_changeable));
4288                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4289                                        &lbp_page_default,
4290                                        sizeof(lbp_page_default));
4291                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4292                                 value = ctl_get_opt(&lun->be_lun->options,
4293                                     "avail-threshold");
4294                                 if (value != NULL &&
4295                                     ctl_expand_number(value, &ival) == 0) {
4296                                         page->descr[0].flags |= SLBPPD_ENABLED |
4297                                             SLBPPD_ARMING_DEC;
4298                                         if (lun->be_lun->blocksize)
4299                                                 ival /= lun->be_lun->blocksize;
4300                                         else
4301                                                 ival /= 512;
4302                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4303                                             page->descr[0].count);
4304                                 }
4305                                 value = ctl_get_opt(&lun->be_lun->options,
4306                                     "used-threshold");
4307                                 if (value != NULL &&
4308                                     ctl_expand_number(value, &ival) == 0) {
4309                                         page->descr[1].flags |= SLBPPD_ENABLED |
4310                                             SLBPPD_ARMING_INC;
4311                                         if (lun->be_lun->blocksize)
4312                                                 ival /= lun->be_lun->blocksize;
4313                                         else
4314                                                 ival /= 512;
4315                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4316                                             page->descr[1].count);
4317                                 }
4318                                 value = ctl_get_opt(&lun->be_lun->options,
4319                                     "pool-avail-threshold");
4320                                 if (value != NULL &&
4321                                     ctl_expand_number(value, &ival) == 0) {
4322                                         page->descr[2].flags |= SLBPPD_ENABLED |
4323                                             SLBPPD_ARMING_DEC;
4324                                         if (lun->be_lun->blocksize)
4325                                                 ival /= lun->be_lun->blocksize;
4326                                         else
4327                                                 ival /= 512;
4328                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4329                                             page->descr[2].count);
4330                                 }
4331                                 value = ctl_get_opt(&lun->be_lun->options,
4332                                     "pool-used-threshold");
4333                                 if (value != NULL &&
4334                                     ctl_expand_number(value, &ival) == 0) {
4335                                         page->descr[3].flags |= SLBPPD_ENABLED |
4336                                             SLBPPD_ARMING_INC;
4337                                         if (lun->be_lun->blocksize)
4338                                                 ival /= lun->be_lun->blocksize;
4339                                         else
4340                                                 ival /= 512;
4341                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4342                                             page->descr[3].count);
4343                                 }
4344                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4345                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4346                                        sizeof(lbp_page_default));
4347                                 page_index->page_data =
4348                                         (uint8_t *)lun->mode_pages.lbp_page;
4349                                 break;
4350                         }
4351                         default:
4352                                 panic("subpage %#x for page %#x is incorrect!",
4353                                       page_index->subpage, page_code);
4354                         }
4355                         break;
4356                 }
4357                 case SMS_CDDVD_CAPS_PAGE:{
4358                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4359                             ("subpage %#x for page %#x is incorrect!",
4360                             page_index->subpage, page_code));
4361                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4362                                &cddvd_page_default,
4363                                sizeof(cddvd_page_default));
4364                         memcpy(&lun->mode_pages.cddvd_page[
4365                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4366                                sizeof(cddvd_page_changeable));
4367                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4368                                &cddvd_page_default,
4369                                sizeof(cddvd_page_default));
4370                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4371                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4372                                sizeof(cddvd_page_default));
4373                         page_index->page_data =
4374                                 (uint8_t *)lun->mode_pages.cddvd_page;
4375                         break;
4376                 }
4377                 default:
4378                         panic("invalid page code value %#x", page_code);
4379                 }
4380         }
4381
4382         return (CTL_RETVAL_COMPLETE);
4383 }
4384
4385 static int
4386 ctl_init_log_page_index(struct ctl_lun *lun)
4387 {
4388         struct ctl_page_index *page_index;
4389         int i, j, k, prev;
4390
4391         memcpy(&lun->log_pages.index, log_page_index_template,
4392                sizeof(log_page_index_template));
4393
4394         prev = -1;
4395         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4396
4397                 page_index = &lun->log_pages.index[i];
4398                 if (lun->be_lun->lun_type == T_DIRECT &&
4399                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4400                         continue;
4401                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4402                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4403                         continue;
4404                 if (lun->be_lun->lun_type == T_CDROM &&
4405                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4406                         continue;
4407
4408                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4409                     lun->backend->lun_attr == NULL)
4410                         continue;
4411
4412                 if (page_index->page_code != prev) {
4413                         lun->log_pages.pages_page[j] = page_index->page_code;
4414                         prev = page_index->page_code;
4415                         j++;
4416                 }
4417                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4418                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4419                 k++;
4420         }
4421         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4422         lun->log_pages.index[0].page_len = j;
4423         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4424         lun->log_pages.index[1].page_len = k * 2;
4425         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4426         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4427         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4428         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4429         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.ie_page;
4430         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.ie_page);
4431
4432         return (CTL_RETVAL_COMPLETE);
4433 }
4434
4435 static int
4436 hex2bin(const char *str, uint8_t *buf, int buf_size)
4437 {
4438         int i;
4439         u_char c;
4440
4441         memset(buf, 0, buf_size);
4442         while (isspace(str[0]))
4443                 str++;
4444         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4445                 str += 2;
4446         buf_size *= 2;
4447         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4448                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4449                         str++;
4450                 c = str[i];
4451                 if (isdigit(c))
4452                         c -= '0';
4453                 else if (isalpha(c))
4454                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4455                 else
4456                         break;
4457                 if (c >= 16)
4458                         break;
4459                 if ((i & 1) == 0)
4460                         buf[i / 2] |= (c << 4);
4461                 else
4462                         buf[i / 2] |= c;
4463         }
4464         return ((i + 1) / 2);
4465 }
4466
4467 /*
4468  * LUN allocation.
4469  *
4470  * Requirements:
4471  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4472  *   wants us to allocate the LUN and he can block.
4473  * - ctl_softc is always set
4474  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4475  *
4476  * Returns 0 for success, non-zero (errno) for failure.
4477  */
4478 static int
4479 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4480               struct ctl_be_lun *const be_lun)
4481 {
4482         struct ctl_lun *nlun, *lun;
4483         struct scsi_vpd_id_descriptor *desc;
4484         struct scsi_vpd_id_t10 *t10id;
4485         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4486         int lun_number, lun_malloced;
4487         int devidlen, idlen1, idlen2 = 0, len;
4488
4489         if (be_lun == NULL)
4490                 return (EINVAL);
4491
4492         /*
4493          * We currently only support Direct Access or Processor LUN types.
4494          */
4495         switch (be_lun->lun_type) {
4496         case T_DIRECT:
4497         case T_PROCESSOR:
4498         case T_CDROM:
4499                 break;
4500         case T_SEQUENTIAL:
4501         case T_CHANGER:
4502         default:
4503                 be_lun->lun_config_status(be_lun->be_lun,
4504                                           CTL_LUN_CONFIG_FAILURE);
4505                 break;
4506         }
4507         if (ctl_lun == NULL) {
4508                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4509                 lun_malloced = 1;
4510         } else {
4511                 lun_malloced = 0;
4512                 lun = ctl_lun;
4513         }
4514
4515         memset(lun, 0, sizeof(*lun));
4516         if (lun_malloced)
4517                 lun->flags = CTL_LUN_MALLOCED;
4518
4519         /* Generate LUN ID. */
4520         devidlen = max(CTL_DEVID_MIN_LEN,
4521             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4522         idlen1 = sizeof(*t10id) + devidlen;
4523         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4524         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4525         if (scsiname != NULL) {
4526                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4527                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4528         }
4529         eui = ctl_get_opt(&be_lun->options, "eui");
4530         if (eui != NULL) {
4531                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4532         }
4533         naa = ctl_get_opt(&be_lun->options, "naa");
4534         if (naa != NULL) {
4535                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4536         }
4537         uuid = ctl_get_opt(&be_lun->options, "uuid");
4538         if (uuid != NULL) {
4539                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4540         }
4541         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4542             M_CTL, M_WAITOK | M_ZERO);
4543         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4544         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4545         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4546         desc->length = idlen1;
4547         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4548         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4549         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4550                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4551         } else {
4552                 strncpy(t10id->vendor, vendor,
4553                     min(sizeof(t10id->vendor), strlen(vendor)));
4554         }
4555         strncpy((char *)t10id->vendor_spec_id,
4556             (char *)be_lun->device_id, devidlen);
4557         if (scsiname != NULL) {
4558                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4559                     desc->length);
4560                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4561                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4562                     SVPD_ID_TYPE_SCSI_NAME;
4563                 desc->length = idlen2;
4564                 strlcpy(desc->identifier, scsiname, idlen2);
4565         }
4566         if (eui != NULL) {
4567                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4568                     desc->length);
4569                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4570                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4571                     SVPD_ID_TYPE_EUI64;
4572                 desc->length = hex2bin(eui, desc->identifier, 16);
4573                 desc->length = desc->length > 12 ? 16 :
4574                     (desc->length > 8 ? 12 : 8);
4575                 len -= 16 - desc->length;
4576         }
4577         if (naa != NULL) {
4578                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4579                     desc->length);
4580                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4581                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4582                     SVPD_ID_TYPE_NAA;
4583                 desc->length = hex2bin(naa, desc->identifier, 16);
4584                 desc->length = desc->length > 8 ? 16 : 8;
4585                 len -= 16 - desc->length;
4586         }
4587         if (uuid != NULL) {
4588                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4589                     desc->length);
4590                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4591                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4592                     SVPD_ID_TYPE_UUID;
4593                 desc->identifier[0] = 0x10;
4594                 hex2bin(uuid, &desc->identifier[2], 16);
4595                 desc->length = 18;
4596         }
4597         lun->lun_devid->len = len;
4598
4599         mtx_lock(&ctl_softc->ctl_lock);
4600         /*
4601          * See if the caller requested a particular LUN number.  If so, see
4602          * if it is available.  Otherwise, allocate the first available LUN.
4603          */
4604         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4605                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4606                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4607                         mtx_unlock(&ctl_softc->ctl_lock);
4608                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4609                                 printf("ctl: requested LUN ID %d is higher "
4610                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4611                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4612                         } else {
4613                                 /*
4614                                  * XXX KDM return an error, or just assign
4615                                  * another LUN ID in this case??
4616                                  */
4617                                 printf("ctl: requested LUN ID %d is already "
4618                                        "in use\n", be_lun->req_lun_id);
4619                         }
4620 fail:
4621                         free(lun->lun_devid, M_CTL);
4622                         if (lun->flags & CTL_LUN_MALLOCED)
4623                                 free(lun, M_CTL);
4624                         be_lun->lun_config_status(be_lun->be_lun,
4625                                                   CTL_LUN_CONFIG_FAILURE);
4626                         return (ENOSPC);
4627                 }
4628                 lun_number = be_lun->req_lun_id;
4629         } else {
4630                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4631                 if (lun_number == -1) {
4632                         mtx_unlock(&ctl_softc->ctl_lock);
4633                         printf("ctl: can't allocate LUN, out of LUNs\n");
4634                         goto fail;
4635                 }
4636         }
4637         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4638         mtx_unlock(&ctl_softc->ctl_lock);
4639
4640         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4641         lun->lun = lun_number;
4642         lun->be_lun = be_lun;
4643         /*
4644          * The processor LUN is always enabled.  Disk LUNs come on line
4645          * disabled, and must be enabled by the backend.
4646          */
4647         lun->flags |= CTL_LUN_DISABLED;
4648         lun->backend = be_lun->be;
4649         be_lun->ctl_lun = lun;
4650         be_lun->lun_id = lun_number;
4651         atomic_add_int(&be_lun->be->num_luns, 1);
4652         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4653                 lun->flags |= CTL_LUN_EJECTED;
4654         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4655                 lun->flags |= CTL_LUN_NO_MEDIA;
4656         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4657                 lun->flags |= CTL_LUN_STOPPED;
4658
4659         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4660                 lun->flags |= CTL_LUN_PRIMARY_SC;
4661
4662         value = ctl_get_opt(&be_lun->options, "removable");
4663         if (value != NULL) {
4664                 if (strcmp(value, "on") == 0)
4665                         lun->flags |= CTL_LUN_REMOVABLE;
4666         } else if (be_lun->lun_type == T_CDROM)
4667                 lun->flags |= CTL_LUN_REMOVABLE;
4668
4669         lun->ctl_softc = ctl_softc;
4670 #ifdef CTL_TIME_IO
4671         lun->last_busy = getsbinuptime();
4672 #endif
4673         TAILQ_INIT(&lun->ooa_queue);
4674         TAILQ_INIT(&lun->blocked_queue);
4675         STAILQ_INIT(&lun->error_list);
4676         lun->ie_reported = 1;
4677         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4678         ctl_tpc_lun_init(lun);
4679         if (lun->flags & CTL_LUN_REMOVABLE) {
4680                 lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4,
4681                     M_CTL, M_WAITOK);
4682         }
4683
4684         /*
4685          * Initialize the mode and log page index.
4686          */
4687         ctl_init_page_index(lun);
4688         ctl_init_log_page_index(lun);
4689
4690         /* Setup statistics gathering */
4691 #ifdef CTL_LEGACY_STATS
4692         lun->legacy_stats.device_type = be_lun->lun_type;
4693         lun->legacy_stats.lun_number = lun_number;
4694         lun->legacy_stats.blocksize = be_lun->blocksize;
4695         if (be_lun->blocksize == 0)
4696                 lun->legacy_stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4697         for (len = 0; len < CTL_MAX_PORTS; len++)
4698                 lun->legacy_stats.ports[len].targ_port = len;
4699 #endif /* CTL_LEGACY_STATS */
4700         lun->stats.item = lun_number;
4701
4702         /*
4703          * Now, before we insert this lun on the lun list, set the lun
4704          * inventory changed UA for all other luns.
4705          */
4706         mtx_lock(&ctl_softc->ctl_lock);
4707         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4708                 mtx_lock(&nlun->lun_lock);
4709                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4710                 mtx_unlock(&nlun->lun_lock);
4711         }
4712         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4713         ctl_softc->ctl_luns[lun_number] = lun;
4714         ctl_softc->num_luns++;
4715         mtx_unlock(&ctl_softc->ctl_lock);
4716
4717         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4718         return (0);
4719 }
4720
4721 /*
4722  * Delete a LUN.
4723  * Assumptions:
4724  * - LUN has already been marked invalid and any pending I/O has been taken
4725  *   care of.
4726  */
4727 static int
4728 ctl_free_lun(struct ctl_lun *lun)
4729 {
4730         struct ctl_softc *softc = lun->ctl_softc;
4731         struct ctl_lun *nlun;
4732         int i;
4733
4734         KASSERT(TAILQ_EMPTY(&lun->ooa_queue),
4735             ("Freeing a LUN %p with outstanding I/O!\n", lun));
4736
4737         mtx_lock(&softc->ctl_lock);
4738         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4739         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4740         softc->ctl_luns[lun->lun] = NULL;
4741         softc->num_luns--;
4742         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4743                 mtx_lock(&nlun->lun_lock);
4744                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4745                 mtx_unlock(&nlun->lun_lock);
4746         }
4747         mtx_unlock(&softc->ctl_lock);
4748
4749         /*
4750          * Tell the backend to free resources, if this LUN has a backend.
4751          */
4752         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4753         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4754
4755         lun->ie_reportcnt = UINT32_MAX;
4756         callout_drain(&lun->ie_callout);
4757         ctl_tpc_lun_shutdown(lun);
4758         mtx_destroy(&lun->lun_lock);
4759         free(lun->lun_devid, M_CTL);
4760         for (i = 0; i < CTL_MAX_PORTS; i++)
4761                 free(lun->pending_ua[i], M_CTL);
4762         for (i = 0; i < CTL_MAX_PORTS; i++)
4763                 free(lun->pr_keys[i], M_CTL);
4764         free(lun->write_buffer, M_CTL);
4765         free(lun->prevent, M_CTL);
4766         if (lun->flags & CTL_LUN_MALLOCED)
4767                 free(lun, M_CTL);
4768
4769         return (0);
4770 }
4771
4772 static void
4773 ctl_create_lun(struct ctl_be_lun *be_lun)
4774 {
4775
4776         /*
4777          * ctl_alloc_lun() should handle all potential failure cases.
4778          */
4779         ctl_alloc_lun(control_softc, NULL, be_lun);
4780 }
4781
4782 int
4783 ctl_add_lun(struct ctl_be_lun *be_lun)
4784 {
4785         struct ctl_softc *softc = control_softc;
4786
4787         mtx_lock(&softc->ctl_lock);
4788         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4789         mtx_unlock(&softc->ctl_lock);
4790         wakeup(&softc->pending_lun_queue);
4791
4792         return (0);
4793 }
4794
4795 int
4796 ctl_enable_lun(struct ctl_be_lun *be_lun)
4797 {
4798         struct ctl_softc *softc;
4799         struct ctl_port *port, *nport;
4800         struct ctl_lun *lun;
4801         int retval;
4802
4803         lun = (struct ctl_lun *)be_lun->ctl_lun;
4804         softc = lun->ctl_softc;
4805
4806         mtx_lock(&softc->ctl_lock);
4807         mtx_lock(&lun->lun_lock);
4808         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4809                 /*
4810                  * eh?  Why did we get called if the LUN is already
4811                  * enabled?
4812                  */
4813                 mtx_unlock(&lun->lun_lock);
4814                 mtx_unlock(&softc->ctl_lock);
4815                 return (0);
4816         }
4817         lun->flags &= ~CTL_LUN_DISABLED;
4818         mtx_unlock(&lun->lun_lock);
4819
4820         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4821                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4822                     port->lun_map != NULL || port->lun_enable == NULL)
4823                         continue;
4824
4825                 /*
4826                  * Drop the lock while we call the FETD's enable routine.
4827                  * This can lead to a callback into CTL (at least in the
4828                  * case of the internal initiator frontend.
4829                  */
4830                 mtx_unlock(&softc->ctl_lock);
4831                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4832                 mtx_lock(&softc->ctl_lock);
4833                 if (retval != 0) {
4834                         printf("%s: FETD %s port %d returned error "
4835                                "%d for lun_enable on lun %jd\n",
4836                                __func__, port->port_name, port->targ_port,
4837                                retval, (intmax_t)lun->lun);
4838                 }
4839         }
4840
4841         mtx_unlock(&softc->ctl_lock);
4842         ctl_isc_announce_lun(lun);
4843
4844         return (0);
4845 }
4846
4847 int
4848 ctl_disable_lun(struct ctl_be_lun *be_lun)
4849 {
4850         struct ctl_softc *softc;
4851         struct ctl_port *port;
4852         struct ctl_lun *lun;
4853         int retval;
4854
4855         lun = (struct ctl_lun *)be_lun->ctl_lun;
4856         softc = lun->ctl_softc;
4857
4858         mtx_lock(&softc->ctl_lock);
4859         mtx_lock(&lun->lun_lock);
4860         if (lun->flags & CTL_LUN_DISABLED) {
4861                 mtx_unlock(&lun->lun_lock);
4862                 mtx_unlock(&softc->ctl_lock);
4863                 return (0);
4864         }
4865         lun->flags |= CTL_LUN_DISABLED;
4866         mtx_unlock(&lun->lun_lock);
4867
4868         STAILQ_FOREACH(port, &softc->port_list, links) {
4869                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4870                     port->lun_map != NULL || port->lun_disable == NULL)
4871                         continue;
4872
4873                 /*
4874                  * Drop the lock before we call the frontend's disable
4875                  * routine, to avoid lock order reversals.
4876                  *
4877                  * XXX KDM what happens if the frontend list changes while
4878                  * we're traversing it?  It's unlikely, but should be handled.
4879                  */
4880                 mtx_unlock(&softc->ctl_lock);
4881                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4882                 mtx_lock(&softc->ctl_lock);
4883                 if (retval != 0) {
4884                         printf("%s: FETD %s port %d returned error "
4885                                "%d for lun_disable on lun %jd\n",
4886                                __func__, port->port_name, port->targ_port,
4887                                retval, (intmax_t)lun->lun);
4888                 }
4889         }
4890
4891         mtx_unlock(&softc->ctl_lock);
4892         ctl_isc_announce_lun(lun);
4893
4894         return (0);
4895 }
4896
4897 int
4898 ctl_start_lun(struct ctl_be_lun *be_lun)
4899 {
4900         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4901
4902         mtx_lock(&lun->lun_lock);
4903         lun->flags &= ~CTL_LUN_STOPPED;
4904         mtx_unlock(&lun->lun_lock);
4905         return (0);
4906 }
4907
4908 int
4909 ctl_stop_lun(struct ctl_be_lun *be_lun)
4910 {
4911         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4912
4913         mtx_lock(&lun->lun_lock);
4914         lun->flags |= CTL_LUN_STOPPED;
4915         mtx_unlock(&lun->lun_lock);
4916         return (0);
4917 }
4918
4919 int
4920 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4921 {
4922         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4923
4924         mtx_lock(&lun->lun_lock);
4925         lun->flags |= CTL_LUN_NO_MEDIA;
4926         mtx_unlock(&lun->lun_lock);
4927         return (0);
4928 }
4929
4930 int
4931 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4932 {
4933         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4934         union ctl_ha_msg msg;
4935
4936         mtx_lock(&lun->lun_lock);
4937         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4938         if (lun->flags & CTL_LUN_REMOVABLE)
4939                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4940         mtx_unlock(&lun->lun_lock);
4941         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4942             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4943                 bzero(&msg.ua, sizeof(msg.ua));
4944                 msg.hdr.msg_type = CTL_MSG_UA;
4945                 msg.hdr.nexus.initid = -1;
4946                 msg.hdr.nexus.targ_port = -1;
4947                 msg.hdr.nexus.targ_lun = lun->lun;
4948                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4949                 msg.ua.ua_all = 1;
4950                 msg.ua.ua_set = 1;
4951                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4952                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4953                     M_WAITOK);
4954         }
4955         return (0);
4956 }
4957
4958 int
4959 ctl_lun_ejected(struct ctl_be_lun *be_lun)
4960 {
4961         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4962
4963         mtx_lock(&lun->lun_lock);
4964         lun->flags |= CTL_LUN_EJECTED;
4965         mtx_unlock(&lun->lun_lock);
4966         return (0);
4967 }
4968
4969 int
4970 ctl_lun_primary(struct ctl_be_lun *be_lun)
4971 {
4972         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4973
4974         mtx_lock(&lun->lun_lock);
4975         lun->flags |= CTL_LUN_PRIMARY_SC;
4976         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4977         mtx_unlock(&lun->lun_lock);
4978         ctl_isc_announce_lun(lun);
4979         return (0);
4980 }
4981
4982 int
4983 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4984 {
4985         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4986
4987         mtx_lock(&lun->lun_lock);
4988         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4989         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4990         mtx_unlock(&lun->lun_lock);
4991         ctl_isc_announce_lun(lun);
4992         return (0);
4993 }
4994
4995 int
4996 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4997 {
4998         struct ctl_softc *softc;
4999         struct ctl_lun *lun;
5000
5001         lun = (struct ctl_lun *)be_lun->ctl_lun;
5002         softc = lun->ctl_softc;
5003
5004         mtx_lock(&lun->lun_lock);
5005
5006         /*
5007          * The LUN needs to be disabled before it can be marked invalid.
5008          */
5009         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
5010                 mtx_unlock(&lun->lun_lock);
5011                 return (-1);
5012         }
5013         /*
5014          * Mark the LUN invalid.
5015          */
5016         lun->flags |= CTL_LUN_INVALID;
5017
5018         /*
5019          * If there is nothing in the OOA queue, go ahead and free the LUN.
5020          * If we have something in the OOA queue, we'll free it when the
5021          * last I/O completes.
5022          */
5023         if (TAILQ_EMPTY(&lun->ooa_queue)) {
5024                 mtx_unlock(&lun->lun_lock);
5025                 ctl_free_lun(lun);
5026         } else
5027                 mtx_unlock(&lun->lun_lock);
5028
5029         return (0);
5030 }
5031
5032 void
5033 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5034 {
5035         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5036         union ctl_ha_msg msg;
5037
5038         mtx_lock(&lun->lun_lock);
5039         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
5040         mtx_unlock(&lun->lun_lock);
5041         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5042                 /* Send msg to other side. */
5043                 bzero(&msg.ua, sizeof(msg.ua));
5044                 msg.hdr.msg_type = CTL_MSG_UA;
5045                 msg.hdr.nexus.initid = -1;
5046                 msg.hdr.nexus.targ_port = -1;
5047                 msg.hdr.nexus.targ_lun = lun->lun;
5048                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5049                 msg.ua.ua_all = 1;
5050                 msg.ua.ua_set = 1;
5051                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
5052                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5053                     M_WAITOK);
5054         }
5055 }
5056
5057 /*
5058  * Backend "memory move is complete" callback for requests that never
5059  * make it down to say RAIDCore's configuration code.
5060  */
5061 int
5062 ctl_config_move_done(union ctl_io *io)
5063 {
5064         int retval;
5065
5066         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5067         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5068             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5069
5070         if ((io->io_hdr.port_status != 0) &&
5071             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5072              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5073                 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
5074                     /*retry_count*/ io->io_hdr.port_status);
5075         } else if (io->scsiio.kern_data_resid != 0 &&
5076             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
5077             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5078              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5079                 ctl_set_invalid_field_ciu(&io->scsiio);
5080         }
5081
5082         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5083                 ctl_data_print(io);
5084         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5085             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5086              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5087             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5088                 /*
5089                  * XXX KDM just assuming a single pointer here, and not a
5090                  * S/G list.  If we start using S/G lists for config data,
5091                  * we'll need to know how to clean them up here as well.
5092                  */
5093                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5094                         free(io->scsiio.kern_data_ptr, M_CTL);
5095                 ctl_done(io);
5096                 retval = CTL_RETVAL_COMPLETE;
5097         } else {
5098                 /*
5099                  * XXX KDM now we need to continue data movement.  Some
5100                  * options:
5101                  * - call ctl_scsiio() again?  We don't do this for data
5102                  *   writes, because for those at least we know ahead of
5103                  *   time where the write will go and how long it is.  For
5104                  *   config writes, though, that information is largely
5105                  *   contained within the write itself, thus we need to
5106                  *   parse out the data again.
5107                  *
5108                  * - Call some other function once the data is in?
5109                  */
5110
5111                 /*
5112                  * XXX KDM call ctl_scsiio() again for now, and check flag
5113                  * bits to see whether we're allocated or not.
5114                  */
5115                 retval = ctl_scsiio(&io->scsiio);
5116         }
5117         return (retval);
5118 }
5119
5120 /*
5121  * This gets called by a backend driver when it is done with a
5122  * data_submit method.
5123  */
5124 void
5125 ctl_data_submit_done(union ctl_io *io)
5126 {
5127         /*
5128          * If the IO_CONT flag is set, we need to call the supplied
5129          * function to continue processing the I/O, instead of completing
5130          * the I/O just yet.
5131          *
5132          * If there is an error, though, we don't want to keep processing.
5133          * Instead, just send status back to the initiator.
5134          */
5135         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5136             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5137             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5138              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5139                 io->scsiio.io_cont(io);
5140                 return;
5141         }
5142         ctl_done(io);
5143 }
5144
5145 /*
5146  * This gets called by a backend driver when it is done with a
5147  * configuration write.
5148  */
5149 void
5150 ctl_config_write_done(union ctl_io *io)
5151 {
5152         uint8_t *buf;
5153
5154         /*
5155          * If the IO_CONT flag is set, we need to call the supplied
5156          * function to continue processing the I/O, instead of completing
5157          * the I/O just yet.
5158          *
5159          * If there is an error, though, we don't want to keep processing.
5160          * Instead, just send status back to the initiator.
5161          */
5162         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5163             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5164             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5165              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5166                 io->scsiio.io_cont(io);
5167                 return;
5168         }
5169         /*
5170          * Since a configuration write can be done for commands that actually
5171          * have data allocated, like write buffer, and commands that have
5172          * no data, like start/stop unit, we need to check here.
5173          */
5174         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5175                 buf = io->scsiio.kern_data_ptr;
5176         else
5177                 buf = NULL;
5178         ctl_done(io);
5179         if (buf)
5180                 free(buf, M_CTL);
5181 }
5182
5183 void
5184 ctl_config_read_done(union ctl_io *io)
5185 {
5186         uint8_t *buf;
5187
5188         /*
5189          * If there is some error -- we are done, skip data transfer.
5190          */
5191         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5192             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5193              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5194                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5195                         buf = io->scsiio.kern_data_ptr;
5196                 else
5197                         buf = NULL;
5198                 ctl_done(io);
5199                 if (buf)
5200                         free(buf, M_CTL);
5201                 return;
5202         }
5203
5204         /*
5205          * If the IO_CONT flag is set, we need to call the supplied
5206          * function to continue processing the I/O, instead of completing
5207          * the I/O just yet.
5208          */
5209         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5210                 io->scsiio.io_cont(io);
5211                 return;
5212         }
5213
5214         ctl_datamove(io);
5215 }
5216
5217 /*
5218  * SCSI release command.
5219  */
5220 int
5221 ctl_scsi_release(struct ctl_scsiio *ctsio)
5222 {
5223         struct ctl_lun *lun = CTL_LUN(ctsio);
5224         uint32_t residx;
5225
5226         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5227
5228         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5229
5230         /*
5231          * XXX KDM right now, we only support LUN reservation.  We don't
5232          * support 3rd party reservations, or extent reservations, which
5233          * might actually need the parameter list.  If we've gotten this
5234          * far, we've got a LUN reservation.  Anything else got kicked out
5235          * above.  So, according to SPC, ignore the length.
5236          */
5237
5238         mtx_lock(&lun->lun_lock);
5239
5240         /*
5241          * According to SPC, it is not an error for an intiator to attempt
5242          * to release a reservation on a LUN that isn't reserved, or that
5243          * is reserved by another initiator.  The reservation can only be
5244          * released, though, by the initiator who made it or by one of
5245          * several reset type events.
5246          */
5247         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5248                         lun->flags &= ~CTL_LUN_RESERVED;
5249
5250         mtx_unlock(&lun->lun_lock);
5251
5252         ctl_set_success(ctsio);
5253         ctl_done((union ctl_io *)ctsio);
5254         return (CTL_RETVAL_COMPLETE);
5255 }
5256
5257 int
5258 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5259 {
5260         struct ctl_lun *lun = CTL_LUN(ctsio);
5261         uint32_t residx;
5262
5263         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5264
5265         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5266
5267         /*
5268          * XXX KDM right now, we only support LUN reservation.  We don't
5269          * support 3rd party reservations, or extent reservations, which
5270          * might actually need the parameter list.  If we've gotten this
5271          * far, we've got a LUN reservation.  Anything else got kicked out
5272          * above.  So, according to SPC, ignore the length.
5273          */
5274
5275         mtx_lock(&lun->lun_lock);
5276         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5277                 ctl_set_reservation_conflict(ctsio);
5278                 goto bailout;
5279         }
5280
5281         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5282         if (lun->flags & CTL_LUN_PR_RESERVED) {
5283                 ctl_set_success(ctsio);
5284                 goto bailout;
5285         }
5286
5287         lun->flags |= CTL_LUN_RESERVED;
5288         lun->res_idx = residx;
5289         ctl_set_success(ctsio);
5290
5291 bailout:
5292         mtx_unlock(&lun->lun_lock);
5293         ctl_done((union ctl_io *)ctsio);
5294         return (CTL_RETVAL_COMPLETE);
5295 }
5296
5297 int
5298 ctl_start_stop(struct ctl_scsiio *ctsio)
5299 {
5300         struct ctl_lun *lun = CTL_LUN(ctsio);
5301         struct scsi_start_stop_unit *cdb;
5302         int retval;
5303
5304         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5305
5306         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5307
5308         if ((cdb->how & SSS_PC_MASK) == 0) {
5309                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5310                     (cdb->how & SSS_START) == 0) {
5311                         uint32_t residx;
5312
5313                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5314                         if (ctl_get_prkey(lun, residx) == 0 ||
5315                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5316
5317                                 ctl_set_reservation_conflict(ctsio);
5318                                 ctl_done((union ctl_io *)ctsio);
5319                                 return (CTL_RETVAL_COMPLETE);
5320                         }
5321                 }
5322
5323                 if ((cdb->how & SSS_LOEJ) &&
5324                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5325                         ctl_set_invalid_field(ctsio,
5326                                               /*sks_valid*/ 1,
5327                                               /*command*/ 1,
5328                                               /*field*/ 4,
5329                                               /*bit_valid*/ 1,
5330                                               /*bit*/ 1);
5331                         ctl_done((union ctl_io *)ctsio);
5332                         return (CTL_RETVAL_COMPLETE);
5333                 }
5334
5335                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5336                     lun->prevent_count > 0) {
5337                         /* "Medium removal prevented" */
5338                         ctl_set_sense(ctsio, /*current_error*/ 1,
5339                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5340                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5341                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5342                         ctl_done((union ctl_io *)ctsio);
5343                         return (CTL_RETVAL_COMPLETE);
5344                 }
5345         }
5346
5347         retval = lun->backend->config_write((union ctl_io *)ctsio);
5348         return (retval);
5349 }
5350
5351 int
5352 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5353 {
5354         struct ctl_lun *lun = CTL_LUN(ctsio);
5355         struct scsi_prevent *cdb;
5356         int retval;
5357         uint32_t initidx;
5358
5359         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5360
5361         cdb = (struct scsi_prevent *)ctsio->cdb;
5362
5363         if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) {
5364                 ctl_set_invalid_opcode(ctsio);
5365                 ctl_done((union ctl_io *)ctsio);
5366                 return (CTL_RETVAL_COMPLETE);
5367         }
5368
5369         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5370         mtx_lock(&lun->lun_lock);
5371         if ((cdb->how & PR_PREVENT) &&
5372             ctl_is_set(lun->prevent, initidx) == 0) {
5373                 ctl_set_mask(lun->prevent, initidx);
5374                 lun->prevent_count++;
5375         } else if ((cdb->how & PR_PREVENT) == 0 &&
5376             ctl_is_set(lun->prevent, initidx)) {
5377                 ctl_clear_mask(lun->prevent, initidx);
5378                 lun->prevent_count--;
5379         }
5380         mtx_unlock(&lun->lun_lock);
5381         retval = lun->backend->config_write((union ctl_io *)ctsio);
5382         return (retval);
5383 }
5384
5385 /*
5386  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5387  * we don't really do anything with the LBA and length fields if the user
5388  * passes them in.  Instead we'll just flush out the cache for the entire
5389  * LUN.
5390  */
5391 int
5392 ctl_sync_cache(struct ctl_scsiio *ctsio)
5393 {
5394         struct ctl_lun *lun = CTL_LUN(ctsio);
5395         struct ctl_lba_len_flags *lbalen;
5396         uint64_t starting_lba;
5397         uint32_t block_count;
5398         int retval;
5399         uint8_t byte2;
5400
5401         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5402
5403         retval = 0;
5404
5405         switch (ctsio->cdb[0]) {
5406         case SYNCHRONIZE_CACHE: {
5407                 struct scsi_sync_cache *cdb;
5408                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5409
5410                 starting_lba = scsi_4btoul(cdb->begin_lba);
5411                 block_count = scsi_2btoul(cdb->lb_count);
5412                 byte2 = cdb->byte2;
5413                 break;
5414         }
5415         case SYNCHRONIZE_CACHE_16: {
5416                 struct scsi_sync_cache_16 *cdb;
5417                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5418
5419                 starting_lba = scsi_8btou64(cdb->begin_lba);
5420                 block_count = scsi_4btoul(cdb->lb_count);
5421                 byte2 = cdb->byte2;
5422                 break;
5423         }
5424         default:
5425                 ctl_set_invalid_opcode(ctsio);
5426                 ctl_done((union ctl_io *)ctsio);
5427                 goto bailout;
5428                 break; /* NOTREACHED */
5429         }
5430
5431         /*
5432          * We check the LBA and length, but don't do anything with them.
5433          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5434          * get flushed.  This check will just help satisfy anyone who wants
5435          * to see an error for an out of range LBA.
5436          */
5437         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5438                 ctl_set_lba_out_of_range(ctsio,
5439                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5440                 ctl_done((union ctl_io *)ctsio);
5441                 goto bailout;
5442         }
5443
5444         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5445         lbalen->lba = starting_lba;
5446         lbalen->len = block_count;
5447         lbalen->flags = byte2;
5448         retval = lun->backend->config_write((union ctl_io *)ctsio);
5449
5450 bailout:
5451         return (retval);
5452 }
5453
5454 int
5455 ctl_format(struct ctl_scsiio *ctsio)
5456 {
5457         struct scsi_format *cdb;
5458         int length, defect_list_len;
5459
5460         CTL_DEBUG_PRINT(("ctl_format\n"));
5461
5462         cdb = (struct scsi_format *)ctsio->cdb;
5463
5464         length = 0;
5465         if (cdb->byte2 & SF_FMTDATA) {
5466                 if (cdb->byte2 & SF_LONGLIST)
5467                         length = sizeof(struct scsi_format_header_long);
5468                 else
5469                         length = sizeof(struct scsi_format_header_short);
5470         }
5471
5472         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5473          && (length > 0)) {
5474                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5475                 ctsio->kern_data_len = length;
5476                 ctsio->kern_total_len = length;
5477                 ctsio->kern_rel_offset = 0;
5478                 ctsio->kern_sg_entries = 0;
5479                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5480                 ctsio->be_move_done = ctl_config_move_done;
5481                 ctl_datamove((union ctl_io *)ctsio);
5482
5483                 return (CTL_RETVAL_COMPLETE);
5484         }
5485
5486         defect_list_len = 0;
5487
5488         if (cdb->byte2 & SF_FMTDATA) {
5489                 if (cdb->byte2 & SF_LONGLIST) {
5490                         struct scsi_format_header_long *header;
5491
5492                         header = (struct scsi_format_header_long *)
5493                                 ctsio->kern_data_ptr;
5494
5495                         defect_list_len = scsi_4btoul(header->defect_list_len);
5496                         if (defect_list_len != 0) {
5497                                 ctl_set_invalid_field(ctsio,
5498                                                       /*sks_valid*/ 1,
5499                                                       /*command*/ 0,
5500                                                       /*field*/ 2,
5501                                                       /*bit_valid*/ 0,
5502                                                       /*bit*/ 0);
5503                                 goto bailout;
5504                         }
5505                 } else {
5506                         struct scsi_format_header_short *header;
5507
5508                         header = (struct scsi_format_header_short *)
5509                                 ctsio->kern_data_ptr;
5510
5511                         defect_list_len = scsi_2btoul(header->defect_list_len);
5512                         if (defect_list_len != 0) {
5513                                 ctl_set_invalid_field(ctsio,
5514                                                       /*sks_valid*/ 1,
5515                                                       /*command*/ 0,
5516                                                       /*field*/ 2,
5517                                                       /*bit_valid*/ 0,
5518                                                       /*bit*/ 0);
5519                                 goto bailout;
5520                         }
5521                 }
5522         }
5523
5524         ctl_set_success(ctsio);
5525 bailout:
5526
5527         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5528                 free(ctsio->kern_data_ptr, M_CTL);
5529                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5530         }
5531
5532         ctl_done((union ctl_io *)ctsio);
5533         return (CTL_RETVAL_COMPLETE);
5534 }
5535
5536 int
5537 ctl_read_buffer(struct ctl_scsiio *ctsio)
5538 {
5539         struct ctl_lun *lun = CTL_LUN(ctsio);
5540         uint64_t buffer_offset;
5541         uint32_t len;
5542         uint8_t byte2;
5543         static uint8_t descr[4];
5544         static uint8_t echo_descr[4] = { 0 };
5545
5546         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5547
5548         switch (ctsio->cdb[0]) {
5549         case READ_BUFFER: {
5550                 struct scsi_read_buffer *cdb;
5551
5552                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5553                 buffer_offset = scsi_3btoul(cdb->offset);
5554                 len = scsi_3btoul(cdb->length);
5555                 byte2 = cdb->byte2;
5556                 break;
5557         }
5558         case READ_BUFFER_16: {
5559                 struct scsi_read_buffer_16 *cdb;
5560
5561                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5562                 buffer_offset = scsi_8btou64(cdb->offset);
5563                 len = scsi_4btoul(cdb->length);
5564                 byte2 = cdb->byte2;
5565                 break;
5566         }
5567         default: /* This shouldn't happen. */
5568                 ctl_set_invalid_opcode(ctsio);
5569                 ctl_done((union ctl_io *)ctsio);
5570                 return (CTL_RETVAL_COMPLETE);
5571         }
5572
5573         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5574             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5575                 ctl_set_invalid_field(ctsio,
5576                                       /*sks_valid*/ 1,
5577                                       /*command*/ 1,
5578                                       /*field*/ 6,
5579                                       /*bit_valid*/ 0,
5580                                       /*bit*/ 0);
5581                 ctl_done((union ctl_io *)ctsio);
5582                 return (CTL_RETVAL_COMPLETE);
5583         }
5584
5585         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5586                 descr[0] = 0;
5587                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5588                 ctsio->kern_data_ptr = descr;
5589                 len = min(len, sizeof(descr));
5590         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5591                 ctsio->kern_data_ptr = echo_descr;
5592                 len = min(len, sizeof(echo_descr));
5593         } else {
5594                 if (lun->write_buffer == NULL) {
5595                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5596                             M_CTL, M_WAITOK);
5597                 }
5598                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5599         }
5600         ctsio->kern_data_len = len;
5601         ctsio->kern_total_len = len;
5602         ctsio->kern_rel_offset = 0;
5603         ctsio->kern_sg_entries = 0;
5604         ctl_set_success(ctsio);
5605         ctsio->be_move_done = ctl_config_move_done;
5606         ctl_datamove((union ctl_io *)ctsio);
5607         return (CTL_RETVAL_COMPLETE);
5608 }
5609
5610 int
5611 ctl_write_buffer(struct ctl_scsiio *ctsio)
5612 {
5613         struct ctl_lun *lun = CTL_LUN(ctsio);
5614         struct scsi_write_buffer *cdb;
5615         int buffer_offset, len;
5616
5617         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5618
5619         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5620
5621         len = scsi_3btoul(cdb->length);
5622         buffer_offset = scsi_3btoul(cdb->offset);
5623
5624         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5625                 ctl_set_invalid_field(ctsio,
5626                                       /*sks_valid*/ 1,
5627                                       /*command*/ 1,
5628                                       /*field*/ 6,
5629                                       /*bit_valid*/ 0,
5630                                       /*bit*/ 0);
5631                 ctl_done((union ctl_io *)ctsio);
5632                 return (CTL_RETVAL_COMPLETE);
5633         }
5634
5635         /*
5636          * If we've got a kernel request that hasn't been malloced yet,
5637          * malloc it and tell the caller the data buffer is here.
5638          */
5639         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5640                 if (lun->write_buffer == NULL) {
5641                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5642                             M_CTL, M_WAITOK);
5643                 }
5644                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5645                 ctsio->kern_data_len = len;
5646                 ctsio->kern_total_len = len;
5647                 ctsio->kern_rel_offset = 0;
5648                 ctsio->kern_sg_entries = 0;
5649                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5650                 ctsio->be_move_done = ctl_config_move_done;
5651                 ctl_datamove((union ctl_io *)ctsio);
5652
5653                 return (CTL_RETVAL_COMPLETE);
5654         }
5655
5656         ctl_set_success(ctsio);
5657         ctl_done((union ctl_io *)ctsio);
5658         return (CTL_RETVAL_COMPLETE);
5659 }
5660
5661 int
5662 ctl_write_same(struct ctl_scsiio *ctsio)
5663 {
5664         struct ctl_lun *lun = CTL_LUN(ctsio);
5665         struct ctl_lba_len_flags *lbalen;
5666         uint64_t lba;
5667         uint32_t num_blocks;
5668         int len, retval;
5669         uint8_t byte2;
5670
5671         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5672
5673         switch (ctsio->cdb[0]) {
5674         case WRITE_SAME_10: {
5675                 struct scsi_write_same_10 *cdb;
5676
5677                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5678
5679                 lba = scsi_4btoul(cdb->addr);
5680                 num_blocks = scsi_2btoul(cdb->length);
5681                 byte2 = cdb->byte2;
5682                 break;
5683         }
5684         case WRITE_SAME_16: {
5685                 struct scsi_write_same_16 *cdb;
5686
5687                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5688
5689                 lba = scsi_8btou64(cdb->addr);
5690                 num_blocks = scsi_4btoul(cdb->length);
5691                 byte2 = cdb->byte2;
5692                 break;
5693         }
5694         default:
5695                 /*
5696                  * We got a command we don't support.  This shouldn't
5697                  * happen, commands should be filtered out above us.
5698                  */
5699                 ctl_set_invalid_opcode(ctsio);
5700                 ctl_done((union ctl_io *)ctsio);
5701
5702                 return (CTL_RETVAL_COMPLETE);
5703                 break; /* NOTREACHED */
5704         }
5705
5706         /* ANCHOR flag can be used only together with UNMAP */
5707         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5708                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5709                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5710                 ctl_done((union ctl_io *)ctsio);
5711                 return (CTL_RETVAL_COMPLETE);
5712         }
5713
5714         /*
5715          * The first check is to make sure we're in bounds, the second
5716          * check is to catch wrap-around problems.  If the lba + num blocks
5717          * is less than the lba, then we've wrapped around and the block
5718          * range is invalid anyway.
5719          */
5720         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5721          || ((lba + num_blocks) < lba)) {
5722                 ctl_set_lba_out_of_range(ctsio,
5723                     MAX(lba, lun->be_lun->maxlba + 1));
5724                 ctl_done((union ctl_io *)ctsio);
5725                 return (CTL_RETVAL_COMPLETE);
5726         }
5727
5728         /* Zero number of blocks means "to the last logical block" */
5729         if (num_blocks == 0) {
5730                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5731                         ctl_set_invalid_field(ctsio,
5732                                               /*sks_valid*/ 0,
5733                                               /*command*/ 1,
5734                                               /*field*/ 0,
5735                                               /*bit_valid*/ 0,
5736                                               /*bit*/ 0);
5737                         ctl_done((union ctl_io *)ctsio);
5738                         return (CTL_RETVAL_COMPLETE);
5739                 }
5740                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5741         }
5742
5743         len = lun->be_lun->blocksize;
5744
5745         /*
5746          * If we've got a kernel request that hasn't been malloced yet,
5747          * malloc it and tell the caller the data buffer is here.
5748          */
5749         if ((byte2 & SWS_NDOB) == 0 &&
5750             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5751                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5752                 ctsio->kern_data_len = len;
5753                 ctsio->kern_total_len = len;
5754                 ctsio->kern_rel_offset = 0;
5755                 ctsio->kern_sg_entries = 0;
5756                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5757                 ctsio->be_move_done = ctl_config_move_done;
5758                 ctl_datamove((union ctl_io *)ctsio);
5759
5760                 return (CTL_RETVAL_COMPLETE);
5761         }
5762
5763         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5764         lbalen->lba = lba;
5765         lbalen->len = num_blocks;
5766         lbalen->flags = byte2;
5767         retval = lun->backend->config_write((union ctl_io *)ctsio);
5768
5769         return (retval);
5770 }
5771
5772 int
5773 ctl_unmap(struct ctl_scsiio *ctsio)
5774 {
5775         struct ctl_lun *lun = CTL_LUN(ctsio);
5776         struct scsi_unmap *cdb;
5777         struct ctl_ptr_len_flags *ptrlen;
5778         struct scsi_unmap_header *hdr;
5779         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5780         uint64_t lba;
5781         uint32_t num_blocks;
5782         int len, retval;
5783         uint8_t byte2;
5784
5785         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5786
5787         cdb = (struct scsi_unmap *)ctsio->cdb;
5788         len = scsi_2btoul(cdb->length);
5789         byte2 = cdb->byte2;
5790
5791         /*
5792          * If we've got a kernel request that hasn't been malloced yet,
5793          * malloc it and tell the caller the data buffer is here.
5794          */
5795         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5796                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5797                 ctsio->kern_data_len = len;
5798                 ctsio->kern_total_len = len;
5799                 ctsio->kern_rel_offset = 0;
5800                 ctsio->kern_sg_entries = 0;
5801                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5802                 ctsio->be_move_done = ctl_config_move_done;
5803                 ctl_datamove((union ctl_io *)ctsio);
5804
5805                 return (CTL_RETVAL_COMPLETE);
5806         }
5807
5808         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5809         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5810         if (len < sizeof (*hdr) ||
5811             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5812             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5813             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5814                 ctl_set_invalid_field(ctsio,
5815                                       /*sks_valid*/ 0,
5816                                       /*command*/ 0,
5817                                       /*field*/ 0,
5818                                       /*bit_valid*/ 0,
5819                                       /*bit*/ 0);
5820                 goto done;
5821         }
5822         len = scsi_2btoul(hdr->desc_length);
5823         buf = (struct scsi_unmap_desc *)(hdr + 1);
5824         end = buf + len / sizeof(*buf);
5825
5826         endnz = buf;
5827         for (range = buf; range < end; range++) {
5828                 lba = scsi_8btou64(range->lba);
5829                 num_blocks = scsi_4btoul(range->length);
5830                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5831                  || ((lba + num_blocks) < lba)) {
5832                         ctl_set_lba_out_of_range(ctsio,
5833                             MAX(lba, lun->be_lun->maxlba + 1));
5834                         ctl_done((union ctl_io *)ctsio);
5835                         return (CTL_RETVAL_COMPLETE);
5836                 }
5837                 if (num_blocks != 0)
5838                         endnz = range + 1;
5839         }
5840
5841         /*
5842          * Block backend can not handle zero last range.
5843          * Filter it out and return if there is nothing left.
5844          */
5845         len = (uint8_t *)endnz - (uint8_t *)buf;
5846         if (len == 0) {
5847                 ctl_set_success(ctsio);
5848                 goto done;
5849         }
5850
5851         mtx_lock(&lun->lun_lock);
5852         ptrlen = (struct ctl_ptr_len_flags *)
5853             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5854         ptrlen->ptr = (void *)buf;
5855         ptrlen->len = len;
5856         ptrlen->flags = byte2;
5857         ctl_check_blocked(lun);
5858         mtx_unlock(&lun->lun_lock);
5859
5860         retval = lun->backend->config_write((union ctl_io *)ctsio);
5861         return (retval);
5862
5863 done:
5864         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5865                 free(ctsio->kern_data_ptr, M_CTL);
5866                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5867         }
5868         ctl_done((union ctl_io *)ctsio);
5869         return (CTL_RETVAL_COMPLETE);
5870 }
5871
5872 int
5873 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5874                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5875 {
5876         struct ctl_lun *lun = CTL_LUN(ctsio);
5877         uint8_t *current_cp;
5878         int set_ua;
5879         uint32_t initidx;
5880
5881         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5882         set_ua = 0;
5883
5884         current_cp = (page_index->page_data + (page_index->page_len *
5885             CTL_PAGE_CURRENT));
5886
5887         mtx_lock(&lun->lun_lock);
5888         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5889                 memcpy(current_cp, page_ptr, page_index->page_len);
5890                 set_ua = 1;
5891         }
5892         if (set_ua != 0)
5893                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5894         mtx_unlock(&lun->lun_lock);
5895         if (set_ua) {
5896                 ctl_isc_announce_mode(lun,
5897                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5898                     page_index->page_code, page_index->subpage);
5899         }
5900         return (CTL_RETVAL_COMPLETE);
5901 }
5902
5903 static void
5904 ctl_ie_timer(void *arg)
5905 {
5906         struct ctl_lun *lun = arg;
5907         uint64_t t;
5908
5909         if (lun->ie_asc == 0)
5910                 return;
5911
5912         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5913                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5914         else
5915                 lun->ie_reported = 0;
5916
5917         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5918                 lun->ie_reportcnt++;
5919                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5920                 if (t == 0 || t == UINT32_MAX)
5921                         t = 3000;  /* 5 min */
5922                 callout_schedule(&lun->ie_callout, t * hz / 10);
5923         }
5924 }
5925
5926 int
5927 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5928                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5929 {
5930         struct ctl_lun *lun = CTL_LUN(ctsio);
5931         struct scsi_info_exceptions_page *pg;
5932         uint64_t t;
5933
5934         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5935
5936         pg = (struct scsi_info_exceptions_page *)page_ptr;
5937         mtx_lock(&lun->lun_lock);
5938         if (pg->info_flags & SIEP_FLAGS_TEST) {
5939                 lun->ie_asc = 0x5d;
5940                 lun->ie_ascq = 0xff;
5941                 if (pg->mrie == SIEP_MRIE_UA) {
5942                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
5943                         lun->ie_reported = 1;
5944                 } else {
5945                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5946                         lun->ie_reported = -1;
5947                 }
5948                 lun->ie_reportcnt = 1;
5949                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
5950                         lun->ie_reportcnt++;
5951                         t = scsi_4btoul(pg->interval_timer);
5952                         if (t == 0 || t == UINT32_MAX)
5953                                 t = 3000;  /* 5 min */
5954                         callout_reset(&lun->ie_callout, t * hz / 10,
5955                             ctl_ie_timer, lun);
5956                 }
5957         } else {
5958                 lun->ie_asc = 0;
5959                 lun->ie_ascq = 0;
5960                 lun->ie_reported = 1;
5961                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5962                 lun->ie_reportcnt = UINT32_MAX;
5963                 callout_stop(&lun->ie_callout);
5964         }
5965         mtx_unlock(&lun->lun_lock);
5966         return (CTL_RETVAL_COMPLETE);
5967 }
5968
5969 static int
5970 ctl_do_mode_select(union ctl_io *io)
5971 {
5972         struct ctl_lun *lun = CTL_LUN(io);
5973         struct scsi_mode_page_header *page_header;
5974         struct ctl_page_index *page_index;
5975         struct ctl_scsiio *ctsio;
5976         int page_len, page_len_offset, page_len_size;
5977         union ctl_modepage_info *modepage_info;
5978         uint16_t *len_left, *len_used;
5979         int retval, i;
5980
5981         ctsio = &io->scsiio;
5982         page_index = NULL;
5983         page_len = 0;
5984
5985         modepage_info = (union ctl_modepage_info *)
5986                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5987         len_left = &modepage_info->header.len_left;
5988         len_used = &modepage_info->header.len_used;
5989
5990 do_next_page:
5991
5992         page_header = (struct scsi_mode_page_header *)
5993                 (ctsio->kern_data_ptr + *len_used);
5994
5995         if (*len_left == 0) {
5996                 free(ctsio->kern_data_ptr, M_CTL);
5997                 ctl_set_success(ctsio);
5998                 ctl_done((union ctl_io *)ctsio);
5999                 return (CTL_RETVAL_COMPLETE);
6000         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6001
6002                 free(ctsio->kern_data_ptr, M_CTL);
6003                 ctl_set_param_len_error(ctsio);
6004                 ctl_done((union ctl_io *)ctsio);
6005                 return (CTL_RETVAL_COMPLETE);
6006
6007         } else if ((page_header->page_code & SMPH_SPF)
6008                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6009
6010                 free(ctsio->kern_data_ptr, M_CTL);
6011                 ctl_set_param_len_error(ctsio);
6012                 ctl_done((union ctl_io *)ctsio);
6013                 return (CTL_RETVAL_COMPLETE);
6014         }
6015
6016
6017         /*
6018          * XXX KDM should we do something with the block descriptor?
6019          */
6020         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6021                 page_index = &lun->mode_pages.index[i];
6022                 if (lun->be_lun->lun_type == T_DIRECT &&
6023                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6024                         continue;
6025                 if (lun->be_lun->lun_type == T_PROCESSOR &&
6026                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6027                         continue;
6028                 if (lun->be_lun->lun_type == T_CDROM &&
6029                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6030                         continue;
6031
6032                 if ((page_index->page_code & SMPH_PC_MASK) !=
6033                     (page_header->page_code & SMPH_PC_MASK))
6034                         continue;
6035
6036                 /*
6037                  * If neither page has a subpage code, then we've got a
6038                  * match.
6039                  */
6040                 if (((page_index->page_code & SMPH_SPF) == 0)
6041                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6042                         page_len = page_header->page_length;
6043                         break;
6044                 }
6045
6046                 /*
6047                  * If both pages have subpages, then the subpage numbers
6048                  * have to match.
6049                  */
6050                 if ((page_index->page_code & SMPH_SPF)
6051                   && (page_header->page_code & SMPH_SPF)) {
6052                         struct scsi_mode_page_header_sp *sph;
6053
6054                         sph = (struct scsi_mode_page_header_sp *)page_header;
6055                         if (page_index->subpage == sph->subpage) {
6056                                 page_len = scsi_2btoul(sph->page_length);
6057                                 break;
6058                         }
6059                 }
6060         }
6061
6062         /*
6063          * If we couldn't find the page, or if we don't have a mode select
6064          * handler for it, send back an error to the user.
6065          */
6066         if ((i >= CTL_NUM_MODE_PAGES)
6067          || (page_index->select_handler == NULL)) {
6068                 ctl_set_invalid_field(ctsio,
6069                                       /*sks_valid*/ 1,
6070                                       /*command*/ 0,
6071                                       /*field*/ *len_used,
6072                                       /*bit_valid*/ 0,
6073                                       /*bit*/ 0);
6074                 free(ctsio->kern_data_ptr, M_CTL);
6075                 ctl_done((union ctl_io *)ctsio);
6076                 return (CTL_RETVAL_COMPLETE);
6077         }
6078
6079         if (page_index->page_code & SMPH_SPF) {
6080                 page_len_offset = 2;
6081                 page_len_size = 2;
6082         } else {
6083                 page_len_size = 1;
6084                 page_len_offset = 1;
6085         }
6086
6087         /*
6088          * If the length the initiator gives us isn't the one we specify in
6089          * the mode page header, or if they didn't specify enough data in
6090          * the CDB to avoid truncating this page, kick out the request.
6091          */
6092         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6093                 ctl_set_invalid_field(ctsio,
6094                                       /*sks_valid*/ 1,
6095                                       /*command*/ 0,
6096                                       /*field*/ *len_used + page_len_offset,
6097                                       /*bit_valid*/ 0,
6098                                       /*bit*/ 0);
6099                 free(ctsio->kern_data_ptr, M_CTL);
6100                 ctl_done((union ctl_io *)ctsio);
6101                 return (CTL_RETVAL_COMPLETE);
6102         }
6103         if (*len_left < page_index->page_len) {
6104                 free(ctsio->kern_data_ptr, M_CTL);
6105                 ctl_set_param_len_error(ctsio);
6106                 ctl_done((union ctl_io *)ctsio);
6107                 return (CTL_RETVAL_COMPLETE);
6108         }
6109
6110         /*
6111          * Run through the mode page, checking to make sure that the bits
6112          * the user changed are actually legal for him to change.
6113          */
6114         for (i = 0; i < page_index->page_len; i++) {
6115                 uint8_t *user_byte, *change_mask, *current_byte;
6116                 int bad_bit;
6117                 int j;
6118
6119                 user_byte = (uint8_t *)page_header + i;
6120                 change_mask = page_index->page_data +
6121                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6122                 current_byte = page_index->page_data +
6123                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6124
6125                 /*
6126                  * Check to see whether the user set any bits in this byte
6127                  * that he is not allowed to set.
6128                  */
6129                 if ((*user_byte & ~(*change_mask)) ==
6130                     (*current_byte & ~(*change_mask)))
6131                         continue;
6132
6133                 /*
6134                  * Go through bit by bit to determine which one is illegal.
6135                  */
6136                 bad_bit = 0;
6137                 for (j = 7; j >= 0; j--) {
6138                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6139                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6140                                 bad_bit = i;
6141                                 break;
6142                         }
6143                 }
6144                 ctl_set_invalid_field(ctsio,
6145                                       /*sks_valid*/ 1,
6146                                       /*command*/ 0,
6147                                       /*field*/ *len_used + i,
6148                                       /*bit_valid*/ 1,
6149                                       /*bit*/ bad_bit);
6150                 free(ctsio->kern_data_ptr, M_CTL);
6151                 ctl_done((union ctl_io *)ctsio);
6152                 return (CTL_RETVAL_COMPLETE);
6153         }
6154
6155         /*
6156          * Decrement these before we call the page handler, since we may
6157          * end up getting called back one way or another before the handler
6158          * returns to this context.
6159          */
6160         *len_left -= page_index->page_len;
6161         *len_used += page_index->page_len;
6162
6163         retval = page_index->select_handler(ctsio, page_index,
6164                                             (uint8_t *)page_header);
6165
6166         /*
6167          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6168          * wait until this queued command completes to finish processing
6169          * the mode page.  If it returns anything other than
6170          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6171          * already set the sense information, freed the data pointer, and
6172          * completed the io for us.
6173          */
6174         if (retval != CTL_RETVAL_COMPLETE)
6175                 goto bailout_no_done;
6176
6177         /*
6178          * If the initiator sent us more than one page, parse the next one.
6179          */
6180         if (*len_left > 0)
6181                 goto do_next_page;
6182
6183         ctl_set_success(ctsio);
6184         free(ctsio->kern_data_ptr, M_CTL);
6185         ctl_done((union ctl_io *)ctsio);
6186
6187 bailout_no_done:
6188
6189         return (CTL_RETVAL_COMPLETE);
6190
6191 }
6192
6193 int
6194 ctl_mode_select(struct ctl_scsiio *ctsio)
6195 {
6196         struct ctl_lun *lun = CTL_LUN(ctsio);
6197         union ctl_modepage_info *modepage_info;
6198         int bd_len, i, header_size, param_len, pf, rtd, sp;
6199         uint32_t initidx;
6200
6201         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6202         switch (ctsio->cdb[0]) {
6203         case MODE_SELECT_6: {
6204                 struct scsi_mode_select_6 *cdb;
6205
6206                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6207
6208                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6209                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6210                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6211                 param_len = cdb->length;
6212                 header_size = sizeof(struct scsi_mode_header_6);
6213                 break;
6214         }
6215         case MODE_SELECT_10: {
6216                 struct scsi_mode_select_10 *cdb;
6217
6218                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6219
6220                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6221                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6222                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6223                 param_len = scsi_2btoul(cdb->length);
6224                 header_size = sizeof(struct scsi_mode_header_10);
6225                 break;
6226         }
6227         default:
6228                 ctl_set_invalid_opcode(ctsio);
6229                 ctl_done((union ctl_io *)ctsio);
6230                 return (CTL_RETVAL_COMPLETE);
6231         }
6232
6233         if (rtd) {
6234                 if (param_len != 0) {
6235                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 0,
6236                             /*command*/ 1, /*field*/ 0,
6237                             /*bit_valid*/ 0, /*bit*/ 0);
6238                         ctl_done((union ctl_io *)ctsio);
6239                         return (CTL_RETVAL_COMPLETE);
6240                 }
6241
6242                 /* Revert to defaults. */
6243                 ctl_init_page_index(lun);
6244                 mtx_lock(&lun->lun_lock);
6245                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6246                 mtx_unlock(&lun->lun_lock);
6247                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6248                         ctl_isc_announce_mode(lun, -1,
6249                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
6250                             lun->mode_pages.index[i].subpage);
6251                 }
6252                 ctl_set_success(ctsio);
6253                 ctl_done((union ctl_io *)ctsio);
6254                 return (CTL_RETVAL_COMPLETE);
6255         }
6256
6257         /*
6258          * From SPC-3:
6259          * "A parameter list length of zero indicates that the Data-Out Buffer
6260          * shall be empty. This condition shall not be considered as an error."
6261          */
6262         if (param_len == 0) {
6263                 ctl_set_success(ctsio);
6264                 ctl_done((union ctl_io *)ctsio);
6265                 return (CTL_RETVAL_COMPLETE);
6266         }
6267
6268         /*
6269          * Since we'll hit this the first time through, prior to
6270          * allocation, we don't need to free a data buffer here.
6271          */
6272         if (param_len < header_size) {
6273                 ctl_set_param_len_error(ctsio);
6274                 ctl_done((union ctl_io *)ctsio);
6275                 return (CTL_RETVAL_COMPLETE);
6276         }
6277
6278         /*
6279          * Allocate the data buffer and grab the user's data.  In theory,
6280          * we shouldn't have to sanity check the parameter list length here
6281          * because the maximum size is 64K.  We should be able to malloc
6282          * that much without too many problems.
6283          */
6284         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6285                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6286                 ctsio->kern_data_len = param_len;
6287                 ctsio->kern_total_len = param_len;
6288                 ctsio->kern_rel_offset = 0;
6289                 ctsio->kern_sg_entries = 0;
6290                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6291                 ctsio->be_move_done = ctl_config_move_done;
6292                 ctl_datamove((union ctl_io *)ctsio);
6293
6294                 return (CTL_RETVAL_COMPLETE);
6295         }
6296
6297         switch (ctsio->cdb[0]) {
6298         case MODE_SELECT_6: {
6299                 struct scsi_mode_header_6 *mh6;
6300
6301                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6302                 bd_len = mh6->blk_desc_len;
6303                 break;
6304         }
6305         case MODE_SELECT_10: {
6306                 struct scsi_mode_header_10 *mh10;
6307
6308                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6309                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6310                 break;
6311         }
6312         default:
6313                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6314         }
6315
6316         if (param_len < (header_size + bd_len)) {
6317                 free(ctsio->kern_data_ptr, M_CTL);
6318                 ctl_set_param_len_error(ctsio);
6319                 ctl_done((union ctl_io *)ctsio);
6320                 return (CTL_RETVAL_COMPLETE);
6321         }
6322
6323         /*
6324          * Set the IO_CONT flag, so that if this I/O gets passed to
6325          * ctl_config_write_done(), it'll get passed back to
6326          * ctl_do_mode_select() for further processing, or completion if
6327          * we're all done.
6328          */
6329         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6330         ctsio->io_cont = ctl_do_mode_select;
6331
6332         modepage_info = (union ctl_modepage_info *)
6333                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6334         memset(modepage_info, 0, sizeof(*modepage_info));
6335         modepage_info->header.len_left = param_len - header_size - bd_len;
6336         modepage_info->header.len_used = header_size + bd_len;
6337
6338         return (ctl_do_mode_select((union ctl_io *)ctsio));
6339 }
6340
6341 int
6342 ctl_mode_sense(struct ctl_scsiio *ctsio)
6343 {
6344         struct ctl_lun *lun = CTL_LUN(ctsio);
6345         int pc, page_code, dbd, llba, subpage;
6346         int alloc_len, page_len, header_len, total_len;
6347         struct scsi_mode_block_descr *block_desc;
6348         struct ctl_page_index *page_index;
6349
6350         dbd = 0;
6351         llba = 0;
6352         block_desc = NULL;
6353
6354         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6355
6356         switch (ctsio->cdb[0]) {
6357         case MODE_SENSE_6: {
6358                 struct scsi_mode_sense_6 *cdb;
6359
6360                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6361
6362                 header_len = sizeof(struct scsi_mode_hdr_6);
6363                 if (cdb->byte2 & SMS_DBD)
6364                         dbd = 1;
6365                 else
6366                         header_len += sizeof(struct scsi_mode_block_descr);
6367
6368                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6369                 page_code = cdb->page & SMS_PAGE_CODE;
6370                 subpage = cdb->subpage;
6371                 alloc_len = cdb->length;
6372                 break;
6373         }
6374         case MODE_SENSE_10: {
6375                 struct scsi_mode_sense_10 *cdb;
6376
6377                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6378
6379                 header_len = sizeof(struct scsi_mode_hdr_10);
6380
6381                 if (cdb->byte2 & SMS_DBD)
6382                         dbd = 1;
6383                 else
6384                         header_len += sizeof(struct scsi_mode_block_descr);
6385                 if (cdb->byte2 & SMS10_LLBAA)
6386                         llba = 1;
6387                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6388                 page_code = cdb->page & SMS_PAGE_CODE;
6389                 subpage = cdb->subpage;
6390                 alloc_len = scsi_2btoul(cdb->length);
6391                 break;
6392         }
6393         default:
6394                 ctl_set_invalid_opcode(ctsio);
6395                 ctl_done((union ctl_io *)ctsio);
6396                 return (CTL_RETVAL_COMPLETE);
6397                 break; /* NOTREACHED */
6398         }
6399
6400         /*
6401          * We have to make a first pass through to calculate the size of
6402          * the pages that match the user's query.  Then we allocate enough
6403          * memory to hold it, and actually copy the data into the buffer.
6404          */
6405         switch (page_code) {
6406         case SMS_ALL_PAGES_PAGE: {
6407                 u_int i;
6408
6409                 page_len = 0;
6410
6411                 /*
6412                  * At the moment, values other than 0 and 0xff here are
6413                  * reserved according to SPC-3.
6414                  */
6415                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6416                  && (subpage != SMS_SUBPAGE_ALL)) {
6417                         ctl_set_invalid_field(ctsio,
6418                                               /*sks_valid*/ 1,
6419                                               /*command*/ 1,
6420                                               /*field*/ 3,
6421                                               /*bit_valid*/ 0,
6422                                               /*bit*/ 0);
6423                         ctl_done((union ctl_io *)ctsio);
6424                         return (CTL_RETVAL_COMPLETE);
6425                 }
6426
6427                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6428                         page_index = &lun->mode_pages.index[i];
6429
6430                         /* Make sure the page is supported for this dev type */
6431                         if (lun->be_lun->lun_type == T_DIRECT &&
6432                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6433                                 continue;
6434                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6435                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6436                                 continue;
6437                         if (lun->be_lun->lun_type == T_CDROM &&
6438                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6439                                 continue;
6440
6441                         /*
6442                          * We don't use this subpage if the user didn't
6443                          * request all subpages.
6444                          */
6445                         if ((page_index->subpage != 0)
6446                          && (subpage == SMS_SUBPAGE_PAGE_0))
6447                                 continue;
6448
6449 #if 0
6450                         printf("found page %#x len %d\n",
6451                                page_index->page_code & SMPH_PC_MASK,
6452                                page_index->page_len);
6453 #endif
6454                         page_len += page_index->page_len;
6455                 }
6456                 break;
6457         }
6458         default: {
6459                 u_int i;
6460
6461                 page_len = 0;
6462
6463                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6464                         page_index = &lun->mode_pages.index[i];
6465
6466                         /* Make sure the page is supported for this dev type */
6467                         if (lun->be_lun->lun_type == T_DIRECT &&
6468                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6469                                 continue;
6470                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6471                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6472                                 continue;
6473                         if (lun->be_lun->lun_type == T_CDROM &&
6474                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6475                                 continue;
6476
6477                         /* Look for the right page code */
6478                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6479                                 continue;
6480
6481                         /* Look for the right subpage or the subpage wildcard*/
6482                         if ((page_index->subpage != subpage)
6483                          && (subpage != SMS_SUBPAGE_ALL))
6484                                 continue;
6485
6486 #if 0
6487                         printf("found page %#x len %d\n",
6488                                page_index->page_code & SMPH_PC_MASK,
6489                                page_index->page_len);
6490 #endif
6491
6492                         page_len += page_index->page_len;
6493                 }
6494
6495                 if (page_len == 0) {
6496                         ctl_set_invalid_field(ctsio,
6497                                               /*sks_valid*/ 1,
6498                                               /*command*/ 1,
6499                                               /*field*/ 2,
6500                                               /*bit_valid*/ 1,
6501                                               /*bit*/ 5);
6502                         ctl_done((union ctl_io *)ctsio);
6503                         return (CTL_RETVAL_COMPLETE);
6504                 }
6505                 break;
6506         }
6507         }
6508
6509         total_len = header_len + page_len;
6510 #if 0
6511         printf("header_len = %d, page_len = %d, total_len = %d\n",
6512                header_len, page_len, total_len);
6513 #endif
6514
6515         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6516         ctsio->kern_sg_entries = 0;
6517         ctsio->kern_rel_offset = 0;
6518         ctsio->kern_data_len = min(total_len, alloc_len);
6519         ctsio->kern_total_len = ctsio->kern_data_len;
6520
6521         switch (ctsio->cdb[0]) {
6522         case MODE_SENSE_6: {
6523                 struct scsi_mode_hdr_6 *header;
6524
6525                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6526
6527                 header->datalen = MIN(total_len - 1, 254);
6528                 if (lun->be_lun->lun_type == T_DIRECT) {
6529                         header->dev_specific = 0x10; /* DPOFUA */
6530                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6531                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6532                                 header->dev_specific |= 0x80; /* WP */
6533                 }
6534                 if (dbd)
6535                         header->block_descr_len = 0;
6536                 else
6537                         header->block_descr_len =
6538                                 sizeof(struct scsi_mode_block_descr);
6539                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6540                 break;
6541         }
6542         case MODE_SENSE_10: {
6543                 struct scsi_mode_hdr_10 *header;
6544                 int datalen;
6545
6546                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6547
6548                 datalen = MIN(total_len - 2, 65533);
6549                 scsi_ulto2b(datalen, header->datalen);
6550                 if (lun->be_lun->lun_type == T_DIRECT) {
6551                         header->dev_specific = 0x10; /* DPOFUA */
6552                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6553                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6554                                 header->dev_specific |= 0x80; /* WP */
6555                 }
6556                 if (dbd)
6557                         scsi_ulto2b(0, header->block_descr_len);
6558                 else
6559                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6560                                     header->block_descr_len);
6561                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6562                 break;
6563         }
6564         default:
6565                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6566         }
6567
6568         /*
6569          * If we've got a disk, use its blocksize in the block
6570          * descriptor.  Otherwise, just set it to 0.
6571          */
6572         if (dbd == 0) {
6573                 if (lun->be_lun->lun_type == T_DIRECT)
6574                         scsi_ulto3b(lun->be_lun->blocksize,
6575                                     block_desc->block_len);
6576                 else
6577                         scsi_ulto3b(0, block_desc->block_len);
6578         }
6579
6580         switch (page_code) {
6581         case SMS_ALL_PAGES_PAGE: {
6582                 int i, data_used;
6583
6584                 data_used = header_len;
6585                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6586                         struct ctl_page_index *page_index;
6587
6588                         page_index = &lun->mode_pages.index[i];
6589                         if (lun->be_lun->lun_type == T_DIRECT &&
6590                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6591                                 continue;
6592                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6593                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6594                                 continue;
6595                         if (lun->be_lun->lun_type == T_CDROM &&
6596                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6597                                 continue;
6598
6599                         /*
6600                          * We don't use this subpage if the user didn't
6601                          * request all subpages.  We already checked (above)
6602                          * to make sure the user only specified a subpage
6603                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6604                          */
6605                         if ((page_index->subpage != 0)
6606                          && (subpage == SMS_SUBPAGE_PAGE_0))
6607                                 continue;
6608
6609                         /*
6610                          * Call the handler, if it exists, to update the
6611                          * page to the latest values.
6612                          */
6613                         if (page_index->sense_handler != NULL)
6614                                 page_index->sense_handler(ctsio, page_index,pc);
6615
6616                         memcpy(ctsio->kern_data_ptr + data_used,
6617                                page_index->page_data +
6618                                (page_index->page_len * pc),
6619                                page_index->page_len);
6620                         data_used += page_index->page_len;
6621                 }
6622                 break;
6623         }
6624         default: {
6625                 int i, data_used;
6626
6627                 data_used = header_len;
6628
6629                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6630                         struct ctl_page_index *page_index;
6631
6632                         page_index = &lun->mode_pages.index[i];
6633
6634                         /* Look for the right page code */
6635                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6636                                 continue;
6637
6638                         /* Look for the right subpage or the subpage wildcard*/
6639                         if ((page_index->subpage != subpage)
6640                          && (subpage != SMS_SUBPAGE_ALL))
6641                                 continue;
6642
6643                         /* Make sure the page is supported for this dev type */
6644                         if (lun->be_lun->lun_type == T_DIRECT &&
6645                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6646                                 continue;
6647                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6648                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6649                                 continue;
6650                         if (lun->be_lun->lun_type == T_CDROM &&
6651                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6652                                 continue;
6653
6654                         /*
6655                          * Call the handler, if it exists, to update the
6656                          * page to the latest values.
6657                          */
6658                         if (page_index->sense_handler != NULL)
6659                                 page_index->sense_handler(ctsio, page_index,pc);
6660
6661                         memcpy(ctsio->kern_data_ptr + data_used,
6662                                page_index->page_data +
6663                                (page_index->page_len * pc),
6664                                page_index->page_len);
6665                         data_used += page_index->page_len;
6666                 }
6667                 break;
6668         }
6669         }
6670
6671         ctl_set_success(ctsio);
6672         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6673         ctsio->be_move_done = ctl_config_move_done;
6674         ctl_datamove((union ctl_io *)ctsio);
6675         return (CTL_RETVAL_COMPLETE);
6676 }
6677
6678 int
6679 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6680                                struct ctl_page_index *page_index,
6681                                int pc)
6682 {
6683         struct ctl_lun *lun = CTL_LUN(ctsio);
6684         struct scsi_log_param_header *phdr;
6685         uint8_t *data;
6686         uint64_t val;
6687
6688         data = page_index->page_data;
6689
6690         if (lun->backend->lun_attr != NULL &&
6691             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6692              != UINT64_MAX) {
6693                 phdr = (struct scsi_log_param_header *)data;
6694                 scsi_ulto2b(0x0001, phdr->param_code);
6695                 phdr->param_control = SLP_LBIN | SLP_LP;
6696                 phdr->param_len = 8;
6697                 data = (uint8_t *)(phdr + 1);
6698                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6699                 data[4] = 0x02; /* per-pool */
6700                 data += phdr->param_len;
6701         }
6702
6703         if (lun->backend->lun_attr != NULL &&
6704             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6705              != UINT64_MAX) {
6706                 phdr = (struct scsi_log_param_header *)data;
6707                 scsi_ulto2b(0x0002, phdr->param_code);
6708                 phdr->param_control = SLP_LBIN | SLP_LP;
6709                 phdr->param_len = 8;
6710                 data = (uint8_t *)(phdr + 1);
6711                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6712                 data[4] = 0x01; /* per-LUN */
6713                 data += phdr->param_len;
6714         }
6715
6716         if (lun->backend->lun_attr != NULL &&
6717             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6718              != UINT64_MAX) {
6719                 phdr = (struct scsi_log_param_header *)data;
6720                 scsi_ulto2b(0x00f1, phdr->param_code);
6721                 phdr->param_control = SLP_LBIN | SLP_LP;
6722                 phdr->param_len = 8;
6723                 data = (uint8_t *)(phdr + 1);
6724                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6725                 data[4] = 0x02; /* per-pool */
6726                 data += phdr->param_len;
6727         }
6728
6729         if (lun->backend->lun_attr != NULL &&
6730             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6731              != UINT64_MAX) {
6732                 phdr = (struct scsi_log_param_header *)data;
6733                 scsi_ulto2b(0x00f2, phdr->param_code);
6734                 phdr->param_control = SLP_LBIN | SLP_LP;
6735                 phdr->param_len = 8;
6736                 data = (uint8_t *)(phdr + 1);
6737                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6738                 data[4] = 0x02; /* per-pool */
6739                 data += phdr->param_len;
6740         }
6741
6742         page_index->page_len = data - page_index->page_data;
6743         return (0);
6744 }
6745
6746 int
6747 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6748                                struct ctl_page_index *page_index,
6749                                int pc)
6750 {
6751         struct ctl_lun *lun = CTL_LUN(ctsio);
6752         struct stat_page *data;
6753         struct bintime *t;
6754
6755         data = (struct stat_page *)page_index->page_data;
6756
6757         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6758         data->sap.hdr.param_control = SLP_LBIN;
6759         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6760             sizeof(struct scsi_log_param_header);
6761         scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
6762             data->sap.read_num);
6763         scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
6764             data->sap.write_num);
6765         if (lun->be_lun->blocksize > 0) {
6766                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
6767                     lun->be_lun->blocksize, data->sap.recvieved_lba);
6768                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
6769                     lun->be_lun->blocksize, data->sap.transmitted_lba);
6770         }
6771         t = &lun->stats.time[CTL_STATS_READ];
6772         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6773             data->sap.read_int);
6774         t = &lun->stats.time[CTL_STATS_WRITE];
6775         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6776             data->sap.write_int);
6777         scsi_u64to8b(0, data->sap.weighted_num);
6778         scsi_u64to8b(0, data->sap.weighted_int);
6779         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6780         data->it.hdr.param_control = SLP_LBIN;
6781         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6782             sizeof(struct scsi_log_param_header);
6783 #ifdef CTL_TIME_IO
6784         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6785 #endif
6786         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6787         data->it.hdr.param_control = SLP_LBIN;
6788         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6789             sizeof(struct scsi_log_param_header);
6790         scsi_ulto4b(3, data->ti.exponent);
6791         scsi_ulto4b(1, data->ti.integer);
6792         return (0);
6793 }
6794
6795 int
6796 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6797                                struct ctl_page_index *page_index,
6798                                int pc)
6799 {
6800         struct ctl_lun *lun = CTL_LUN(ctsio);
6801         struct scsi_log_informational_exceptions *data;
6802
6803         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6804
6805         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6806         data->hdr.param_control = SLP_LBIN;
6807         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6808             sizeof(struct scsi_log_param_header);
6809         data->ie_asc = lun->ie_asc;
6810         data->ie_ascq = lun->ie_ascq;
6811         data->temperature = 0xff;
6812         return (0);
6813 }
6814
6815 int
6816 ctl_log_sense(struct ctl_scsiio *ctsio)
6817 {
6818         struct ctl_lun *lun = CTL_LUN(ctsio);
6819         int i, pc, page_code, subpage;
6820         int alloc_len, total_len;
6821         struct ctl_page_index *page_index;
6822         struct scsi_log_sense *cdb;
6823         struct scsi_log_header *header;
6824
6825         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6826
6827         cdb = (struct scsi_log_sense *)ctsio->cdb;
6828         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6829         page_code = cdb->page & SLS_PAGE_CODE;
6830         subpage = cdb->subpage;
6831         alloc_len = scsi_2btoul(cdb->length);
6832
6833         page_index = NULL;
6834         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6835                 page_index = &lun->log_pages.index[i];
6836
6837                 /* Look for the right page code */
6838                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6839                         continue;
6840
6841                 /* Look for the right subpage or the subpage wildcard*/
6842                 if (page_index->subpage != subpage)
6843                         continue;
6844
6845                 break;
6846         }
6847         if (i >= CTL_NUM_LOG_PAGES) {
6848                 ctl_set_invalid_field(ctsio,
6849                                       /*sks_valid*/ 1,
6850                                       /*command*/ 1,
6851                                       /*field*/ 2,
6852                                       /*bit_valid*/ 0,
6853                                       /*bit*/ 0);
6854                 ctl_done((union ctl_io *)ctsio);
6855                 return (CTL_RETVAL_COMPLETE);
6856         }
6857
6858         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6859
6860         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6861         ctsio->kern_sg_entries = 0;
6862         ctsio->kern_rel_offset = 0;
6863         ctsio->kern_data_len = min(total_len, alloc_len);
6864         ctsio->kern_total_len = ctsio->kern_data_len;
6865
6866         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6867         header->page = page_index->page_code;
6868         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6869                 header->page |= SL_DS;
6870         if (page_index->subpage) {
6871                 header->page |= SL_SPF;
6872                 header->subpage = page_index->subpage;
6873         }
6874         scsi_ulto2b(page_index->page_len, header->datalen);
6875
6876         /*
6877          * Call the handler, if it exists, to update the
6878          * page to the latest values.
6879          */
6880         if (page_index->sense_handler != NULL)
6881                 page_index->sense_handler(ctsio, page_index, pc);
6882
6883         memcpy(header + 1, page_index->page_data, page_index->page_len);
6884
6885         ctl_set_success(ctsio);
6886         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6887         ctsio->be_move_done = ctl_config_move_done;
6888         ctl_datamove((union ctl_io *)ctsio);
6889         return (CTL_RETVAL_COMPLETE);
6890 }
6891
6892 int
6893 ctl_read_capacity(struct ctl_scsiio *ctsio)
6894 {
6895         struct ctl_lun *lun = CTL_LUN(ctsio);
6896         struct scsi_read_capacity *cdb;
6897         struct scsi_read_capacity_data *data;
6898         uint32_t lba;
6899
6900         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6901
6902         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6903
6904         lba = scsi_4btoul(cdb->addr);
6905         if (((cdb->pmi & SRC_PMI) == 0)
6906          && (lba != 0)) {
6907                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6908                                       /*sks_valid*/ 1,
6909                                       /*command*/ 1,
6910                                       /*field*/ 2,
6911                                       /*bit_valid*/ 0,
6912                                       /*bit*/ 0);
6913                 ctl_done((union ctl_io *)ctsio);
6914                 return (CTL_RETVAL_COMPLETE);
6915         }
6916
6917         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6918         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6919         ctsio->kern_data_len = sizeof(*data);
6920         ctsio->kern_total_len = sizeof(*data);
6921         ctsio->kern_rel_offset = 0;
6922         ctsio->kern_sg_entries = 0;
6923
6924         /*
6925          * If the maximum LBA is greater than 0xfffffffe, the user must
6926          * issue a SERVICE ACTION IN (16) command, with the read capacity
6927          * serivce action set.
6928          */
6929         if (lun->be_lun->maxlba > 0xfffffffe)
6930                 scsi_ulto4b(0xffffffff, data->addr);
6931         else
6932                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6933
6934         /*
6935          * XXX KDM this may not be 512 bytes...
6936          */
6937         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6938
6939         ctl_set_success(ctsio);
6940         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6941         ctsio->be_move_done = ctl_config_move_done;
6942         ctl_datamove((union ctl_io *)ctsio);
6943         return (CTL_RETVAL_COMPLETE);
6944 }
6945
6946 int
6947 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6948 {
6949         struct ctl_lun *lun = CTL_LUN(ctsio);
6950         struct scsi_read_capacity_16 *cdb;
6951         struct scsi_read_capacity_data_long *data;
6952         uint64_t lba;
6953         uint32_t alloc_len;
6954
6955         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6956
6957         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6958
6959         alloc_len = scsi_4btoul(cdb->alloc_len);
6960         lba = scsi_8btou64(cdb->addr);
6961
6962         if ((cdb->reladr & SRC16_PMI)
6963          && (lba != 0)) {
6964                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6965                                       /*sks_valid*/ 1,
6966                                       /*command*/ 1,
6967                                       /*field*/ 2,
6968                                       /*bit_valid*/ 0,
6969                                       /*bit*/ 0);
6970                 ctl_done((union ctl_io *)ctsio);
6971                 return (CTL_RETVAL_COMPLETE);
6972         }
6973
6974         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6975         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6976         ctsio->kern_rel_offset = 0;
6977         ctsio->kern_sg_entries = 0;
6978         ctsio->kern_data_len = min(sizeof(*data), alloc_len);
6979         ctsio->kern_total_len = ctsio->kern_data_len;
6980
6981         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6982         /* XXX KDM this may not be 512 bytes... */
6983         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6984         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6985         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6986         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6987                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6988
6989         ctl_set_success(ctsio);
6990         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6991         ctsio->be_move_done = ctl_config_move_done;
6992         ctl_datamove((union ctl_io *)ctsio);
6993         return (CTL_RETVAL_COMPLETE);
6994 }
6995
6996 int
6997 ctl_get_lba_status(struct ctl_scsiio *ctsio)
6998 {
6999         struct ctl_lun *lun = CTL_LUN(ctsio);
7000         struct scsi_get_lba_status *cdb;
7001         struct scsi_get_lba_status_data *data;
7002         struct ctl_lba_len_flags *lbalen;
7003         uint64_t lba;
7004         uint32_t alloc_len, total_len;
7005         int retval;
7006
7007         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7008
7009         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7010         lba = scsi_8btou64(cdb->addr);
7011         alloc_len = scsi_4btoul(cdb->alloc_len);
7012
7013         if (lba > lun->be_lun->maxlba) {
7014                 ctl_set_lba_out_of_range(ctsio, lba);
7015                 ctl_done((union ctl_io *)ctsio);
7016                 return (CTL_RETVAL_COMPLETE);
7017         }
7018
7019         total_len = sizeof(*data) + sizeof(data->descr[0]);
7020         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7021         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7022         ctsio->kern_rel_offset = 0;
7023         ctsio->kern_sg_entries = 0;
7024         ctsio->kern_data_len = min(total_len, alloc_len);
7025         ctsio->kern_total_len = ctsio->kern_data_len;
7026
7027         /* Fill dummy data in case backend can't tell anything. */
7028         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7029         scsi_u64to8b(lba, data->descr[0].addr);
7030         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7031             data->descr[0].length);
7032         data->descr[0].status = 0; /* Mapped or unknown. */
7033
7034         ctl_set_success(ctsio);
7035         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7036         ctsio->be_move_done = ctl_config_move_done;
7037
7038         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7039         lbalen->lba = lba;
7040         lbalen->len = total_len;
7041         lbalen->flags = 0;
7042         retval = lun->backend->config_read((union ctl_io *)ctsio);
7043         return (retval);
7044 }
7045
7046 int
7047 ctl_read_defect(struct ctl_scsiio *ctsio)
7048 {
7049         struct scsi_read_defect_data_10 *ccb10;
7050         struct scsi_read_defect_data_12 *ccb12;
7051         struct scsi_read_defect_data_hdr_10 *data10;
7052         struct scsi_read_defect_data_hdr_12 *data12;
7053         uint32_t alloc_len, data_len;
7054         uint8_t format;
7055
7056         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7057
7058         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7059                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7060                 format = ccb10->format;
7061                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7062                 data_len = sizeof(*data10);
7063         } else {
7064                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7065                 format = ccb12->format;
7066                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7067                 data_len = sizeof(*data12);
7068         }
7069         if (alloc_len == 0) {
7070                 ctl_set_success(ctsio);
7071                 ctl_done((union ctl_io *)ctsio);
7072                 return (CTL_RETVAL_COMPLETE);
7073         }
7074
7075         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7076         ctsio->kern_rel_offset = 0;
7077         ctsio->kern_sg_entries = 0;
7078         ctsio->kern_data_len = min(data_len, alloc_len);
7079         ctsio->kern_total_len = ctsio->kern_data_len;
7080
7081         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7082                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7083                     ctsio->kern_data_ptr;
7084                 data10->format = format;
7085                 scsi_ulto2b(0, data10->length);
7086         } else {
7087                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7088                     ctsio->kern_data_ptr;
7089                 data12->format = format;
7090                 scsi_ulto2b(0, data12->generation);
7091                 scsi_ulto4b(0, data12->length);
7092         }
7093
7094         ctl_set_success(ctsio);
7095         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7096         ctsio->be_move_done = ctl_config_move_done;
7097         ctl_datamove((union ctl_io *)ctsio);
7098         return (CTL_RETVAL_COMPLETE);
7099 }
7100
7101 int
7102 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7103 {
7104         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7105         struct ctl_lun *lun = CTL_LUN(ctsio);
7106         struct scsi_maintenance_in *cdb;
7107         int retval;
7108         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7109         int num_ha_groups, num_target_ports, shared_group;
7110         struct ctl_port *port;
7111         struct scsi_target_group_data *rtg_ptr;
7112         struct scsi_target_group_data_extended *rtg_ext_ptr;
7113         struct scsi_target_port_group_descriptor *tpg_desc;
7114
7115         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7116
7117         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7118         retval = CTL_RETVAL_COMPLETE;
7119
7120         switch (cdb->byte2 & STG_PDF_MASK) {
7121         case STG_PDF_LENGTH:
7122                 ext = 0;
7123                 break;
7124         case STG_PDF_EXTENDED:
7125                 ext = 1;
7126                 break;
7127         default:
7128                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7129                                       /*sks_valid*/ 1,
7130                                       /*command*/ 1,
7131                                       /*field*/ 2,
7132                                       /*bit_valid*/ 1,
7133                                       /*bit*/ 5);
7134                 ctl_done((union ctl_io *)ctsio);
7135                 return(retval);
7136         }
7137
7138         num_target_ports = 0;
7139         shared_group = (softc->is_single != 0);
7140         mtx_lock(&softc->ctl_lock);
7141         STAILQ_FOREACH(port, &softc->port_list, links) {
7142                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7143                         continue;
7144                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7145                         continue;
7146                 num_target_ports++;
7147                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7148                         shared_group = 1;
7149         }
7150         mtx_unlock(&softc->ctl_lock);
7151         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7152
7153         if (ext)
7154                 total_len = sizeof(struct scsi_target_group_data_extended);
7155         else
7156                 total_len = sizeof(struct scsi_target_group_data);
7157         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7158                 (shared_group + num_ha_groups) +
7159             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7160
7161         alloc_len = scsi_4btoul(cdb->length);
7162
7163         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7164         ctsio->kern_sg_entries = 0;
7165         ctsio->kern_rel_offset = 0;
7166         ctsio->kern_data_len = min(total_len, alloc_len);
7167         ctsio->kern_total_len = ctsio->kern_data_len;
7168
7169         if (ext) {
7170                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7171                     ctsio->kern_data_ptr;
7172                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7173                 rtg_ext_ptr->format_type = 0x10;
7174                 rtg_ext_ptr->implicit_transition_time = 0;
7175                 tpg_desc = &rtg_ext_ptr->groups[0];
7176         } else {
7177                 rtg_ptr = (struct scsi_target_group_data *)
7178                     ctsio->kern_data_ptr;
7179                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7180                 tpg_desc = &rtg_ptr->groups[0];
7181         }
7182
7183         mtx_lock(&softc->ctl_lock);
7184         pg = softc->port_min / softc->port_cnt;
7185         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7186                 /* Some shelf is known to be primary. */
7187                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7188                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7189                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7190                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7191                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7192                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7193                 else
7194                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7195                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7196                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7197                 } else {
7198                         ts = os;
7199                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7200                 }
7201         } else {
7202                 /* No known primary shelf. */
7203                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7204                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7205                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7206                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7207                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7208                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7209                 } else {
7210                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7211                 }
7212         }
7213         if (shared_group) {
7214                 tpg_desc->pref_state = ts;
7215                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7216                     TPG_U_SUP | TPG_T_SUP;
7217                 scsi_ulto2b(1, tpg_desc->target_port_group);
7218                 tpg_desc->status = TPG_IMPLICIT;
7219                 pc = 0;
7220                 STAILQ_FOREACH(port, &softc->port_list, links) {
7221                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7222                                 continue;
7223                         if (!softc->is_single &&
7224                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7225                                 continue;
7226                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7227                                 continue;
7228                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7229                             relative_target_port_identifier);
7230                         pc++;
7231                 }
7232                 tpg_desc->target_port_count = pc;
7233                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7234                     &tpg_desc->descriptors[pc];
7235         }
7236         for (g = 0; g < num_ha_groups; g++) {
7237                 tpg_desc->pref_state = (g == pg) ? ts : os;
7238                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7239                     TPG_U_SUP | TPG_T_SUP;
7240                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7241                 tpg_desc->status = TPG_IMPLICIT;
7242                 pc = 0;
7243                 STAILQ_FOREACH(port, &softc->port_list, links) {
7244                         if (port->targ_port < g * softc->port_cnt ||
7245                             port->targ_port >= (g + 1) * softc->port_cnt)
7246                                 continue;
7247                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7248                                 continue;
7249                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7250                                 continue;
7251                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7252                                 continue;
7253                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7254                             relative_target_port_identifier);
7255                         pc++;
7256                 }
7257                 tpg_desc->target_port_count = pc;
7258                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7259                     &tpg_desc->descriptors[pc];
7260         }
7261         mtx_unlock(&softc->ctl_lock);
7262
7263         ctl_set_success(ctsio);
7264         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7265         ctsio->be_move_done = ctl_config_move_done;
7266         ctl_datamove((union ctl_io *)ctsio);
7267         return(retval);
7268 }
7269
7270 int
7271 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7272 {
7273         struct ctl_lun *lun = CTL_LUN(ctsio);
7274         struct scsi_report_supported_opcodes *cdb;
7275         const struct ctl_cmd_entry *entry, *sentry;
7276         struct scsi_report_supported_opcodes_all *all;
7277         struct scsi_report_supported_opcodes_descr *descr;
7278         struct scsi_report_supported_opcodes_one *one;
7279         int retval;
7280         int alloc_len, total_len;
7281         int opcode, service_action, i, j, num;
7282
7283         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7284
7285         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7286         retval = CTL_RETVAL_COMPLETE;
7287
7288         opcode = cdb->requested_opcode;
7289         service_action = scsi_2btoul(cdb->requested_service_action);
7290         switch (cdb->options & RSO_OPTIONS_MASK) {
7291         case RSO_OPTIONS_ALL:
7292                 num = 0;
7293                 for (i = 0; i < 256; i++) {
7294                         entry = &ctl_cmd_table[i];
7295                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7296                                 for (j = 0; j < 32; j++) {
7297                                         sentry = &((const struct ctl_cmd_entry *)
7298                                             entry->execute)[j];
7299                                         if (ctl_cmd_applicable(
7300                                             lun->be_lun->lun_type, sentry))
7301                                                 num++;
7302                                 }
7303                         } else {
7304                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7305                                     entry))
7306                                         num++;
7307                         }
7308                 }
7309                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7310                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7311                 break;
7312         case RSO_OPTIONS_OC:
7313                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7314                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7315                                               /*sks_valid*/ 1,
7316                                               /*command*/ 1,
7317                                               /*field*/ 2,
7318                                               /*bit_valid*/ 1,
7319                                               /*bit*/ 2);
7320                         ctl_done((union ctl_io *)ctsio);
7321                         return (CTL_RETVAL_COMPLETE);
7322                 }
7323                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7324                 break;
7325         case RSO_OPTIONS_OC_SA:
7326                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7327                     service_action >= 32) {
7328                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7329                                               /*sks_valid*/ 1,
7330                                               /*command*/ 1,
7331                                               /*field*/ 2,
7332                                               /*bit_valid*/ 1,
7333                                               /*bit*/ 2);
7334                         ctl_done((union ctl_io *)ctsio);
7335                         return (CTL_RETVAL_COMPLETE);
7336                 }
7337                 /* FALLTHROUGH */
7338         case RSO_OPTIONS_OC_ASA:
7339                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7340                 break;
7341         default:
7342                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7343                                       /*sks_valid*/ 1,
7344                                       /*command*/ 1,
7345                                       /*field*/ 2,
7346                                       /*bit_valid*/ 1,
7347                                       /*bit*/ 2);
7348                 ctl_done((union ctl_io *)ctsio);
7349                 return (CTL_RETVAL_COMPLETE);
7350         }
7351
7352         alloc_len = scsi_4btoul(cdb->length);
7353
7354         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7355         ctsio->kern_sg_entries = 0;
7356         ctsio->kern_rel_offset = 0;
7357         ctsio->kern_data_len = min(total_len, alloc_len);
7358         ctsio->kern_total_len = ctsio->kern_data_len;
7359
7360         switch (cdb->options & RSO_OPTIONS_MASK) {
7361         case RSO_OPTIONS_ALL:
7362                 all = (struct scsi_report_supported_opcodes_all *)
7363                     ctsio->kern_data_ptr;
7364                 num = 0;
7365                 for (i = 0; i < 256; i++) {
7366                         entry = &ctl_cmd_table[i];
7367                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7368                                 for (j = 0; j < 32; j++) {
7369                                         sentry = &((const struct ctl_cmd_entry *)
7370                                             entry->execute)[j];
7371                                         if (!ctl_cmd_applicable(
7372                                             lun->be_lun->lun_type, sentry))
7373                                                 continue;
7374                                         descr = &all->descr[num++];
7375                                         descr->opcode = i;
7376                                         scsi_ulto2b(j, descr->service_action);
7377                                         descr->flags = RSO_SERVACTV;
7378                                         scsi_ulto2b(sentry->length,
7379                                             descr->cdb_length);
7380                                 }
7381                         } else {
7382                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7383                                     entry))
7384                                         continue;
7385                                 descr = &all->descr[num++];
7386                                 descr->opcode = i;
7387                                 scsi_ulto2b(0, descr->service_action);
7388                                 descr->flags = 0;
7389                                 scsi_ulto2b(entry->length, descr->cdb_length);
7390                         }
7391                 }
7392                 scsi_ulto4b(
7393                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7394                     all->length);
7395                 break;
7396         case RSO_OPTIONS_OC:
7397                 one = (struct scsi_report_supported_opcodes_one *)
7398                     ctsio->kern_data_ptr;
7399                 entry = &ctl_cmd_table[opcode];
7400                 goto fill_one;
7401         case RSO_OPTIONS_OC_SA:
7402                 one = (struct scsi_report_supported_opcodes_one *)
7403                     ctsio->kern_data_ptr;
7404                 entry = &ctl_cmd_table[opcode];
7405                 entry = &((const struct ctl_cmd_entry *)
7406                     entry->execute)[service_action];
7407 fill_one:
7408                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7409                         one->support = 3;
7410                         scsi_ulto2b(entry->length, one->cdb_length);
7411                         one->cdb_usage[0] = opcode;
7412                         memcpy(&one->cdb_usage[1], entry->usage,
7413                             entry->length - 1);
7414                 } else
7415                         one->support = 1;
7416                 break;
7417         case RSO_OPTIONS_OC_ASA:
7418                 one = (struct scsi_report_supported_opcodes_one *)
7419                     ctsio->kern_data_ptr;
7420                 entry = &ctl_cmd_table[opcode];
7421                 if (entry->flags & CTL_CMD_FLAG_SA5) {
7422                         entry = &((const struct ctl_cmd_entry *)
7423                             entry->execute)[service_action];
7424                 } else if (service_action != 0) {
7425                         one->support = 1;
7426                         break;
7427                 }
7428                 goto fill_one;
7429         }
7430
7431         ctl_set_success(ctsio);
7432         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7433         ctsio->be_move_done = ctl_config_move_done;
7434         ctl_datamove((union ctl_io *)ctsio);
7435         return(retval);
7436 }
7437
7438 int
7439 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7440 {
7441         struct scsi_report_supported_tmf *cdb;
7442         struct scsi_report_supported_tmf_ext_data *data;
7443         int retval;
7444         int alloc_len, total_len;
7445
7446         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7447
7448         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7449
7450         retval = CTL_RETVAL_COMPLETE;
7451
7452         if (cdb->options & RST_REPD)
7453                 total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7454         else
7455                 total_len = sizeof(struct scsi_report_supported_tmf_data);
7456         alloc_len = scsi_4btoul(cdb->length);
7457
7458         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7459         ctsio->kern_sg_entries = 0;
7460         ctsio->kern_rel_offset = 0;
7461         ctsio->kern_data_len = min(total_len, alloc_len);
7462         ctsio->kern_total_len = ctsio->kern_data_len;
7463
7464         data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7465         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7466             RST_TRS;
7467         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7468         data->length = total_len - 4;
7469
7470         ctl_set_success(ctsio);
7471         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7472         ctsio->be_move_done = ctl_config_move_done;
7473         ctl_datamove((union ctl_io *)ctsio);
7474         return (retval);
7475 }
7476
7477 int
7478 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7479 {
7480         struct scsi_report_timestamp *cdb;
7481         struct scsi_report_timestamp_data *data;
7482         struct timeval tv;
7483         int64_t timestamp;
7484         int retval;
7485         int alloc_len, total_len;
7486
7487         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7488
7489         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7490
7491         retval = CTL_RETVAL_COMPLETE;
7492
7493         total_len = sizeof(struct scsi_report_timestamp_data);
7494         alloc_len = scsi_4btoul(cdb->length);
7495
7496         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7497         ctsio->kern_sg_entries = 0;
7498         ctsio->kern_rel_offset = 0;
7499         ctsio->kern_data_len = min(total_len, alloc_len);
7500         ctsio->kern_total_len = ctsio->kern_data_len;
7501
7502         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7503         scsi_ulto2b(sizeof(*data) - 2, data->length);
7504         data->origin = RTS_ORIG_OUTSIDE;
7505         getmicrotime(&tv);
7506         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7507         scsi_ulto4b(timestamp >> 16, data->timestamp);
7508         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7509
7510         ctl_set_success(ctsio);
7511         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7512         ctsio->be_move_done = ctl_config_move_done;
7513         ctl_datamove((union ctl_io *)ctsio);
7514         return (retval);
7515 }
7516
7517 int
7518 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7519 {
7520         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7521         struct ctl_lun *lun = CTL_LUN(ctsio);
7522         struct scsi_per_res_in *cdb;
7523         int alloc_len, total_len = 0;
7524         /* struct scsi_per_res_in_rsrv in_data; */
7525         uint64_t key;
7526
7527         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7528
7529         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7530
7531         alloc_len = scsi_2btoul(cdb->length);
7532
7533 retry:
7534         mtx_lock(&lun->lun_lock);
7535         switch (cdb->action) {
7536         case SPRI_RK: /* read keys */
7537                 total_len = sizeof(struct scsi_per_res_in_keys) +
7538                         lun->pr_key_count *
7539                         sizeof(struct scsi_per_res_key);
7540                 break;
7541         case SPRI_RR: /* read reservation */
7542                 if (lun->flags & CTL_LUN_PR_RESERVED)
7543                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7544                 else
7545                         total_len = sizeof(struct scsi_per_res_in_header);
7546                 break;
7547         case SPRI_RC: /* report capabilities */
7548                 total_len = sizeof(struct scsi_per_res_cap);
7549                 break;
7550         case SPRI_RS: /* read full status */
7551                 total_len = sizeof(struct scsi_per_res_in_header) +
7552                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7553                     lun->pr_key_count;
7554                 break;
7555         default:
7556                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7557         }
7558         mtx_unlock(&lun->lun_lock);
7559
7560         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7561         ctsio->kern_rel_offset = 0;
7562         ctsio->kern_sg_entries = 0;
7563         ctsio->kern_data_len = min(total_len, alloc_len);
7564         ctsio->kern_total_len = ctsio->kern_data_len;
7565
7566         mtx_lock(&lun->lun_lock);
7567         switch (cdb->action) {
7568         case SPRI_RK: { // read keys
7569         struct scsi_per_res_in_keys *res_keys;
7570                 int i, key_count;
7571
7572                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7573
7574                 /*
7575                  * We had to drop the lock to allocate our buffer, which
7576                  * leaves time for someone to come in with another
7577                  * persistent reservation.  (That is unlikely, though,
7578                  * since this should be the only persistent reservation
7579                  * command active right now.)
7580                  */
7581                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7582                     (lun->pr_key_count *
7583                      sizeof(struct scsi_per_res_key)))){
7584                         mtx_unlock(&lun->lun_lock);
7585                         free(ctsio->kern_data_ptr, M_CTL);
7586                         printf("%s: reservation length changed, retrying\n",
7587                                __func__);
7588                         goto retry;
7589                 }
7590
7591                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7592
7593                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7594                              lun->pr_key_count, res_keys->header.length);
7595
7596                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7597                         if ((key = ctl_get_prkey(lun, i)) == 0)
7598                                 continue;
7599
7600                         /*
7601                          * We used lun->pr_key_count to calculate the
7602                          * size to allocate.  If it turns out the number of
7603                          * initiators with the registered flag set is
7604                          * larger than that (i.e. they haven't been kept in
7605                          * sync), we've got a problem.
7606                          */
7607                         if (key_count >= lun->pr_key_count) {
7608                                 key_count++;
7609                                 continue;
7610                         }
7611                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7612                         key_count++;
7613                 }
7614                 break;
7615         }
7616         case SPRI_RR: { // read reservation
7617                 struct scsi_per_res_in_rsrv *res;
7618                 int tmp_len, header_only;
7619
7620                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7621
7622                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7623
7624                 if (lun->flags & CTL_LUN_PR_RESERVED)
7625                 {
7626                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7627                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7628                                     res->header.length);
7629                         header_only = 0;
7630                 } else {
7631                         tmp_len = sizeof(struct scsi_per_res_in_header);
7632                         scsi_ulto4b(0, res->header.length);
7633                         header_only = 1;
7634                 }
7635
7636                 /*
7637                  * We had to drop the lock to allocate our buffer, which
7638                  * leaves time for someone to come in with another
7639                  * persistent reservation.  (That is unlikely, though,
7640                  * since this should be the only persistent reservation
7641                  * command active right now.)
7642                  */
7643                 if (tmp_len != total_len) {
7644                         mtx_unlock(&lun->lun_lock);
7645                         free(ctsio->kern_data_ptr, M_CTL);
7646                         printf("%s: reservation status changed, retrying\n",
7647                                __func__);
7648                         goto retry;
7649                 }
7650
7651                 /*
7652                  * No reservation held, so we're done.
7653                  */
7654                 if (header_only != 0)
7655                         break;
7656
7657                 /*
7658                  * If the registration is an All Registrants type, the key
7659                  * is 0, since it doesn't really matter.
7660                  */
7661                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7662                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7663                             res->data.reservation);
7664                 }
7665                 res->data.scopetype = lun->pr_res_type;
7666                 break;
7667         }
7668         case SPRI_RC:     //report capabilities
7669         {
7670                 struct scsi_per_res_cap *res_cap;
7671                 uint16_t type_mask;
7672
7673                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7674                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7675                 res_cap->flags1 = SPRI_CRH;
7676                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7677                 type_mask = SPRI_TM_WR_EX_AR |
7678                             SPRI_TM_EX_AC_RO |
7679                             SPRI_TM_WR_EX_RO |
7680                             SPRI_TM_EX_AC |
7681                             SPRI_TM_WR_EX |
7682                             SPRI_TM_EX_AC_AR;
7683                 scsi_ulto2b(type_mask, res_cap->type_mask);
7684                 break;
7685         }
7686         case SPRI_RS: { // read full status
7687                 struct scsi_per_res_in_full *res_status;
7688                 struct scsi_per_res_in_full_desc *res_desc;
7689                 struct ctl_port *port;
7690                 int i, len;
7691
7692                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7693
7694                 /*
7695                  * We had to drop the lock to allocate our buffer, which
7696                  * leaves time for someone to come in with another
7697                  * persistent reservation.  (That is unlikely, though,
7698                  * since this should be the only persistent reservation
7699                  * command active right now.)
7700                  */
7701                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7702                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7703                      lun->pr_key_count)){
7704                         mtx_unlock(&lun->lun_lock);
7705                         free(ctsio->kern_data_ptr, M_CTL);
7706                         printf("%s: reservation length changed, retrying\n",
7707                                __func__);
7708                         goto retry;
7709                 }
7710
7711                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7712
7713                 res_desc = &res_status->desc[0];
7714                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7715                         if ((key = ctl_get_prkey(lun, i)) == 0)
7716                                 continue;
7717
7718                         scsi_u64to8b(key, res_desc->res_key.key);
7719                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7720                             (lun->pr_res_idx == i ||
7721                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7722                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7723                                 res_desc->scopetype = lun->pr_res_type;
7724                         }
7725                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7726                             res_desc->rel_trgt_port_id);
7727                         len = 0;
7728                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7729                         if (port != NULL)
7730                                 len = ctl_create_iid(port,
7731                                     i % CTL_MAX_INIT_PER_PORT,
7732                                     res_desc->transport_id);
7733                         scsi_ulto4b(len, res_desc->additional_length);
7734                         res_desc = (struct scsi_per_res_in_full_desc *)
7735                             &res_desc->transport_id[len];
7736                 }
7737                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7738                     res_status->header.length);
7739                 break;
7740         }
7741         default:
7742                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7743         }
7744         mtx_unlock(&lun->lun_lock);
7745
7746         ctl_set_success(ctsio);
7747         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7748         ctsio->be_move_done = ctl_config_move_done;
7749         ctl_datamove((union ctl_io *)ctsio);
7750         return (CTL_RETVAL_COMPLETE);
7751 }
7752
7753 /*
7754  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7755  * it should return.
7756  */
7757 static int
7758 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7759                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7760                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7761                 struct scsi_per_res_out_parms* param)
7762 {
7763         union ctl_ha_msg persis_io;
7764         int i;
7765
7766         mtx_lock(&lun->lun_lock);
7767         if (sa_res_key == 0) {
7768                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7769                         /* validate scope and type */
7770                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7771                              SPR_LU_SCOPE) {
7772                                 mtx_unlock(&lun->lun_lock);
7773                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7774                                                       /*sks_valid*/ 1,
7775                                                       /*command*/ 1,
7776                                                       /*field*/ 2,
7777                                                       /*bit_valid*/ 1,
7778                                                       /*bit*/ 4);
7779                                 ctl_done((union ctl_io *)ctsio);
7780                                 return (1);
7781                         }
7782
7783                         if (type>8 || type==2 || type==4 || type==0) {
7784                                 mtx_unlock(&lun->lun_lock);
7785                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7786                                                       /*sks_valid*/ 1,
7787                                                       /*command*/ 1,
7788                                                       /*field*/ 2,
7789                                                       /*bit_valid*/ 1,
7790                                                       /*bit*/ 0);
7791                                 ctl_done((union ctl_io *)ctsio);
7792                                 return (1);
7793                         }
7794
7795                         /*
7796                          * Unregister everybody else and build UA for
7797                          * them
7798                          */
7799                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7800                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7801                                         continue;
7802
7803                                 ctl_clr_prkey(lun, i);
7804                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7805                         }
7806                         lun->pr_key_count = 1;
7807                         lun->pr_res_type = type;
7808                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7809                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7810                                 lun->pr_res_idx = residx;
7811                         lun->pr_generation++;
7812                         mtx_unlock(&lun->lun_lock);
7813
7814                         /* send msg to other side */
7815                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7816                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7817                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7818                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7819                         persis_io.pr.pr_info.res_type = type;
7820                         memcpy(persis_io.pr.pr_info.sa_res_key,
7821                                param->serv_act_res_key,
7822                                sizeof(param->serv_act_res_key));
7823                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7824                             sizeof(persis_io.pr), M_WAITOK);
7825                 } else {
7826                         /* not all registrants */
7827                         mtx_unlock(&lun->lun_lock);
7828                         free(ctsio->kern_data_ptr, M_CTL);
7829                         ctl_set_invalid_field(ctsio,
7830                                               /*sks_valid*/ 1,
7831                                               /*command*/ 0,
7832                                               /*field*/ 8,
7833                                               /*bit_valid*/ 0,
7834                                               /*bit*/ 0);
7835                         ctl_done((union ctl_io *)ctsio);
7836                         return (1);
7837                 }
7838         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7839                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7840                 int found = 0;
7841
7842                 if (res_key == sa_res_key) {
7843                         /* special case */
7844                         /*
7845                          * The spec implies this is not good but doesn't
7846                          * say what to do. There are two choices either
7847                          * generate a res conflict or check condition
7848                          * with illegal field in parameter data. Since
7849                          * that is what is done when the sa_res_key is
7850                          * zero I'll take that approach since this has
7851                          * to do with the sa_res_key.
7852                          */
7853                         mtx_unlock(&lun->lun_lock);
7854                         free(ctsio->kern_data_ptr, M_CTL);
7855                         ctl_set_invalid_field(ctsio,
7856                                               /*sks_valid*/ 1,
7857                                               /*command*/ 0,
7858                                               /*field*/ 8,
7859                                               /*bit_valid*/ 0,
7860                                               /*bit*/ 0);
7861                         ctl_done((union ctl_io *)ctsio);
7862                         return (1);
7863                 }
7864
7865                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7866                         if (ctl_get_prkey(lun, i) != sa_res_key)
7867                                 continue;
7868
7869                         found = 1;
7870                         ctl_clr_prkey(lun, i);
7871                         lun->pr_key_count--;
7872                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7873                 }
7874                 if (!found) {
7875                         mtx_unlock(&lun->lun_lock);
7876                         free(ctsio->kern_data_ptr, M_CTL);
7877                         ctl_set_reservation_conflict(ctsio);
7878                         ctl_done((union ctl_io *)ctsio);
7879                         return (CTL_RETVAL_COMPLETE);
7880                 }
7881                 lun->pr_generation++;
7882                 mtx_unlock(&lun->lun_lock);
7883
7884                 /* send msg to other side */
7885                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7886                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7887                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7888                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7889                 persis_io.pr.pr_info.res_type = type;
7890                 memcpy(persis_io.pr.pr_info.sa_res_key,
7891                        param->serv_act_res_key,
7892                        sizeof(param->serv_act_res_key));
7893                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7894                     sizeof(persis_io.pr), M_WAITOK);
7895         } else {
7896                 /* Reserved but not all registrants */
7897                 /* sa_res_key is res holder */
7898                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7899                         /* validate scope and type */
7900                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7901                              SPR_LU_SCOPE) {
7902                                 mtx_unlock(&lun->lun_lock);
7903                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7904                                                       /*sks_valid*/ 1,
7905                                                       /*command*/ 1,
7906                                                       /*field*/ 2,
7907                                                       /*bit_valid*/ 1,
7908                                                       /*bit*/ 4);
7909                                 ctl_done((union ctl_io *)ctsio);
7910                                 return (1);
7911                         }
7912
7913                         if (type>8 || type==2 || type==4 || type==0) {
7914                                 mtx_unlock(&lun->lun_lock);
7915                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7916                                                       /*sks_valid*/ 1,
7917                                                       /*command*/ 1,
7918                                                       /*field*/ 2,
7919                                                       /*bit_valid*/ 1,
7920                                                       /*bit*/ 0);
7921                                 ctl_done((union ctl_io *)ctsio);
7922                                 return (1);
7923                         }
7924
7925                         /*
7926                          * Do the following:
7927                          * if sa_res_key != res_key remove all
7928                          * registrants w/sa_res_key and generate UA
7929                          * for these registrants(Registrations
7930                          * Preempted) if it wasn't an exclusive
7931                          * reservation generate UA(Reservations
7932                          * Preempted) for all other registered nexuses
7933                          * if the type has changed. Establish the new
7934                          * reservation and holder. If res_key and
7935                          * sa_res_key are the same do the above
7936                          * except don't unregister the res holder.
7937                          */
7938
7939                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7940                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7941                                         continue;
7942
7943                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
7944                                         ctl_clr_prkey(lun, i);
7945                                         lun->pr_key_count--;
7946                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7947                                 } else if (type != lun->pr_res_type &&
7948                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
7949                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
7950                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
7951                                 }
7952                         }
7953                         lun->pr_res_type = type;
7954                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7955                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7956                                 lun->pr_res_idx = residx;
7957                         else
7958                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7959                         lun->pr_generation++;
7960                         mtx_unlock(&lun->lun_lock);
7961
7962                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7963                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7964                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7965                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7966                         persis_io.pr.pr_info.res_type = type;
7967                         memcpy(persis_io.pr.pr_info.sa_res_key,
7968                                param->serv_act_res_key,
7969                                sizeof(param->serv_act_res_key));
7970                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7971                             sizeof(persis_io.pr), M_WAITOK);
7972                 } else {
7973                         /*
7974                          * sa_res_key is not the res holder just
7975                          * remove registrants
7976                          */
7977                         int found=0;
7978
7979                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7980                                 if (sa_res_key != ctl_get_prkey(lun, i))
7981                                         continue;
7982
7983                                 found = 1;
7984                                 ctl_clr_prkey(lun, i);
7985                                 lun->pr_key_count--;
7986                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7987                         }
7988
7989                         if (!found) {
7990                                 mtx_unlock(&lun->lun_lock);
7991                                 free(ctsio->kern_data_ptr, M_CTL);
7992                                 ctl_set_reservation_conflict(ctsio);
7993                                 ctl_done((union ctl_io *)ctsio);
7994                                 return (1);
7995                         }
7996                         lun->pr_generation++;
7997                         mtx_unlock(&lun->lun_lock);
7998
7999                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8000                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8001                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8002                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8003                         persis_io.pr.pr_info.res_type = type;
8004                         memcpy(persis_io.pr.pr_info.sa_res_key,
8005                                param->serv_act_res_key,
8006                                sizeof(param->serv_act_res_key));
8007                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8008                             sizeof(persis_io.pr), M_WAITOK);
8009                 }
8010         }
8011         return (0);
8012 }
8013
8014 static void
8015 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8016 {
8017         uint64_t sa_res_key;
8018         int i;
8019
8020         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8021
8022         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8023          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8024          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8025                 if (sa_res_key == 0) {
8026                         /*
8027                          * Unregister everybody else and build UA for
8028                          * them
8029                          */
8030                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8031                                 if (i == msg->pr.pr_info.residx ||
8032                                     ctl_get_prkey(lun, i) == 0)
8033                                         continue;
8034
8035                                 ctl_clr_prkey(lun, i);
8036                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8037                         }
8038
8039                         lun->pr_key_count = 1;
8040                         lun->pr_res_type = msg->pr.pr_info.res_type;
8041                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8042                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8043                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8044                 } else {
8045                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8046                                 if (sa_res_key == ctl_get_prkey(lun, i))
8047                                         continue;
8048
8049                                 ctl_clr_prkey(lun, i);
8050                                 lun->pr_key_count--;
8051                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8052                         }
8053                 }
8054         } else {
8055                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8056                         if (i == msg->pr.pr_info.residx ||
8057                             ctl_get_prkey(lun, i) == 0)
8058                                 continue;
8059
8060                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8061                                 ctl_clr_prkey(lun, i);
8062                                 lun->pr_key_count--;
8063                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8064                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8065                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8066                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8067                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8068                         }
8069                 }
8070                 lun->pr_res_type = msg->pr.pr_info.res_type;
8071                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8072                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8073                         lun->pr_res_idx = msg->pr.pr_info.residx;
8074                 else
8075                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8076         }
8077         lun->pr_generation++;
8078
8079 }
8080
8081
8082 int
8083 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8084 {
8085         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8086         struct ctl_lun *lun = CTL_LUN(ctsio);
8087         int retval;
8088         u_int32_t param_len;
8089         struct scsi_per_res_out *cdb;
8090         struct scsi_per_res_out_parms* param;
8091         uint32_t residx;
8092         uint64_t res_key, sa_res_key, key;
8093         uint8_t type;
8094         union ctl_ha_msg persis_io;
8095         int    i;
8096
8097         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8098
8099         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8100         retval = CTL_RETVAL_COMPLETE;
8101
8102         /*
8103          * We only support whole-LUN scope.  The scope & type are ignored for
8104          * register, register and ignore existing key and clear.
8105          * We sometimes ignore scope and type on preempts too!!
8106          * Verify reservation type here as well.
8107          */
8108         type = cdb->scope_type & SPR_TYPE_MASK;
8109         if ((cdb->action == SPRO_RESERVE)
8110          || (cdb->action == SPRO_RELEASE)) {
8111                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8112                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8113                                               /*sks_valid*/ 1,
8114                                               /*command*/ 1,
8115                                               /*field*/ 2,
8116                                               /*bit_valid*/ 1,
8117                                               /*bit*/ 4);
8118                         ctl_done((union ctl_io *)ctsio);
8119                         return (CTL_RETVAL_COMPLETE);
8120                 }
8121
8122                 if (type>8 || type==2 || type==4 || type==0) {
8123                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8124                                               /*sks_valid*/ 1,
8125                                               /*command*/ 1,
8126                                               /*field*/ 2,
8127                                               /*bit_valid*/ 1,
8128                                               /*bit*/ 0);
8129                         ctl_done((union ctl_io *)ctsio);
8130                         return (CTL_RETVAL_COMPLETE);
8131                 }
8132         }
8133
8134         param_len = scsi_4btoul(cdb->length);
8135
8136         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8137                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8138                 ctsio->kern_data_len = param_len;
8139                 ctsio->kern_total_len = param_len;
8140                 ctsio->kern_rel_offset = 0;
8141                 ctsio->kern_sg_entries = 0;
8142                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8143                 ctsio->be_move_done = ctl_config_move_done;
8144                 ctl_datamove((union ctl_io *)ctsio);
8145
8146                 return (CTL_RETVAL_COMPLETE);
8147         }
8148
8149         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8150
8151         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8152         res_key = scsi_8btou64(param->res_key.key);
8153         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8154
8155         /*
8156          * Validate the reservation key here except for SPRO_REG_IGNO
8157          * This must be done for all other service actions
8158          */
8159         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8160                 mtx_lock(&lun->lun_lock);
8161                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8162                         if (res_key != key) {
8163                                 /*
8164                                  * The current key passed in doesn't match
8165                                  * the one the initiator previously
8166                                  * registered.
8167                                  */
8168                                 mtx_unlock(&lun->lun_lock);
8169                                 free(ctsio->kern_data_ptr, M_CTL);
8170                                 ctl_set_reservation_conflict(ctsio);
8171                                 ctl_done((union ctl_io *)ctsio);
8172                                 return (CTL_RETVAL_COMPLETE);
8173                         }
8174                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8175                         /*
8176                          * We are not registered
8177                          */
8178                         mtx_unlock(&lun->lun_lock);
8179                         free(ctsio->kern_data_ptr, M_CTL);
8180                         ctl_set_reservation_conflict(ctsio);
8181                         ctl_done((union ctl_io *)ctsio);
8182                         return (CTL_RETVAL_COMPLETE);
8183                 } else if (res_key != 0) {
8184                         /*
8185                          * We are not registered and trying to register but
8186                          * the register key isn't zero.
8187                          */
8188                         mtx_unlock(&lun->lun_lock);
8189                         free(ctsio->kern_data_ptr, M_CTL);
8190                         ctl_set_reservation_conflict(ctsio);
8191                         ctl_done((union ctl_io *)ctsio);
8192                         return (CTL_RETVAL_COMPLETE);
8193                 }
8194                 mtx_unlock(&lun->lun_lock);
8195         }
8196
8197         switch (cdb->action & SPRO_ACTION_MASK) {
8198         case SPRO_REGISTER:
8199         case SPRO_REG_IGNO: {
8200
8201 #if 0
8202                 printf("Registration received\n");
8203 #endif
8204
8205                 /*
8206                  * We don't support any of these options, as we report in
8207                  * the read capabilities request (see
8208                  * ctl_persistent_reserve_in(), above).
8209                  */
8210                 if ((param->flags & SPR_SPEC_I_PT)
8211                  || (param->flags & SPR_ALL_TG_PT)
8212                  || (param->flags & SPR_APTPL)) {
8213                         int bit_ptr;
8214
8215                         if (param->flags & SPR_APTPL)
8216                                 bit_ptr = 0;
8217                         else if (param->flags & SPR_ALL_TG_PT)
8218                                 bit_ptr = 2;
8219                         else /* SPR_SPEC_I_PT */
8220                                 bit_ptr = 3;
8221
8222                         free(ctsio->kern_data_ptr, M_CTL);
8223                         ctl_set_invalid_field(ctsio,
8224                                               /*sks_valid*/ 1,
8225                                               /*command*/ 0,
8226                                               /*field*/ 20,
8227                                               /*bit_valid*/ 1,
8228                                               /*bit*/ bit_ptr);
8229                         ctl_done((union ctl_io *)ctsio);
8230                         return (CTL_RETVAL_COMPLETE);
8231                 }
8232
8233                 mtx_lock(&lun->lun_lock);
8234
8235                 /*
8236                  * The initiator wants to clear the
8237                  * key/unregister.
8238                  */
8239                 if (sa_res_key == 0) {
8240                         if ((res_key == 0
8241                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8242                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8243                           && ctl_get_prkey(lun, residx) == 0)) {
8244                                 mtx_unlock(&lun->lun_lock);
8245                                 goto done;
8246                         }
8247
8248                         ctl_clr_prkey(lun, residx);
8249                         lun->pr_key_count--;
8250
8251                         if (residx == lun->pr_res_idx) {
8252                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8253                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8254
8255                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8256                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8257                                     lun->pr_key_count) {
8258                                         /*
8259                                          * If the reservation is a registrants
8260                                          * only type we need to generate a UA
8261                                          * for other registered inits.  The
8262                                          * sense code should be RESERVATIONS
8263                                          * RELEASED
8264                                          */
8265
8266                                         for (i = softc->init_min; i < softc->init_max; i++){
8267                                                 if (ctl_get_prkey(lun, i) == 0)
8268                                                         continue;
8269                                                 ctl_est_ua(lun, i,
8270                                                     CTL_UA_RES_RELEASE);
8271                                         }
8272                                 }
8273                                 lun->pr_res_type = 0;
8274                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8275                                 if (lun->pr_key_count==0) {
8276                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8277                                         lun->pr_res_type = 0;
8278                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8279                                 }
8280                         }
8281                         lun->pr_generation++;
8282                         mtx_unlock(&lun->lun_lock);
8283
8284                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8285                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8286                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8287                         persis_io.pr.pr_info.residx = residx;
8288                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8289                             sizeof(persis_io.pr), M_WAITOK);
8290                 } else /* sa_res_key != 0 */ {
8291
8292                         /*
8293                          * If we aren't registered currently then increment
8294                          * the key count and set the registered flag.
8295                          */
8296                         ctl_alloc_prkey(lun, residx);
8297                         if (ctl_get_prkey(lun, residx) == 0)
8298                                 lun->pr_key_count++;
8299                         ctl_set_prkey(lun, residx, sa_res_key);
8300                         lun->pr_generation++;
8301                         mtx_unlock(&lun->lun_lock);
8302
8303                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8304                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8305                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8306                         persis_io.pr.pr_info.residx = residx;
8307                         memcpy(persis_io.pr.pr_info.sa_res_key,
8308                                param->serv_act_res_key,
8309                                sizeof(param->serv_act_res_key));
8310                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8311                             sizeof(persis_io.pr), M_WAITOK);
8312                 }
8313
8314                 break;
8315         }
8316         case SPRO_RESERVE:
8317 #if 0
8318                 printf("Reserve executed type %d\n", type);
8319 #endif
8320                 mtx_lock(&lun->lun_lock);
8321                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8322                         /*
8323                          * if this isn't the reservation holder and it's
8324                          * not a "all registrants" type or if the type is
8325                          * different then we have a conflict
8326                          */
8327                         if ((lun->pr_res_idx != residx
8328                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8329                          || lun->pr_res_type != type) {
8330                                 mtx_unlock(&lun->lun_lock);
8331                                 free(ctsio->kern_data_ptr, M_CTL);
8332                                 ctl_set_reservation_conflict(ctsio);
8333                                 ctl_done((union ctl_io *)ctsio);
8334                                 return (CTL_RETVAL_COMPLETE);
8335                         }
8336                         mtx_unlock(&lun->lun_lock);
8337                 } else /* create a reservation */ {
8338                         /*
8339                          * If it's not an "all registrants" type record
8340                          * reservation holder
8341                          */
8342                         if (type != SPR_TYPE_WR_EX_AR
8343                          && type != SPR_TYPE_EX_AC_AR)
8344                                 lun->pr_res_idx = residx; /* Res holder */
8345                         else
8346                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8347
8348                         lun->flags |= CTL_LUN_PR_RESERVED;
8349                         lun->pr_res_type = type;
8350
8351                         mtx_unlock(&lun->lun_lock);
8352
8353                         /* send msg to other side */
8354                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8355                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8356                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8357                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8358                         persis_io.pr.pr_info.res_type = type;
8359                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8360                             sizeof(persis_io.pr), M_WAITOK);
8361                 }
8362                 break;
8363
8364         case SPRO_RELEASE:
8365                 mtx_lock(&lun->lun_lock);
8366                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8367                         /* No reservation exists return good status */
8368                         mtx_unlock(&lun->lun_lock);
8369                         goto done;
8370                 }
8371                 /*
8372                  * Is this nexus a reservation holder?
8373                  */
8374                 if (lun->pr_res_idx != residx
8375                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8376                         /*
8377                          * not a res holder return good status but
8378                          * do nothing
8379                          */
8380                         mtx_unlock(&lun->lun_lock);
8381                         goto done;
8382                 }
8383
8384                 if (lun->pr_res_type != type) {
8385                         mtx_unlock(&lun->lun_lock);
8386                         free(ctsio->kern_data_ptr, M_CTL);
8387                         ctl_set_illegal_pr_release(ctsio);
8388                         ctl_done((union ctl_io *)ctsio);
8389                         return (CTL_RETVAL_COMPLETE);
8390                 }
8391
8392                 /* okay to release */
8393                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8394                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8395                 lun->pr_res_type = 0;
8396
8397                 /*
8398                  * If this isn't an exclusive access reservation and NUAR
8399                  * is not set, generate UA for all other registrants.
8400                  */
8401                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8402                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8403                         for (i = softc->init_min; i < softc->init_max; i++) {
8404                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8405                                         continue;
8406                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8407                         }
8408                 }
8409                 mtx_unlock(&lun->lun_lock);
8410
8411                 /* Send msg to other side */
8412                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8413                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8414                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8415                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8416                      sizeof(persis_io.pr), M_WAITOK);
8417                 break;
8418
8419         case SPRO_CLEAR:
8420                 /* send msg to other side */
8421
8422                 mtx_lock(&lun->lun_lock);
8423                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8424                 lun->pr_res_type = 0;
8425                 lun->pr_key_count = 0;
8426                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8427
8428                 ctl_clr_prkey(lun, residx);
8429                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8430                         if (ctl_get_prkey(lun, i) != 0) {
8431                                 ctl_clr_prkey(lun, i);
8432                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8433                         }
8434                 lun->pr_generation++;
8435                 mtx_unlock(&lun->lun_lock);
8436
8437                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8438                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8439                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8440                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8441                      sizeof(persis_io.pr), M_WAITOK);
8442                 break;
8443
8444         case SPRO_PREEMPT:
8445         case SPRO_PRE_ABO: {
8446                 int nretval;
8447
8448                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8449                                           residx, ctsio, cdb, param);
8450                 if (nretval != 0)
8451                         return (CTL_RETVAL_COMPLETE);
8452                 break;
8453         }
8454         default:
8455                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8456         }
8457
8458 done:
8459         free(ctsio->kern_data_ptr, M_CTL);
8460         ctl_set_success(ctsio);
8461         ctl_done((union ctl_io *)ctsio);
8462
8463         return (retval);
8464 }
8465
8466 /*
8467  * This routine is for handling a message from the other SC pertaining to
8468  * persistent reserve out. All the error checking will have been done
8469  * so only perorming the action need be done here to keep the two
8470  * in sync.
8471  */
8472 static void
8473 ctl_hndl_per_res_out_on_other_sc(union ctl_io *io)
8474 {
8475         struct ctl_softc *softc = CTL_SOFTC(io);
8476         union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg;
8477         struct ctl_lun *lun;
8478         int i;
8479         uint32_t residx, targ_lun;
8480
8481         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8482         mtx_lock(&softc->ctl_lock);
8483         if (targ_lun >= CTL_MAX_LUNS ||
8484             (lun = softc->ctl_luns[targ_lun]) == NULL) {
8485                 mtx_unlock(&softc->ctl_lock);
8486                 return;
8487         }
8488         mtx_lock(&lun->lun_lock);
8489         mtx_unlock(&softc->ctl_lock);
8490         if (lun->flags & CTL_LUN_DISABLED) {
8491                 mtx_unlock(&lun->lun_lock);
8492                 return;
8493         }
8494         residx = ctl_get_initindex(&msg->hdr.nexus);
8495         switch(msg->pr.pr_info.action) {
8496         case CTL_PR_REG_KEY:
8497                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8498                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8499                         lun->pr_key_count++;
8500                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8501                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8502                 lun->pr_generation++;
8503                 break;
8504
8505         case CTL_PR_UNREG_KEY:
8506                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8507                 lun->pr_key_count--;
8508
8509                 /* XXX Need to see if the reservation has been released */
8510                 /* if so do we need to generate UA? */
8511                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8512                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8513                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8514
8515                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8516                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8517                             lun->pr_key_count) {
8518                                 /*
8519                                  * If the reservation is a registrants
8520                                  * only type we need to generate a UA
8521                                  * for other registered inits.  The
8522                                  * sense code should be RESERVATIONS
8523                                  * RELEASED
8524                                  */
8525
8526                                 for (i = softc->init_min; i < softc->init_max; i++) {
8527                                         if (ctl_get_prkey(lun, i) == 0)
8528                                                 continue;
8529
8530                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8531                                 }
8532                         }
8533                         lun->pr_res_type = 0;
8534                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8535                         if (lun->pr_key_count==0) {
8536                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8537                                 lun->pr_res_type = 0;
8538                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8539                         }
8540                 }
8541                 lun->pr_generation++;
8542                 break;
8543
8544         case CTL_PR_RESERVE:
8545                 lun->flags |= CTL_LUN_PR_RESERVED;
8546                 lun->pr_res_type = msg->pr.pr_info.res_type;
8547                 lun->pr_res_idx = msg->pr.pr_info.residx;
8548
8549                 break;
8550
8551         case CTL_PR_RELEASE:
8552                 /*
8553                  * If this isn't an exclusive access reservation and NUAR
8554                  * is not set, generate UA for all other registrants.
8555                  */
8556                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8557                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8558                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8559                         for (i = softc->init_min; i < softc->init_max; i++)
8560                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8561                                         continue;
8562                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8563                 }
8564
8565                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8566                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8567                 lun->pr_res_type = 0;
8568                 break;
8569
8570         case CTL_PR_PREEMPT:
8571                 ctl_pro_preempt_other(lun, msg);
8572                 break;
8573         case CTL_PR_CLEAR:
8574                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8575                 lun->pr_res_type = 0;
8576                 lun->pr_key_count = 0;
8577                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8578
8579                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8580                         if (ctl_get_prkey(lun, i) == 0)
8581                                 continue;
8582                         ctl_clr_prkey(lun, i);
8583                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8584                 }
8585                 lun->pr_generation++;
8586                 break;
8587         }
8588
8589         mtx_unlock(&lun->lun_lock);
8590 }
8591
8592 int
8593 ctl_read_write(struct ctl_scsiio *ctsio)
8594 {
8595         struct ctl_lun *lun = CTL_LUN(ctsio);
8596         struct ctl_lba_len_flags *lbalen;
8597         uint64_t lba;
8598         uint32_t num_blocks;
8599         int flags, retval;
8600         int isread;
8601
8602         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8603
8604         flags = 0;
8605         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8606               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8607         switch (ctsio->cdb[0]) {
8608         case READ_6:
8609         case WRITE_6: {
8610                 struct scsi_rw_6 *cdb;
8611
8612                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8613
8614                 lba = scsi_3btoul(cdb->addr);
8615                 /* only 5 bits are valid in the most significant address byte */
8616                 lba &= 0x1fffff;
8617                 num_blocks = cdb->length;
8618                 /*
8619                  * This is correct according to SBC-2.
8620                  */
8621                 if (num_blocks == 0)
8622                         num_blocks = 256;
8623                 break;
8624         }
8625         case READ_10:
8626         case WRITE_10: {
8627                 struct scsi_rw_10 *cdb;
8628
8629                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8630                 if (cdb->byte2 & SRW10_FUA)
8631                         flags |= CTL_LLF_FUA;
8632                 if (cdb->byte2 & SRW10_DPO)
8633                         flags |= CTL_LLF_DPO;
8634                 lba = scsi_4btoul(cdb->addr);
8635                 num_blocks = scsi_2btoul(cdb->length);
8636                 break;
8637         }
8638         case WRITE_VERIFY_10: {
8639                 struct scsi_write_verify_10 *cdb;
8640
8641                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8642                 flags |= CTL_LLF_FUA;
8643                 if (cdb->byte2 & SWV_DPO)
8644                         flags |= CTL_LLF_DPO;
8645                 lba = scsi_4btoul(cdb->addr);
8646                 num_blocks = scsi_2btoul(cdb->length);
8647                 break;
8648         }
8649         case READ_12:
8650         case WRITE_12: {
8651                 struct scsi_rw_12 *cdb;
8652
8653                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8654                 if (cdb->byte2 & SRW12_FUA)
8655                         flags |= CTL_LLF_FUA;
8656                 if (cdb->byte2 & SRW12_DPO)
8657                         flags |= CTL_LLF_DPO;
8658                 lba = scsi_4btoul(cdb->addr);
8659                 num_blocks = scsi_4btoul(cdb->length);
8660                 break;
8661         }
8662         case WRITE_VERIFY_12: {
8663                 struct scsi_write_verify_12 *cdb;
8664
8665                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8666                 flags |= CTL_LLF_FUA;
8667                 if (cdb->byte2 & SWV_DPO)
8668                         flags |= CTL_LLF_DPO;
8669                 lba = scsi_4btoul(cdb->addr);
8670                 num_blocks = scsi_4btoul(cdb->length);
8671                 break;
8672         }
8673         case READ_16:
8674         case WRITE_16: {
8675                 struct scsi_rw_16 *cdb;
8676
8677                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8678                 if (cdb->byte2 & SRW12_FUA)
8679                         flags |= CTL_LLF_FUA;
8680                 if (cdb->byte2 & SRW12_DPO)
8681                         flags |= CTL_LLF_DPO;
8682                 lba = scsi_8btou64(cdb->addr);
8683                 num_blocks = scsi_4btoul(cdb->length);
8684                 break;
8685         }
8686         case WRITE_ATOMIC_16: {
8687                 struct scsi_write_atomic_16 *cdb;
8688
8689                 if (lun->be_lun->atomicblock == 0) {
8690                         ctl_set_invalid_opcode(ctsio);
8691                         ctl_done((union ctl_io *)ctsio);
8692                         return (CTL_RETVAL_COMPLETE);
8693                 }
8694
8695                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8696                 if (cdb->byte2 & SRW12_FUA)
8697                         flags |= CTL_LLF_FUA;
8698                 if (cdb->byte2 & SRW12_DPO)
8699                         flags |= CTL_LLF_DPO;
8700                 lba = scsi_8btou64(cdb->addr);
8701                 num_blocks = scsi_2btoul(cdb->length);
8702                 if (num_blocks > lun->be_lun->atomicblock) {
8703                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8704                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8705                             /*bit*/ 0);
8706                         ctl_done((union ctl_io *)ctsio);
8707                         return (CTL_RETVAL_COMPLETE);
8708                 }
8709                 break;
8710         }
8711         case WRITE_VERIFY_16: {
8712                 struct scsi_write_verify_16 *cdb;
8713
8714                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8715                 flags |= CTL_LLF_FUA;
8716                 if (cdb->byte2 & SWV_DPO)
8717                         flags |= CTL_LLF_DPO;
8718                 lba = scsi_8btou64(cdb->addr);
8719                 num_blocks = scsi_4btoul(cdb->length);
8720                 break;
8721         }
8722         default:
8723                 /*
8724                  * We got a command we don't support.  This shouldn't
8725                  * happen, commands should be filtered out above us.
8726                  */
8727                 ctl_set_invalid_opcode(ctsio);
8728                 ctl_done((union ctl_io *)ctsio);
8729
8730                 return (CTL_RETVAL_COMPLETE);
8731                 break; /* NOTREACHED */
8732         }
8733
8734         /*
8735          * The first check is to make sure we're in bounds, the second
8736          * check is to catch wrap-around problems.  If the lba + num blocks
8737          * is less than the lba, then we've wrapped around and the block
8738          * range is invalid anyway.
8739          */
8740         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8741          || ((lba + num_blocks) < lba)) {
8742                 ctl_set_lba_out_of_range(ctsio,
8743                     MAX(lba, lun->be_lun->maxlba + 1));
8744                 ctl_done((union ctl_io *)ctsio);
8745                 return (CTL_RETVAL_COMPLETE);
8746         }
8747
8748         /*
8749          * According to SBC-3, a transfer length of 0 is not an error.
8750          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8751          * translates to 256 blocks for those commands.
8752          */
8753         if (num_blocks == 0) {
8754                 ctl_set_success(ctsio);
8755                 ctl_done((union ctl_io *)ctsio);
8756                 return (CTL_RETVAL_COMPLETE);
8757         }
8758
8759         /* Set FUA and/or DPO if caches are disabled. */
8760         if (isread) {
8761                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8762                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8763         } else {
8764                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8765                         flags |= CTL_LLF_FUA;
8766         }
8767
8768         lbalen = (struct ctl_lba_len_flags *)
8769             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8770         lbalen->lba = lba;
8771         lbalen->len = num_blocks;
8772         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8773
8774         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8775         ctsio->kern_rel_offset = 0;
8776
8777         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8778
8779         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8780         return (retval);
8781 }
8782
8783 static int
8784 ctl_cnw_cont(union ctl_io *io)
8785 {
8786         struct ctl_lun *lun = CTL_LUN(io);
8787         struct ctl_scsiio *ctsio;
8788         struct ctl_lba_len_flags *lbalen;
8789         int retval;
8790
8791         ctsio = &io->scsiio;
8792         ctsio->io_hdr.status = CTL_STATUS_NONE;
8793         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8794         lbalen = (struct ctl_lba_len_flags *)
8795             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8796         lbalen->flags &= ~CTL_LLF_COMPARE;
8797         lbalen->flags |= CTL_LLF_WRITE;
8798
8799         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8800         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8801         return (retval);
8802 }
8803
8804 int
8805 ctl_cnw(struct ctl_scsiio *ctsio)
8806 {
8807         struct ctl_lun *lun = CTL_LUN(ctsio);
8808         struct ctl_lba_len_flags *lbalen;
8809         uint64_t lba;
8810         uint32_t num_blocks;
8811         int flags, retval;
8812
8813         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8814
8815         flags = 0;
8816         switch (ctsio->cdb[0]) {
8817         case COMPARE_AND_WRITE: {
8818                 struct scsi_compare_and_write *cdb;
8819
8820                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8821                 if (cdb->byte2 & SRW10_FUA)
8822                         flags |= CTL_LLF_FUA;
8823                 if (cdb->byte2 & SRW10_DPO)
8824                         flags |= CTL_LLF_DPO;
8825                 lba = scsi_8btou64(cdb->addr);
8826                 num_blocks = cdb->length;
8827                 break;
8828         }
8829         default:
8830                 /*
8831                  * We got a command we don't support.  This shouldn't
8832                  * happen, commands should be filtered out above us.
8833                  */
8834                 ctl_set_invalid_opcode(ctsio);
8835                 ctl_done((union ctl_io *)ctsio);
8836
8837                 return (CTL_RETVAL_COMPLETE);
8838                 break; /* NOTREACHED */
8839         }
8840
8841         /*
8842          * The first check is to make sure we're in bounds, the second
8843          * check is to catch wrap-around problems.  If the lba + num blocks
8844          * is less than the lba, then we've wrapped around and the block
8845          * range is invalid anyway.
8846          */
8847         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8848          || ((lba + num_blocks) < lba)) {
8849                 ctl_set_lba_out_of_range(ctsio,
8850                     MAX(lba, lun->be_lun->maxlba + 1));
8851                 ctl_done((union ctl_io *)ctsio);
8852                 return (CTL_RETVAL_COMPLETE);
8853         }
8854
8855         /*
8856          * According to SBC-3, a transfer length of 0 is not an error.
8857          */
8858         if (num_blocks == 0) {
8859                 ctl_set_success(ctsio);
8860                 ctl_done((union ctl_io *)ctsio);
8861                 return (CTL_RETVAL_COMPLETE);
8862         }
8863
8864         /* Set FUA if write cache is disabled. */
8865         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8866                 flags |= CTL_LLF_FUA;
8867
8868         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8869         ctsio->kern_rel_offset = 0;
8870
8871         /*
8872          * Set the IO_CONT flag, so that if this I/O gets passed to
8873          * ctl_data_submit_done(), it'll get passed back to
8874          * ctl_ctl_cnw_cont() for further processing.
8875          */
8876         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8877         ctsio->io_cont = ctl_cnw_cont;
8878
8879         lbalen = (struct ctl_lba_len_flags *)
8880             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8881         lbalen->lba = lba;
8882         lbalen->len = num_blocks;
8883         lbalen->flags = CTL_LLF_COMPARE | flags;
8884
8885         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8886         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8887         return (retval);
8888 }
8889
8890 int
8891 ctl_verify(struct ctl_scsiio *ctsio)
8892 {
8893         struct ctl_lun *lun = CTL_LUN(ctsio);
8894         struct ctl_lba_len_flags *lbalen;
8895         uint64_t lba;
8896         uint32_t num_blocks;
8897         int bytchk, flags;
8898         int retval;
8899
8900         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8901
8902         bytchk = 0;
8903         flags = CTL_LLF_FUA;
8904         switch (ctsio->cdb[0]) {
8905         case VERIFY_10: {
8906                 struct scsi_verify_10 *cdb;
8907
8908                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8909                 if (cdb->byte2 & SVFY_BYTCHK)
8910                         bytchk = 1;
8911                 if (cdb->byte2 & SVFY_DPO)
8912                         flags |= CTL_LLF_DPO;
8913                 lba = scsi_4btoul(cdb->addr);
8914                 num_blocks = scsi_2btoul(cdb->length);
8915                 break;
8916         }
8917         case VERIFY_12: {
8918                 struct scsi_verify_12 *cdb;
8919
8920                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8921                 if (cdb->byte2 & SVFY_BYTCHK)
8922                         bytchk = 1;
8923                 if (cdb->byte2 & SVFY_DPO)
8924                         flags |= CTL_LLF_DPO;
8925                 lba = scsi_4btoul(cdb->addr);
8926                 num_blocks = scsi_4btoul(cdb->length);
8927                 break;
8928         }
8929         case VERIFY_16: {
8930                 struct scsi_rw_16 *cdb;
8931
8932                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8933                 if (cdb->byte2 & SVFY_BYTCHK)
8934                         bytchk = 1;
8935                 if (cdb->byte2 & SVFY_DPO)
8936                         flags |= CTL_LLF_DPO;
8937                 lba = scsi_8btou64(cdb->addr);
8938                 num_blocks = scsi_4btoul(cdb->length);
8939                 break;
8940         }
8941         default:
8942                 /*
8943                  * We got a command we don't support.  This shouldn't
8944                  * happen, commands should be filtered out above us.
8945                  */
8946                 ctl_set_invalid_opcode(ctsio);
8947                 ctl_done((union ctl_io *)ctsio);
8948                 return (CTL_RETVAL_COMPLETE);
8949         }
8950
8951         /*
8952          * The first check is to make sure we're in bounds, the second
8953          * check is to catch wrap-around problems.  If the lba + num blocks
8954          * is less than the lba, then we've wrapped around and the block
8955          * range is invalid anyway.
8956          */
8957         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8958          || ((lba + num_blocks) < lba)) {
8959                 ctl_set_lba_out_of_range(ctsio,
8960                     MAX(lba, lun->be_lun->maxlba + 1));
8961                 ctl_done((union ctl_io *)ctsio);
8962                 return (CTL_RETVAL_COMPLETE);
8963         }
8964
8965         /*
8966          * According to SBC-3, a transfer length of 0 is not an error.
8967          */
8968         if (num_blocks == 0) {
8969                 ctl_set_success(ctsio);
8970                 ctl_done((union ctl_io *)ctsio);
8971                 return (CTL_RETVAL_COMPLETE);
8972         }
8973
8974         lbalen = (struct ctl_lba_len_flags *)
8975             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8976         lbalen->lba = lba;
8977         lbalen->len = num_blocks;
8978         if (bytchk) {
8979                 lbalen->flags = CTL_LLF_COMPARE | flags;
8980                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8981         } else {
8982                 lbalen->flags = CTL_LLF_VERIFY | flags;
8983                 ctsio->kern_total_len = 0;
8984         }
8985         ctsio->kern_rel_offset = 0;
8986
8987         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
8988         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8989         return (retval);
8990 }
8991
8992 int
8993 ctl_report_luns(struct ctl_scsiio *ctsio)
8994 {
8995         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8996         struct ctl_port *port = CTL_PORT(ctsio);
8997         struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio);
8998         struct scsi_report_luns *cdb;
8999         struct scsi_report_luns_data *lun_data;
9000         int num_filled, num_luns, num_port_luns, retval;
9001         uint32_t alloc_len, lun_datalen;
9002         uint32_t initidx, targ_lun_id, lun_id;
9003
9004         retval = CTL_RETVAL_COMPLETE;
9005         cdb = (struct scsi_report_luns *)ctsio->cdb;
9006
9007         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9008
9009         num_luns = 0;
9010         num_port_luns = port->lun_map ? port->lun_map_size : CTL_MAX_LUNS;
9011         mtx_lock(&softc->ctl_lock);
9012         for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) {
9013                 if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX)
9014                         num_luns++;
9015         }
9016         mtx_unlock(&softc->ctl_lock);
9017
9018         switch (cdb->select_report) {
9019         case RPL_REPORT_DEFAULT:
9020         case RPL_REPORT_ALL:
9021         case RPL_REPORT_NONSUBSID:
9022                 break;
9023         case RPL_REPORT_WELLKNOWN:
9024         case RPL_REPORT_ADMIN:
9025         case RPL_REPORT_CONGLOM:
9026                 num_luns = 0;
9027                 break;
9028         default:
9029                 ctl_set_invalid_field(ctsio,
9030                                       /*sks_valid*/ 1,
9031                                       /*command*/ 1,
9032                                       /*field*/ 2,
9033                                       /*bit_valid*/ 0,
9034                                       /*bit*/ 0);
9035                 ctl_done((union ctl_io *)ctsio);
9036                 return (retval);
9037                 break; /* NOTREACHED */
9038         }
9039
9040         alloc_len = scsi_4btoul(cdb->length);
9041         /*
9042          * The initiator has to allocate at least 16 bytes for this request,
9043          * so he can at least get the header and the first LUN.  Otherwise
9044          * we reject the request (per SPC-3 rev 14, section 6.21).
9045          */
9046         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9047             sizeof(struct scsi_report_luns_lundata))) {
9048                 ctl_set_invalid_field(ctsio,
9049                                       /*sks_valid*/ 1,
9050                                       /*command*/ 1,
9051                                       /*field*/ 6,
9052                                       /*bit_valid*/ 0,
9053                                       /*bit*/ 0);
9054                 ctl_done((union ctl_io *)ctsio);
9055                 return (retval);
9056         }
9057
9058         lun_datalen = sizeof(*lun_data) +
9059                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9060
9061         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9062         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9063         ctsio->kern_sg_entries = 0;
9064
9065         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9066
9067         mtx_lock(&softc->ctl_lock);
9068         for (targ_lun_id = 0, num_filled = 0;
9069             targ_lun_id < num_port_luns && num_filled < num_luns;
9070             targ_lun_id++) {
9071                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9072                 if (lun_id == UINT32_MAX)
9073                         continue;
9074                 lun = softc->ctl_luns[lun_id];
9075                 if (lun == NULL)
9076                         continue;
9077
9078                 be64enc(lun_data->luns[num_filled++].lundata,
9079                     ctl_encode_lun(targ_lun_id));
9080
9081                 /*
9082                  * According to SPC-3, rev 14 section 6.21:
9083                  *
9084                  * "The execution of a REPORT LUNS command to any valid and
9085                  * installed logical unit shall clear the REPORTED LUNS DATA
9086                  * HAS CHANGED unit attention condition for all logical
9087                  * units of that target with respect to the requesting
9088                  * initiator. A valid and installed logical unit is one
9089                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9090                  * INQUIRY data (see 6.4.2)."
9091                  *
9092                  * If request_lun is NULL, the LUN this report luns command
9093                  * was issued to is either disabled or doesn't exist. In that
9094                  * case, we shouldn't clear any pending lun change unit
9095                  * attention.
9096                  */
9097                 if (request_lun != NULL) {
9098                         mtx_lock(&lun->lun_lock);
9099                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9100                         mtx_unlock(&lun->lun_lock);
9101                 }
9102         }
9103         mtx_unlock(&softc->ctl_lock);
9104
9105         /*
9106          * It's quite possible that we've returned fewer LUNs than we allocated
9107          * space for.  Trim it.
9108          */
9109         lun_datalen = sizeof(*lun_data) +
9110                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9111         ctsio->kern_rel_offset = 0;
9112         ctsio->kern_sg_entries = 0;
9113         ctsio->kern_data_len = min(lun_datalen, alloc_len);
9114         ctsio->kern_total_len = ctsio->kern_data_len;
9115
9116         /*
9117          * We set this to the actual data length, regardless of how much
9118          * space we actually have to return results.  If the user looks at
9119          * this value, he'll know whether or not he allocated enough space
9120          * and reissue the command if necessary.  We don't support well
9121          * known logical units, so if the user asks for that, return none.
9122          */
9123         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9124
9125         /*
9126          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9127          * this request.
9128          */
9129         ctl_set_success(ctsio);
9130         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9131         ctsio->be_move_done = ctl_config_move_done;
9132         ctl_datamove((union ctl_io *)ctsio);
9133         return (retval);
9134 }
9135
9136 int
9137 ctl_request_sense(struct ctl_scsiio *ctsio)
9138 {
9139         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9140         struct ctl_lun *lun = CTL_LUN(ctsio);
9141         struct scsi_request_sense *cdb;
9142         struct scsi_sense_data *sense_ptr, *ps;
9143         uint32_t initidx;
9144         int have_error;
9145         u_int sense_len = SSD_FULL_SIZE;
9146         scsi_sense_data_type sense_format;
9147         ctl_ua_type ua_type;
9148         uint8_t asc = 0, ascq = 0;
9149
9150         cdb = (struct scsi_request_sense *)ctsio->cdb;
9151
9152         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9153
9154         /*
9155          * Determine which sense format the user wants.
9156          */
9157         if (cdb->byte2 & SRS_DESC)
9158                 sense_format = SSD_TYPE_DESC;
9159         else
9160                 sense_format = SSD_TYPE_FIXED;
9161
9162         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9163         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9164         ctsio->kern_sg_entries = 0;
9165         ctsio->kern_rel_offset = 0;
9166
9167         /*
9168          * struct scsi_sense_data, which is currently set to 256 bytes, is
9169          * larger than the largest allowed value for the length field in the
9170          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9171          */
9172         ctsio->kern_data_len = cdb->length;
9173         ctsio->kern_total_len = cdb->length;
9174
9175         /*
9176          * If we don't have a LUN, we don't have any pending sense.
9177          */
9178         if (lun == NULL ||
9179             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9180              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9181                 /* "Logical unit not supported" */
9182                 ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9183                     /*current_error*/ 1,
9184                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9185                     /*asc*/ 0x25,
9186                     /*ascq*/ 0x00,
9187                     SSD_ELEM_NONE);
9188                 goto send;
9189         }
9190
9191         have_error = 0;
9192         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9193         /*
9194          * Check for pending sense, and then for pending unit attentions.
9195          * Pending sense gets returned first, then pending unit attentions.
9196          */
9197         mtx_lock(&lun->lun_lock);
9198         ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
9199         if (ps != NULL)
9200                 ps += initidx % CTL_MAX_INIT_PER_PORT;
9201         if (ps != NULL && ps->error_code != 0) {
9202                 scsi_sense_data_type stored_format;
9203
9204                 /*
9205                  * Check to see which sense format was used for the stored
9206                  * sense data.
9207                  */
9208                 stored_format = scsi_sense_type(ps);
9209
9210                 /*
9211                  * If the user requested a different sense format than the
9212                  * one we stored, then we need to convert it to the other
9213                  * format.  If we're going from descriptor to fixed format
9214                  * sense data, we may lose things in translation, depending
9215                  * on what options were used.
9216                  *
9217                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9218                  * for some reason we'll just copy it out as-is.
9219                  */
9220                 if ((stored_format == SSD_TYPE_FIXED)
9221                  && (sense_format == SSD_TYPE_DESC))
9222                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9223                             ps, (struct scsi_sense_data_desc *)sense_ptr);
9224                 else if ((stored_format == SSD_TYPE_DESC)
9225                       && (sense_format == SSD_TYPE_FIXED))
9226                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9227                             ps, (struct scsi_sense_data_fixed *)sense_ptr);
9228                 else
9229                         memcpy(sense_ptr, ps, sizeof(*sense_ptr));
9230
9231                 ps->error_code = 0;
9232                 have_error = 1;
9233         } else {
9234                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9235                     sense_format);
9236                 if (ua_type != CTL_UA_NONE)
9237                         have_error = 1;
9238         }
9239         if (have_error == 0) {
9240                 /*
9241                  * Report informational exception if have one and allowed.
9242                  */
9243                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9244                         asc = lun->ie_asc;
9245                         ascq = lun->ie_ascq;
9246                 }
9247                 ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9248                     /*current_error*/ 1,
9249                     /*sense_key*/ SSD_KEY_NO_SENSE,
9250                     /*asc*/ asc,
9251                     /*ascq*/ ascq,
9252                     SSD_ELEM_NONE);
9253         }
9254         mtx_unlock(&lun->lun_lock);
9255
9256 send:
9257         /*
9258          * We report the SCSI status as OK, since the status of the command
9259          * itself is OK.  We're reporting sense as parameter data.
9260          */
9261         ctl_set_success(ctsio);
9262         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9263         ctsio->be_move_done = ctl_config_move_done;
9264         ctl_datamove((union ctl_io *)ctsio);
9265         return (CTL_RETVAL_COMPLETE);
9266 }
9267
9268 int
9269 ctl_tur(struct ctl_scsiio *ctsio)
9270 {
9271
9272         CTL_DEBUG_PRINT(("ctl_tur\n"));
9273
9274         ctl_set_success(ctsio);
9275         ctl_done((union ctl_io *)ctsio);
9276
9277         return (CTL_RETVAL_COMPLETE);
9278 }
9279
9280 /*
9281  * SCSI VPD page 0x00, the Supported VPD Pages page.
9282  */
9283 static int
9284 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9285 {
9286         struct ctl_lun *lun = CTL_LUN(ctsio);
9287         struct scsi_vpd_supported_pages *pages;
9288         int sup_page_size;
9289         int p;
9290
9291         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9292             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9293         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9294         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9295         ctsio->kern_rel_offset = 0;
9296         ctsio->kern_sg_entries = 0;
9297         ctsio->kern_data_len = min(sup_page_size, alloc_len);
9298         ctsio->kern_total_len = ctsio->kern_data_len;
9299
9300         /*
9301          * The control device is always connected.  The disk device, on the
9302          * other hand, may not be online all the time.  Need to change this
9303          * to figure out whether the disk device is actually online or not.
9304          */
9305         if (lun != NULL)
9306                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9307                                 lun->be_lun->lun_type;
9308         else
9309                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9310
9311         p = 0;
9312         /* Supported VPD pages */
9313         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9314         /* Serial Number */
9315         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9316         /* Device Identification */
9317         pages->page_list[p++] = SVPD_DEVICE_ID;
9318         /* Extended INQUIRY Data */
9319         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9320         /* Mode Page Policy */
9321         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9322         /* SCSI Ports */
9323         pages->page_list[p++] = SVPD_SCSI_PORTS;
9324         /* Third-party Copy */
9325         pages->page_list[p++] = SVPD_SCSI_TPC;
9326         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9327                 /* Block limits */
9328                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9329                 /* Block Device Characteristics */
9330                 pages->page_list[p++] = SVPD_BDC;
9331                 /* Logical Block Provisioning */
9332                 pages->page_list[p++] = SVPD_LBP;
9333         }
9334         pages->length = p;
9335
9336         ctl_set_success(ctsio);
9337         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9338         ctsio->be_move_done = ctl_config_move_done;
9339         ctl_datamove((union ctl_io *)ctsio);
9340         return (CTL_RETVAL_COMPLETE);
9341 }
9342
9343 /*
9344  * SCSI VPD page 0x80, the Unit Serial Number page.
9345  */
9346 static int
9347 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9348 {
9349         struct ctl_lun *lun = CTL_LUN(ctsio);
9350         struct scsi_vpd_unit_serial_number *sn_ptr;
9351         int data_len;
9352
9353         data_len = 4 + CTL_SN_LEN;
9354         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9355         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9356         ctsio->kern_rel_offset = 0;
9357         ctsio->kern_sg_entries = 0;
9358         ctsio->kern_data_len = min(data_len, alloc_len);
9359         ctsio->kern_total_len = ctsio->kern_data_len;
9360
9361         /*
9362          * The control device is always connected.  The disk device, on the
9363          * other hand, may not be online all the time.  Need to change this
9364          * to figure out whether the disk device is actually online or not.
9365          */
9366         if (lun != NULL)
9367                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9368                                   lun->be_lun->lun_type;
9369         else
9370                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9371
9372         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9373         sn_ptr->length = CTL_SN_LEN;
9374         /*
9375          * If we don't have a LUN, we just leave the serial number as
9376          * all spaces.
9377          */
9378         if (lun != NULL) {
9379                 strncpy((char *)sn_ptr->serial_num,
9380                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9381         } else
9382                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9383
9384         ctl_set_success(ctsio);
9385         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9386         ctsio->be_move_done = ctl_config_move_done;
9387         ctl_datamove((union ctl_io *)ctsio);
9388         return (CTL_RETVAL_COMPLETE);
9389 }
9390
9391
9392 /*
9393  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9394  */
9395 static int
9396 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9397 {
9398         struct ctl_lun *lun = CTL_LUN(ctsio);
9399         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9400         int data_len;
9401
9402         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9403         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9404         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9405         ctsio->kern_sg_entries = 0;
9406         ctsio->kern_rel_offset = 0;
9407         ctsio->kern_data_len = min(data_len, alloc_len);
9408         ctsio->kern_total_len = ctsio->kern_data_len;
9409
9410         /*
9411          * The control device is always connected.  The disk device, on the
9412          * other hand, may not be online all the time.
9413          */
9414         if (lun != NULL)
9415                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9416                                      lun->be_lun->lun_type;
9417         else
9418                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9419         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9420         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9421         /*
9422          * We support head of queue, ordered and simple tags.
9423          */
9424         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9425         /*
9426          * Volatile cache supported.
9427          */
9428         eid_ptr->flags3 = SVPD_EID_V_SUP;
9429
9430         /*
9431          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9432          * attention for a particular IT nexus on all LUNs once we report
9433          * it to that nexus once.  This bit is required as of SPC-4.
9434          */
9435         eid_ptr->flags4 = SVPD_EID_LUICLR;
9436
9437         /*
9438          * We support revert to defaults (RTD) bit in MODE SELECT.
9439          */
9440         eid_ptr->flags5 = SVPD_EID_RTD_SUP;
9441
9442         /*
9443          * XXX KDM in order to correctly answer this, we would need
9444          * information from the SIM to determine how much sense data it
9445          * can send.  So this would really be a path inquiry field, most
9446          * likely.  This can be set to a maximum of 252 according to SPC-4,
9447          * but the hardware may or may not be able to support that much.
9448          * 0 just means that the maximum sense data length is not reported.
9449          */
9450         eid_ptr->max_sense_length = 0;
9451
9452         ctl_set_success(ctsio);
9453         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9454         ctsio->be_move_done = ctl_config_move_done;
9455         ctl_datamove((union ctl_io *)ctsio);
9456         return (CTL_RETVAL_COMPLETE);
9457 }
9458
9459 static int
9460 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9461 {
9462         struct ctl_lun *lun = CTL_LUN(ctsio);
9463         struct scsi_vpd_mode_page_policy *mpp_ptr;
9464         int data_len;
9465
9466         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9467             sizeof(struct scsi_vpd_mode_page_policy_descr);
9468
9469         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9470         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9471         ctsio->kern_rel_offset = 0;
9472         ctsio->kern_sg_entries = 0;
9473         ctsio->kern_data_len = min(data_len, alloc_len);
9474         ctsio->kern_total_len = ctsio->kern_data_len;
9475
9476         /*
9477          * The control device is always connected.  The disk device, on the
9478          * other hand, may not be online all the time.
9479          */
9480         if (lun != NULL)
9481                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9482                                      lun->be_lun->lun_type;
9483         else
9484                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9485         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9486         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9487         mpp_ptr->descr[0].page_code = 0x3f;
9488         mpp_ptr->descr[0].subpage_code = 0xff;
9489         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9490
9491         ctl_set_success(ctsio);
9492         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9493         ctsio->be_move_done = ctl_config_move_done;
9494         ctl_datamove((union ctl_io *)ctsio);
9495         return (CTL_RETVAL_COMPLETE);
9496 }
9497
9498 /*
9499  * SCSI VPD page 0x83, the Device Identification page.
9500  */
9501 static int
9502 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9503 {
9504         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9505         struct ctl_port *port = CTL_PORT(ctsio);
9506         struct ctl_lun *lun = CTL_LUN(ctsio);
9507         struct scsi_vpd_device_id *devid_ptr;
9508         struct scsi_vpd_id_descriptor *desc;
9509         int data_len, g;
9510         uint8_t proto;
9511
9512         data_len = sizeof(struct scsi_vpd_device_id) +
9513             sizeof(struct scsi_vpd_id_descriptor) +
9514                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9515             sizeof(struct scsi_vpd_id_descriptor) +
9516                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9517         if (lun && lun->lun_devid)
9518                 data_len += lun->lun_devid->len;
9519         if (port && port->port_devid)
9520                 data_len += port->port_devid->len;
9521         if (port && port->target_devid)
9522                 data_len += port->target_devid->len;
9523
9524         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9525         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9526         ctsio->kern_sg_entries = 0;
9527         ctsio->kern_rel_offset = 0;
9528         ctsio->kern_sg_entries = 0;
9529         ctsio->kern_data_len = min(data_len, alloc_len);
9530         ctsio->kern_total_len = ctsio->kern_data_len;
9531
9532         /*
9533          * The control device is always connected.  The disk device, on the
9534          * other hand, may not be online all the time.
9535          */
9536         if (lun != NULL)
9537                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9538                                      lun->be_lun->lun_type;
9539         else
9540                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9541         devid_ptr->page_code = SVPD_DEVICE_ID;
9542         scsi_ulto2b(data_len - 4, devid_ptr->length);
9543
9544         if (port && port->port_type == CTL_PORT_FC)
9545                 proto = SCSI_PROTO_FC << 4;
9546         else if (port && port->port_type == CTL_PORT_SAS)
9547                 proto = SCSI_PROTO_SAS << 4;
9548         else if (port && port->port_type == CTL_PORT_ISCSI)
9549                 proto = SCSI_PROTO_ISCSI << 4;
9550         else
9551                 proto = SCSI_PROTO_SPI << 4;
9552         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9553
9554         /*
9555          * We're using a LUN association here.  i.e., this device ID is a
9556          * per-LUN identifier.
9557          */
9558         if (lun && lun->lun_devid) {
9559                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9560                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9561                     lun->lun_devid->len);
9562         }
9563
9564         /*
9565          * This is for the WWPN which is a port association.
9566          */
9567         if (port && port->port_devid) {
9568                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9569                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9570                     port->port_devid->len);
9571         }
9572
9573         /*
9574          * This is for the Relative Target Port(type 4h) identifier
9575          */
9576         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9577         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9578             SVPD_ID_TYPE_RELTARG;
9579         desc->length = 4;
9580         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9581         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9582             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9583
9584         /*
9585          * This is for the Target Port Group(type 5h) identifier
9586          */
9587         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9588         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9589             SVPD_ID_TYPE_TPORTGRP;
9590         desc->length = 4;
9591         if (softc->is_single ||
9592             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9593                 g = 1;
9594         else
9595                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9596         scsi_ulto2b(g, &desc->identifier[2]);
9597         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9598             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9599
9600         /*
9601          * This is for the Target identifier
9602          */
9603         if (port && port->target_devid) {
9604                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9605         }
9606
9607         ctl_set_success(ctsio);
9608         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9609         ctsio->be_move_done = ctl_config_move_done;
9610         ctl_datamove((union ctl_io *)ctsio);
9611         return (CTL_RETVAL_COMPLETE);
9612 }
9613
9614 static int
9615 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9616 {
9617         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9618         struct ctl_lun *lun = CTL_LUN(ctsio);
9619         struct scsi_vpd_scsi_ports *sp;
9620         struct scsi_vpd_port_designation *pd;
9621         struct scsi_vpd_port_designation_cont *pdc;
9622         struct ctl_port *port;
9623         int data_len, num_target_ports, iid_len, id_len;
9624
9625         num_target_ports = 0;
9626         iid_len = 0;
9627         id_len = 0;
9628         mtx_lock(&softc->ctl_lock);
9629         STAILQ_FOREACH(port, &softc->port_list, links) {
9630                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9631                         continue;
9632                 if (lun != NULL &&
9633                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9634                         continue;
9635                 num_target_ports++;
9636                 if (port->init_devid)
9637                         iid_len += port->init_devid->len;
9638                 if (port->port_devid)
9639                         id_len += port->port_devid->len;
9640         }
9641         mtx_unlock(&softc->ctl_lock);
9642
9643         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9644             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9645              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9646         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9647         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9648         ctsio->kern_sg_entries = 0;
9649         ctsio->kern_rel_offset = 0;
9650         ctsio->kern_sg_entries = 0;
9651         ctsio->kern_data_len = min(data_len, alloc_len);
9652         ctsio->kern_total_len = ctsio->kern_data_len;
9653
9654         /*
9655          * The control device is always connected.  The disk device, on the
9656          * other hand, may not be online all the time.  Need to change this
9657          * to figure out whether the disk device is actually online or not.
9658          */
9659         if (lun != NULL)
9660                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9661                                   lun->be_lun->lun_type;
9662         else
9663                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9664
9665         sp->page_code = SVPD_SCSI_PORTS;
9666         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9667             sp->page_length);
9668         pd = &sp->design[0];
9669
9670         mtx_lock(&softc->ctl_lock);
9671         STAILQ_FOREACH(port, &softc->port_list, links) {
9672                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9673                         continue;
9674                 if (lun != NULL &&
9675                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9676                         continue;
9677                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9678                 if (port->init_devid) {
9679                         iid_len = port->init_devid->len;
9680                         memcpy(pd->initiator_transportid,
9681                             port->init_devid->data, port->init_devid->len);
9682                 } else
9683                         iid_len = 0;
9684                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9685                 pdc = (struct scsi_vpd_port_designation_cont *)
9686                     (&pd->initiator_transportid[iid_len]);
9687                 if (port->port_devid) {
9688                         id_len = port->port_devid->len;
9689                         memcpy(pdc->target_port_descriptors,
9690                             port->port_devid->data, port->port_devid->len);
9691                 } else
9692                         id_len = 0;
9693                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9694                 pd = (struct scsi_vpd_port_designation *)
9695                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9696         }
9697         mtx_unlock(&softc->ctl_lock);
9698
9699         ctl_set_success(ctsio);
9700         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9701         ctsio->be_move_done = ctl_config_move_done;
9702         ctl_datamove((union ctl_io *)ctsio);
9703         return (CTL_RETVAL_COMPLETE);
9704 }
9705
9706 static int
9707 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9708 {
9709         struct ctl_lun *lun = CTL_LUN(ctsio);
9710         struct scsi_vpd_block_limits *bl_ptr;
9711         uint64_t ival;
9712
9713         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9714         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9715         ctsio->kern_sg_entries = 0;
9716         ctsio->kern_rel_offset = 0;
9717         ctsio->kern_sg_entries = 0;
9718         ctsio->kern_data_len = min(sizeof(*bl_ptr), alloc_len);
9719         ctsio->kern_total_len = ctsio->kern_data_len;
9720
9721         /*
9722          * The control device is always connected.  The disk device, on the
9723          * other hand, may not be online all the time.  Need to change this
9724          * to figure out whether the disk device is actually online or not.
9725          */
9726         if (lun != NULL)
9727                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9728                                   lun->be_lun->lun_type;
9729         else
9730                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9731
9732         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9733         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9734         bl_ptr->max_cmp_write_len = 0xff;
9735         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9736         if (lun != NULL) {
9737                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9738                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9739                         ival = 0xffffffff;
9740                         ctl_get_opt_number(&lun->be_lun->options,
9741                             "unmap_max_lba", &ival);
9742                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9743                         ival = 0xffffffff;
9744                         ctl_get_opt_number(&lun->be_lun->options,
9745                             "unmap_max_descr", &ival);
9746                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9747                         if (lun->be_lun->ublockexp != 0) {
9748                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9749                                     bl_ptr->opt_unmap_grain);
9750                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9751                                     bl_ptr->unmap_grain_align);
9752                         }
9753                 }
9754                 scsi_ulto4b(lun->be_lun->atomicblock,
9755                     bl_ptr->max_atomic_transfer_length);
9756                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9757                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9758                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9759                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9760                 ival = UINT64_MAX;
9761                 ctl_get_opt_number(&lun->be_lun->options, "write_same_max_lba", &ival);
9762                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9763         }
9764
9765         ctl_set_success(ctsio);
9766         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9767         ctsio->be_move_done = ctl_config_move_done;
9768         ctl_datamove((union ctl_io *)ctsio);
9769         return (CTL_RETVAL_COMPLETE);
9770 }
9771
9772 static int
9773 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9774 {
9775         struct ctl_lun *lun = CTL_LUN(ctsio);
9776         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9777         const char *value;
9778         u_int i;
9779
9780         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9781         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9782         ctsio->kern_sg_entries = 0;
9783         ctsio->kern_rel_offset = 0;
9784         ctsio->kern_data_len = min(sizeof(*bdc_ptr), alloc_len);
9785         ctsio->kern_total_len = ctsio->kern_data_len;
9786
9787         /*
9788          * The control device is always connected.  The disk device, on the
9789          * other hand, may not be online all the time.  Need to change this
9790          * to figure out whether the disk device is actually online or not.
9791          */
9792         if (lun != NULL)
9793                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9794                                   lun->be_lun->lun_type;
9795         else
9796                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9797         bdc_ptr->page_code = SVPD_BDC;
9798         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9799         if (lun != NULL &&
9800             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9801                 i = strtol(value, NULL, 0);
9802         else
9803                 i = CTL_DEFAULT_ROTATION_RATE;
9804         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9805         if (lun != NULL &&
9806             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9807                 i = strtol(value, NULL, 0);
9808         else
9809                 i = 0;
9810         bdc_ptr->wab_wac_ff = (i & 0x0f);
9811         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9812
9813         ctl_set_success(ctsio);
9814         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9815         ctsio->be_move_done = ctl_config_move_done;
9816         ctl_datamove((union ctl_io *)ctsio);
9817         return (CTL_RETVAL_COMPLETE);
9818 }
9819
9820 static int
9821 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9822 {
9823         struct ctl_lun *lun = CTL_LUN(ctsio);
9824         struct scsi_vpd_logical_block_prov *lbp_ptr;
9825         const char *value;
9826
9827         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9828         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9829         ctsio->kern_sg_entries = 0;
9830         ctsio->kern_rel_offset = 0;
9831         ctsio->kern_data_len = min(sizeof(*lbp_ptr), alloc_len);
9832         ctsio->kern_total_len = ctsio->kern_data_len;
9833
9834         /*
9835          * The control device is always connected.  The disk device, on the
9836          * other hand, may not be online all the time.  Need to change this
9837          * to figure out whether the disk device is actually online or not.
9838          */
9839         if (lun != NULL)
9840                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9841                                   lun->be_lun->lun_type;
9842         else
9843                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9844
9845         lbp_ptr->page_code = SVPD_LBP;
9846         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9847         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
9848         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9849                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
9850                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
9851                 value = ctl_get_opt(&lun->be_lun->options, "provisioning_type");
9852                 if (value != NULL) {
9853                         if (strcmp(value, "resource") == 0)
9854                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
9855                         else if (strcmp(value, "thin") == 0)
9856                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
9857                 } else
9858                         lbp_ptr->prov_type = SVPD_LBP_THIN;
9859         }
9860
9861         ctl_set_success(ctsio);
9862         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9863         ctsio->be_move_done = ctl_config_move_done;
9864         ctl_datamove((union ctl_io *)ctsio);
9865         return (CTL_RETVAL_COMPLETE);
9866 }
9867
9868 /*
9869  * INQUIRY with the EVPD bit set.
9870  */
9871 static int
9872 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9873 {
9874         struct ctl_lun *lun = CTL_LUN(ctsio);
9875         struct scsi_inquiry *cdb;
9876         int alloc_len, retval;
9877
9878         cdb = (struct scsi_inquiry *)ctsio->cdb;
9879         alloc_len = scsi_2btoul(cdb->length);
9880
9881         switch (cdb->page_code) {
9882         case SVPD_SUPPORTED_PAGES:
9883                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9884                 break;
9885         case SVPD_UNIT_SERIAL_NUMBER:
9886                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9887                 break;
9888         case SVPD_DEVICE_ID:
9889                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9890                 break;
9891         case SVPD_EXTENDED_INQUIRY_DATA:
9892                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
9893                 break;
9894         case SVPD_MODE_PAGE_POLICY:
9895                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
9896                 break;
9897         case SVPD_SCSI_PORTS:
9898                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
9899                 break;
9900         case SVPD_SCSI_TPC:
9901                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
9902                 break;
9903         case SVPD_BLOCK_LIMITS:
9904                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9905                         goto err;
9906                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9907                 break;
9908         case SVPD_BDC:
9909                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9910                         goto err;
9911                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
9912                 break;
9913         case SVPD_LBP:
9914                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9915                         goto err;
9916                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9917                 break;
9918         default:
9919 err:
9920                 ctl_set_invalid_field(ctsio,
9921                                       /*sks_valid*/ 1,
9922                                       /*command*/ 1,
9923                                       /*field*/ 2,
9924                                       /*bit_valid*/ 0,
9925                                       /*bit*/ 0);
9926                 ctl_done((union ctl_io *)ctsio);
9927                 retval = CTL_RETVAL_COMPLETE;
9928                 break;
9929         }
9930
9931         return (retval);
9932 }
9933
9934 /*
9935  * Standard INQUIRY data.
9936  */
9937 static int
9938 ctl_inquiry_std(struct ctl_scsiio *ctsio)
9939 {
9940         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9941         struct ctl_port *port = CTL_PORT(ctsio);
9942         struct ctl_lun *lun = CTL_LUN(ctsio);
9943         struct scsi_inquiry_data *inq_ptr;
9944         struct scsi_inquiry *cdb;
9945         char *val;
9946         uint32_t alloc_len, data_len;
9947         ctl_port_type port_type;
9948
9949         port_type = port->port_type;
9950         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
9951                 port_type = CTL_PORT_SCSI;
9952
9953         cdb = (struct scsi_inquiry *)ctsio->cdb;
9954         alloc_len = scsi_2btoul(cdb->length);
9955
9956         /*
9957          * We malloc the full inquiry data size here and fill it
9958          * in.  If the user only asks for less, we'll give him
9959          * that much.
9960          */
9961         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
9962         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9963         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9964         ctsio->kern_sg_entries = 0;
9965         ctsio->kern_rel_offset = 0;
9966         ctsio->kern_data_len = min(data_len, alloc_len);
9967         ctsio->kern_total_len = ctsio->kern_data_len;
9968
9969         if (lun != NULL) {
9970                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
9971                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
9972                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9973                             lun->be_lun->lun_type;
9974                 } else {
9975                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
9976                             lun->be_lun->lun_type;
9977                 }
9978                 if (lun->flags & CTL_LUN_REMOVABLE)
9979                         inq_ptr->dev_qual2 |= SID_RMB;
9980         } else
9981                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9982
9983         /* RMB in byte 2 is 0 */
9984         inq_ptr->version = SCSI_REV_SPC5;
9985
9986         /*
9987          * According to SAM-3, even if a device only supports a single
9988          * level of LUN addressing, it should still set the HISUP bit:
9989          *
9990          * 4.9.1 Logical unit numbers overview
9991          *
9992          * All logical unit number formats described in this standard are
9993          * hierarchical in structure even when only a single level in that
9994          * hierarchy is used. The HISUP bit shall be set to one in the
9995          * standard INQUIRY data (see SPC-2) when any logical unit number
9996          * format described in this standard is used.  Non-hierarchical
9997          * formats are outside the scope of this standard.
9998          *
9999          * Therefore we set the HiSup bit here.
10000          *
10001          * The response format is 2, per SPC-3.
10002          */
10003         inq_ptr->response_format = SID_HiSup | 2;
10004
10005         inq_ptr->additional_length = data_len -
10006             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10007         CTL_DEBUG_PRINT(("additional_length = %d\n",
10008                          inq_ptr->additional_length));
10009
10010         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10011         if (port_type == CTL_PORT_SCSI)
10012                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10013         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10014         inq_ptr->flags = SID_CmdQue;
10015         if (port_type == CTL_PORT_SCSI)
10016                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10017
10018         /*
10019          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10020          * We have 8 bytes for the vendor name, and 16 bytes for the device
10021          * name and 4 bytes for the revision.
10022          */
10023         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10024             "vendor")) == NULL) {
10025                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10026         } else {
10027                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10028                 strncpy(inq_ptr->vendor, val,
10029                     min(sizeof(inq_ptr->vendor), strlen(val)));
10030         }
10031         if (lun == NULL) {
10032                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10033                     sizeof(inq_ptr->product));
10034         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10035                 switch (lun->be_lun->lun_type) {
10036                 case T_DIRECT:
10037                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10038                             sizeof(inq_ptr->product));
10039                         break;
10040                 case T_PROCESSOR:
10041                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10042                             sizeof(inq_ptr->product));
10043                         break;
10044                 case T_CDROM:
10045                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10046                             sizeof(inq_ptr->product));
10047                         break;
10048                 default:
10049                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10050                             sizeof(inq_ptr->product));
10051                         break;
10052                 }
10053         } else {
10054                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10055                 strncpy(inq_ptr->product, val,
10056                     min(sizeof(inq_ptr->product), strlen(val)));
10057         }
10058
10059         /*
10060          * XXX make this a macro somewhere so it automatically gets
10061          * incremented when we make changes.
10062          */
10063         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10064             "revision")) == NULL) {
10065                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10066         } else {
10067                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10068                 strncpy(inq_ptr->revision, val,
10069                     min(sizeof(inq_ptr->revision), strlen(val)));
10070         }
10071
10072         /*
10073          * For parallel SCSI, we support double transition and single
10074          * transition clocking.  We also support QAS (Quick Arbitration
10075          * and Selection) and Information Unit transfers on both the
10076          * control and array devices.
10077          */
10078         if (port_type == CTL_PORT_SCSI)
10079                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10080                                     SID_SPI_IUS;
10081
10082         /* SAM-6 (no version claimed) */
10083         scsi_ulto2b(0x00C0, inq_ptr->version1);
10084         /* SPC-5 (no version claimed) */
10085         scsi_ulto2b(0x05C0, inq_ptr->version2);
10086         if (port_type == CTL_PORT_FC) {
10087                 /* FCP-2 ANSI INCITS.350:2003 */
10088                 scsi_ulto2b(0x0917, inq_ptr->version3);
10089         } else if (port_type == CTL_PORT_SCSI) {
10090                 /* SPI-4 ANSI INCITS.362:200x */
10091                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10092         } else if (port_type == CTL_PORT_ISCSI) {
10093                 /* iSCSI (no version claimed) */
10094                 scsi_ulto2b(0x0960, inq_ptr->version3);
10095         } else if (port_type == CTL_PORT_SAS) {
10096                 /* SAS (no version claimed) */
10097                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10098         } else if (port_type == CTL_PORT_UMASS) {
10099                 /* USB Mass Storage Class Bulk-Only Transport, Revision 1.0 */
10100                 scsi_ulto2b(0x1730, inq_ptr->version3);
10101         }
10102
10103         if (lun == NULL) {
10104                 /* SBC-4 (no version claimed) */
10105                 scsi_ulto2b(0x0600, inq_ptr->version4);
10106         } else {
10107                 switch (lun->be_lun->lun_type) {
10108                 case T_DIRECT:
10109                         /* SBC-4 (no version claimed) */
10110                         scsi_ulto2b(0x0600, inq_ptr->version4);
10111                         break;
10112                 case T_PROCESSOR:
10113                         break;
10114                 case T_CDROM:
10115                         /* MMC-6 (no version claimed) */
10116                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10117                         break;
10118                 default:
10119                         break;
10120                 }
10121         }
10122
10123         ctl_set_success(ctsio);
10124         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10125         ctsio->be_move_done = ctl_config_move_done;
10126         ctl_datamove((union ctl_io *)ctsio);
10127         return (CTL_RETVAL_COMPLETE);
10128 }
10129
10130 int
10131 ctl_inquiry(struct ctl_scsiio *ctsio)
10132 {
10133         struct scsi_inquiry *cdb;
10134         int retval;
10135
10136         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10137
10138         cdb = (struct scsi_inquiry *)ctsio->cdb;
10139         if (cdb->byte2 & SI_EVPD)
10140                 retval = ctl_inquiry_evpd(ctsio);
10141         else if (cdb->page_code == 0)
10142                 retval = ctl_inquiry_std(ctsio);
10143         else {
10144                 ctl_set_invalid_field(ctsio,
10145                                       /*sks_valid*/ 1,
10146                                       /*command*/ 1,
10147                                       /*field*/ 2,
10148                                       /*bit_valid*/ 0,
10149                                       /*bit*/ 0);
10150                 ctl_done((union ctl_io *)ctsio);
10151                 return (CTL_RETVAL_COMPLETE);
10152         }
10153
10154         return (retval);
10155 }
10156
10157 int
10158 ctl_get_config(struct ctl_scsiio *ctsio)
10159 {
10160         struct ctl_lun *lun = CTL_LUN(ctsio);
10161         struct scsi_get_config_header *hdr;
10162         struct scsi_get_config_feature *feature;
10163         struct scsi_get_config *cdb;
10164         uint32_t alloc_len, data_len;
10165         int rt, starting;
10166
10167         cdb = (struct scsi_get_config *)ctsio->cdb;
10168         rt = (cdb->rt & SGC_RT_MASK);
10169         starting = scsi_2btoul(cdb->starting_feature);
10170         alloc_len = scsi_2btoul(cdb->length);
10171
10172         data_len = sizeof(struct scsi_get_config_header) +
10173             sizeof(struct scsi_get_config_feature) + 8 +
10174             sizeof(struct scsi_get_config_feature) + 8 +
10175             sizeof(struct scsi_get_config_feature) + 4 +
10176             sizeof(struct scsi_get_config_feature) + 4 +
10177             sizeof(struct scsi_get_config_feature) + 8 +
10178             sizeof(struct scsi_get_config_feature) +
10179             sizeof(struct scsi_get_config_feature) + 4 +
10180             sizeof(struct scsi_get_config_feature) + 4 +
10181             sizeof(struct scsi_get_config_feature) + 4 +
10182             sizeof(struct scsi_get_config_feature) + 4 +
10183             sizeof(struct scsi_get_config_feature) + 4 +
10184             sizeof(struct scsi_get_config_feature) + 4;
10185         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10186         ctsio->kern_sg_entries = 0;
10187         ctsio->kern_rel_offset = 0;
10188
10189         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10190         if (lun->flags & CTL_LUN_NO_MEDIA)
10191                 scsi_ulto2b(0x0000, hdr->current_profile);
10192         else
10193                 scsi_ulto2b(0x0010, hdr->current_profile);
10194         feature = (struct scsi_get_config_feature *)(hdr + 1);
10195
10196         if (starting > 0x003b)
10197                 goto done;
10198         if (starting > 0x003a)
10199                 goto f3b;
10200         if (starting > 0x002b)
10201                 goto f3a;
10202         if (starting > 0x002a)
10203                 goto f2b;
10204         if (starting > 0x001f)
10205                 goto f2a;
10206         if (starting > 0x001e)
10207                 goto f1f;
10208         if (starting > 0x001d)
10209                 goto f1e;
10210         if (starting > 0x0010)
10211                 goto f1d;
10212         if (starting > 0x0003)
10213                 goto f10;
10214         if (starting > 0x0002)
10215                 goto f3;
10216         if (starting > 0x0001)
10217                 goto f2;
10218         if (starting > 0x0000)
10219                 goto f1;
10220
10221         /* Profile List */
10222         scsi_ulto2b(0x0000, feature->feature_code);
10223         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10224         feature->add_length = 8;
10225         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10226         feature->feature_data[2] = 0x00;
10227         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10228         feature->feature_data[6] = 0x01;
10229         feature = (struct scsi_get_config_feature *)
10230             &feature->feature_data[feature->add_length];
10231
10232 f1:     /* Core */
10233         scsi_ulto2b(0x0001, feature->feature_code);
10234         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10235         feature->add_length = 8;
10236         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10237         feature->feature_data[4] = 0x03;
10238         feature = (struct scsi_get_config_feature *)
10239             &feature->feature_data[feature->add_length];
10240
10241 f2:     /* Morphing */
10242         scsi_ulto2b(0x0002, feature->feature_code);
10243         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10244         feature->add_length = 4;
10245         feature->feature_data[0] = 0x02;
10246         feature = (struct scsi_get_config_feature *)
10247             &feature->feature_data[feature->add_length];
10248
10249 f3:     /* Removable Medium */
10250         scsi_ulto2b(0x0003, feature->feature_code);
10251         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10252         feature->add_length = 4;
10253         feature->feature_data[0] = 0x39;
10254         feature = (struct scsi_get_config_feature *)
10255             &feature->feature_data[feature->add_length];
10256
10257         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10258                 goto done;
10259
10260 f10:    /* Random Read */
10261         scsi_ulto2b(0x0010, feature->feature_code);
10262         feature->flags = 0x00;
10263         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10264                 feature->flags |= SGC_F_CURRENT;
10265         feature->add_length = 8;
10266         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10267         scsi_ulto2b(1, &feature->feature_data[4]);
10268         feature->feature_data[6] = 0x00;
10269         feature = (struct scsi_get_config_feature *)
10270             &feature->feature_data[feature->add_length];
10271
10272 f1d:    /* Multi-Read */
10273         scsi_ulto2b(0x001D, feature->feature_code);
10274         feature->flags = 0x00;
10275         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10276                 feature->flags |= SGC_F_CURRENT;
10277         feature->add_length = 0;
10278         feature = (struct scsi_get_config_feature *)
10279             &feature->feature_data[feature->add_length];
10280
10281 f1e:    /* CD Read */
10282         scsi_ulto2b(0x001E, feature->feature_code);
10283         feature->flags = 0x00;
10284         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10285                 feature->flags |= SGC_F_CURRENT;
10286         feature->add_length = 4;
10287         feature->feature_data[0] = 0x00;
10288         feature = (struct scsi_get_config_feature *)
10289             &feature->feature_data[feature->add_length];
10290
10291 f1f:    /* DVD Read */
10292         scsi_ulto2b(0x001F, feature->feature_code);
10293         feature->flags = 0x08;
10294         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10295                 feature->flags |= SGC_F_CURRENT;
10296         feature->add_length = 4;
10297         feature->feature_data[0] = 0x01;
10298         feature->feature_data[2] = 0x03;
10299         feature = (struct scsi_get_config_feature *)
10300             &feature->feature_data[feature->add_length];
10301
10302 f2a:    /* DVD+RW */
10303         scsi_ulto2b(0x002A, feature->feature_code);
10304         feature->flags = 0x04;
10305         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10306                 feature->flags |= SGC_F_CURRENT;
10307         feature->add_length = 4;
10308         feature->feature_data[0] = 0x00;
10309         feature->feature_data[1] = 0x00;
10310         feature = (struct scsi_get_config_feature *)
10311             &feature->feature_data[feature->add_length];
10312
10313 f2b:    /* DVD+R */
10314         scsi_ulto2b(0x002B, feature->feature_code);
10315         feature->flags = 0x00;
10316         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10317                 feature->flags |= SGC_F_CURRENT;
10318         feature->add_length = 4;
10319         feature->feature_data[0] = 0x00;
10320         feature = (struct scsi_get_config_feature *)
10321             &feature->feature_data[feature->add_length];
10322
10323 f3a:    /* DVD+RW Dual Layer */
10324         scsi_ulto2b(0x003A, feature->feature_code);
10325         feature->flags = 0x00;
10326         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10327                 feature->flags |= SGC_F_CURRENT;
10328         feature->add_length = 4;
10329         feature->feature_data[0] = 0x00;
10330         feature->feature_data[1] = 0x00;
10331         feature = (struct scsi_get_config_feature *)
10332             &feature->feature_data[feature->add_length];
10333
10334 f3b:    /* DVD+R Dual Layer */
10335         scsi_ulto2b(0x003B, feature->feature_code);
10336         feature->flags = 0x00;
10337         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10338                 feature->flags |= SGC_F_CURRENT;
10339         feature->add_length = 4;
10340         feature->feature_data[0] = 0x00;
10341         feature = (struct scsi_get_config_feature *)
10342             &feature->feature_data[feature->add_length];
10343
10344 done:
10345         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10346         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10347                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10348                 if (scsi_2btoul(feature->feature_code) == starting)
10349                         feature = (struct scsi_get_config_feature *)
10350                             &feature->feature_data[feature->add_length];
10351                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10352         }
10353         scsi_ulto4b(data_len - 4, hdr->data_length);
10354         ctsio->kern_data_len = min(data_len, alloc_len);
10355         ctsio->kern_total_len = ctsio->kern_data_len;
10356
10357         ctl_set_success(ctsio);
10358         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10359         ctsio->be_move_done = ctl_config_move_done;
10360         ctl_datamove((union ctl_io *)ctsio);
10361         return (CTL_RETVAL_COMPLETE);
10362 }
10363
10364 int
10365 ctl_get_event_status(struct ctl_scsiio *ctsio)
10366 {
10367         struct scsi_get_event_status_header *hdr;
10368         struct scsi_get_event_status *cdb;
10369         uint32_t alloc_len, data_len;
10370         int notif_class;
10371
10372         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10373         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10374                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10375                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10376                 ctl_done((union ctl_io *)ctsio);
10377                 return (CTL_RETVAL_COMPLETE);
10378         }
10379         notif_class = cdb->notif_class;
10380         alloc_len = scsi_2btoul(cdb->length);
10381
10382         data_len = sizeof(struct scsi_get_event_status_header);
10383         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10384         ctsio->kern_sg_entries = 0;
10385         ctsio->kern_rel_offset = 0;
10386         ctsio->kern_data_len = min(data_len, alloc_len);
10387         ctsio->kern_total_len = ctsio->kern_data_len;
10388
10389         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10390         scsi_ulto2b(0, hdr->descr_length);
10391         hdr->nea_class = SGESN_NEA;
10392         hdr->supported_class = 0;
10393
10394         ctl_set_success(ctsio);
10395         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10396         ctsio->be_move_done = ctl_config_move_done;
10397         ctl_datamove((union ctl_io *)ctsio);
10398         return (CTL_RETVAL_COMPLETE);
10399 }
10400
10401 int
10402 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10403 {
10404         struct scsi_mechanism_status_header *hdr;
10405         struct scsi_mechanism_status *cdb;
10406         uint32_t alloc_len, data_len;
10407
10408         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10409         alloc_len = scsi_2btoul(cdb->length);
10410
10411         data_len = sizeof(struct scsi_mechanism_status_header);
10412         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10413         ctsio->kern_sg_entries = 0;
10414         ctsio->kern_rel_offset = 0;
10415         ctsio->kern_data_len = min(data_len, alloc_len);
10416         ctsio->kern_total_len = ctsio->kern_data_len;
10417
10418         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10419         hdr->state1 = 0x00;
10420         hdr->state2 = 0xe0;
10421         scsi_ulto3b(0, hdr->lba);
10422         hdr->slots_num = 0;
10423         scsi_ulto2b(0, hdr->slots_length);
10424
10425         ctl_set_success(ctsio);
10426         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10427         ctsio->be_move_done = ctl_config_move_done;
10428         ctl_datamove((union ctl_io *)ctsio);
10429         return (CTL_RETVAL_COMPLETE);
10430 }
10431
10432 static void
10433 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10434 {
10435
10436         lba += 150;
10437         buf[0] = 0;
10438         buf[1] = bin2bcd((lba / 75) / 60);
10439         buf[2] = bin2bcd((lba / 75) % 60);
10440         buf[3] = bin2bcd(lba % 75);
10441 }
10442
10443 int
10444 ctl_read_toc(struct ctl_scsiio *ctsio)
10445 {
10446         struct ctl_lun *lun = CTL_LUN(ctsio);
10447         struct scsi_read_toc_hdr *hdr;
10448         struct scsi_read_toc_type01_descr *descr;
10449         struct scsi_read_toc *cdb;
10450         uint32_t alloc_len, data_len;
10451         int format, msf;
10452
10453         cdb = (struct scsi_read_toc *)ctsio->cdb;
10454         msf = (cdb->byte2 & CD_MSF) != 0;
10455         format = cdb->format;
10456         alloc_len = scsi_2btoul(cdb->data_len);
10457
10458         data_len = sizeof(struct scsi_read_toc_hdr);
10459         if (format == 0)
10460                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10461         else
10462                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10463         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10464         ctsio->kern_sg_entries = 0;
10465         ctsio->kern_rel_offset = 0;
10466         ctsio->kern_data_len = min(data_len, alloc_len);
10467         ctsio->kern_total_len = ctsio->kern_data_len;
10468
10469         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10470         if (format == 0) {
10471                 scsi_ulto2b(0x12, hdr->data_length);
10472                 hdr->first = 1;
10473                 hdr->last = 1;
10474                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10475                 descr->addr_ctl = 0x14;
10476                 descr->track_number = 1;
10477                 if (msf)
10478                         ctl_ultomsf(0, descr->track_start);
10479                 else
10480                         scsi_ulto4b(0, descr->track_start);
10481                 descr++;
10482                 descr->addr_ctl = 0x14;
10483                 descr->track_number = 0xaa;
10484                 if (msf)
10485                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10486                 else
10487                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10488         } else {
10489                 scsi_ulto2b(0x0a, hdr->data_length);
10490                 hdr->first = 1;
10491                 hdr->last = 1;
10492                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10493                 descr->addr_ctl = 0x14;
10494                 descr->track_number = 1;
10495                 if (msf)
10496                         ctl_ultomsf(0, descr->track_start);
10497                 else
10498                         scsi_ulto4b(0, descr->track_start);
10499         }
10500
10501         ctl_set_success(ctsio);
10502         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10503         ctsio->be_move_done = ctl_config_move_done;
10504         ctl_datamove((union ctl_io *)ctsio);
10505         return (CTL_RETVAL_COMPLETE);
10506 }
10507
10508 /*
10509  * For known CDB types, parse the LBA and length.
10510  */
10511 static int
10512 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10513 {
10514         if (io->io_hdr.io_type != CTL_IO_SCSI)
10515                 return (1);
10516
10517         switch (io->scsiio.cdb[0]) {
10518         case COMPARE_AND_WRITE: {
10519                 struct scsi_compare_and_write *cdb;
10520
10521                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10522
10523                 *lba = scsi_8btou64(cdb->addr);
10524                 *len = cdb->length;
10525                 break;
10526         }
10527         case READ_6:
10528         case WRITE_6: {
10529                 struct scsi_rw_6 *cdb;
10530
10531                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10532
10533                 *lba = scsi_3btoul(cdb->addr);
10534                 /* only 5 bits are valid in the most significant address byte */
10535                 *lba &= 0x1fffff;
10536                 *len = cdb->length;
10537                 break;
10538         }
10539         case READ_10:
10540         case WRITE_10: {
10541                 struct scsi_rw_10 *cdb;
10542
10543                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10544
10545                 *lba = scsi_4btoul(cdb->addr);
10546                 *len = scsi_2btoul(cdb->length);
10547                 break;
10548         }
10549         case WRITE_VERIFY_10: {
10550                 struct scsi_write_verify_10 *cdb;
10551
10552                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10553
10554                 *lba = scsi_4btoul(cdb->addr);
10555                 *len = scsi_2btoul(cdb->length);
10556                 break;
10557         }
10558         case READ_12:
10559         case WRITE_12: {
10560                 struct scsi_rw_12 *cdb;
10561
10562                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10563
10564                 *lba = scsi_4btoul(cdb->addr);
10565                 *len = scsi_4btoul(cdb->length);
10566                 break;
10567         }
10568         case WRITE_VERIFY_12: {
10569                 struct scsi_write_verify_12 *cdb;
10570
10571                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10572
10573                 *lba = scsi_4btoul(cdb->addr);
10574                 *len = scsi_4btoul(cdb->length);
10575                 break;
10576         }
10577         case READ_16:
10578         case WRITE_16: {
10579                 struct scsi_rw_16 *cdb;
10580
10581                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10582
10583                 *lba = scsi_8btou64(cdb->addr);
10584                 *len = scsi_4btoul(cdb->length);
10585                 break;
10586         }
10587         case WRITE_ATOMIC_16: {
10588                 struct scsi_write_atomic_16 *cdb;
10589
10590                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10591
10592                 *lba = scsi_8btou64(cdb->addr);
10593                 *len = scsi_2btoul(cdb->length);
10594                 break;
10595         }
10596         case WRITE_VERIFY_16: {
10597                 struct scsi_write_verify_16 *cdb;
10598
10599                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10600
10601                 *lba = scsi_8btou64(cdb->addr);
10602                 *len = scsi_4btoul(cdb->length);
10603                 break;
10604         }
10605         case WRITE_SAME_10: {
10606                 struct scsi_write_same_10 *cdb;
10607
10608                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10609
10610                 *lba = scsi_4btoul(cdb->addr);
10611                 *len = scsi_2btoul(cdb->length);
10612                 break;
10613         }
10614         case WRITE_SAME_16: {
10615                 struct scsi_write_same_16 *cdb;
10616
10617                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10618
10619                 *lba = scsi_8btou64(cdb->addr);
10620                 *len = scsi_4btoul(cdb->length);
10621                 break;
10622         }
10623         case VERIFY_10: {
10624                 struct scsi_verify_10 *cdb;
10625
10626                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10627
10628                 *lba = scsi_4btoul(cdb->addr);
10629                 *len = scsi_2btoul(cdb->length);
10630                 break;
10631         }
10632         case VERIFY_12: {
10633                 struct scsi_verify_12 *cdb;
10634
10635                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10636
10637                 *lba = scsi_4btoul(cdb->addr);
10638                 *len = scsi_4btoul(cdb->length);
10639                 break;
10640         }
10641         case VERIFY_16: {
10642                 struct scsi_verify_16 *cdb;
10643
10644                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10645
10646                 *lba = scsi_8btou64(cdb->addr);
10647                 *len = scsi_4btoul(cdb->length);
10648                 break;
10649         }
10650         case UNMAP: {
10651                 *lba = 0;
10652                 *len = UINT64_MAX;
10653                 break;
10654         }
10655         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10656                 struct scsi_get_lba_status *cdb;
10657
10658                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10659                 *lba = scsi_8btou64(cdb->addr);
10660                 *len = UINT32_MAX;
10661                 break;
10662         }
10663         default:
10664                 return (1);
10665                 break; /* NOTREACHED */
10666         }
10667
10668         return (0);
10669 }
10670
10671 static ctl_action
10672 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10673     bool seq)
10674 {
10675         uint64_t endlba1, endlba2;
10676
10677         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10678         endlba2 = lba2 + len2 - 1;
10679
10680         if ((endlba1 < lba2) || (endlba2 < lba1))
10681                 return (CTL_ACTION_PASS);
10682         else
10683                 return (CTL_ACTION_BLOCK);
10684 }
10685
10686 static int
10687 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10688 {
10689         struct ctl_ptr_len_flags *ptrlen;
10690         struct scsi_unmap_desc *buf, *end, *range;
10691         uint64_t lba;
10692         uint32_t len;
10693
10694         /* If not UNMAP -- go other way. */
10695         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10696             io->scsiio.cdb[0] != UNMAP)
10697                 return (CTL_ACTION_ERROR);
10698
10699         /* If UNMAP without data -- block and wait for data. */
10700         ptrlen = (struct ctl_ptr_len_flags *)
10701             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10702         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10703             ptrlen->ptr == NULL)
10704                 return (CTL_ACTION_BLOCK);
10705
10706         /* UNMAP with data -- check for collision. */
10707         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10708         end = buf + ptrlen->len / sizeof(*buf);
10709         for (range = buf; range < end; range++) {
10710                 lba = scsi_8btou64(range->lba);
10711                 len = scsi_4btoul(range->length);
10712                 if ((lba < lba2 + len2) && (lba + len > lba2))
10713                         return (CTL_ACTION_BLOCK);
10714         }
10715         return (CTL_ACTION_PASS);
10716 }
10717
10718 static ctl_action
10719 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10720 {
10721         uint64_t lba1, lba2;
10722         uint64_t len1, len2;
10723         int retval;
10724
10725         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10726                 return (CTL_ACTION_ERROR);
10727
10728         retval = ctl_extent_check_unmap(io1, lba2, len2);
10729         if (retval != CTL_ACTION_ERROR)
10730                 return (retval);
10731
10732         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10733                 return (CTL_ACTION_ERROR);
10734
10735         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10736                 seq = FALSE;
10737         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10738 }
10739
10740 static ctl_action
10741 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10742 {
10743         uint64_t lba1, lba2;
10744         uint64_t len1, len2;
10745
10746         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10747                 return (CTL_ACTION_PASS);
10748         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10749                 return (CTL_ACTION_ERROR);
10750         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10751                 return (CTL_ACTION_ERROR);
10752
10753         if (lba1 + len1 == lba2)
10754                 return (CTL_ACTION_BLOCK);
10755         return (CTL_ACTION_PASS);
10756 }
10757
10758 static ctl_action
10759 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10760     union ctl_io *ooa_io)
10761 {
10762         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10763         const ctl_serialize_action *serialize_row;
10764
10765         /*
10766          * The initiator attempted multiple untagged commands at the same
10767          * time.  Can't do that.
10768          */
10769         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10770          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10771          && ((pending_io->io_hdr.nexus.targ_port ==
10772               ooa_io->io_hdr.nexus.targ_port)
10773           && (pending_io->io_hdr.nexus.initid ==
10774               ooa_io->io_hdr.nexus.initid))
10775          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10776               CTL_FLAG_STATUS_SENT)) == 0))
10777                 return (CTL_ACTION_OVERLAP);
10778
10779         /*
10780          * The initiator attempted to send multiple tagged commands with
10781          * the same ID.  (It's fine if different initiators have the same
10782          * tag ID.)
10783          *
10784          * Even if all of those conditions are true, we don't kill the I/O
10785          * if the command ahead of us has been aborted.  We won't end up
10786          * sending it to the FETD, and it's perfectly legal to resend a
10787          * command with the same tag number as long as the previous
10788          * instance of this tag number has been aborted somehow.
10789          */
10790         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10791          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10792          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10793          && ((pending_io->io_hdr.nexus.targ_port ==
10794               ooa_io->io_hdr.nexus.targ_port)
10795           && (pending_io->io_hdr.nexus.initid ==
10796               ooa_io->io_hdr.nexus.initid))
10797          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10798               CTL_FLAG_STATUS_SENT)) == 0))
10799                 return (CTL_ACTION_OVERLAP_TAG);
10800
10801         /*
10802          * If we get a head of queue tag, SAM-3 says that we should
10803          * immediately execute it.
10804          *
10805          * What happens if this command would normally block for some other
10806          * reason?  e.g. a request sense with a head of queue tag
10807          * immediately after a write.  Normally that would block, but this
10808          * will result in its getting executed immediately...
10809          *
10810          * We currently return "pass" instead of "skip", so we'll end up
10811          * going through the rest of the queue to check for overlapped tags.
10812          *
10813          * XXX KDM check for other types of blockage first??
10814          */
10815         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10816                 return (CTL_ACTION_PASS);
10817
10818         /*
10819          * Ordered tags have to block until all items ahead of them
10820          * have completed.  If we get called with an ordered tag, we always
10821          * block, if something else is ahead of us in the queue.
10822          */
10823         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10824                 return (CTL_ACTION_BLOCK);
10825
10826         /*
10827          * Simple tags get blocked until all head of queue and ordered tags
10828          * ahead of them have completed.  I'm lumping untagged commands in
10829          * with simple tags here.  XXX KDM is that the right thing to do?
10830          */
10831         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10832           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10833          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10834           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10835                 return (CTL_ACTION_BLOCK);
10836
10837         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10838         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
10839             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
10840              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
10841              pending_io->scsiio.cdb[1], pending_io));
10842         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10843         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
10844                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
10845         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
10846             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
10847              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
10848              ooa_io->scsiio.cdb[1], ooa_io));
10849
10850         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10851
10852         switch (serialize_row[pending_entry->seridx]) {
10853         case CTL_SER_BLOCK:
10854                 return (CTL_ACTION_BLOCK);
10855         case CTL_SER_EXTENT:
10856                 return (ctl_extent_check(ooa_io, pending_io,
10857                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10858         case CTL_SER_EXTENTOPT:
10859                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10860                     SCP_QUEUE_ALG_UNRESTRICTED)
10861                         return (ctl_extent_check(ooa_io, pending_io,
10862                             (lun->be_lun &&
10863                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10864                 return (CTL_ACTION_PASS);
10865         case CTL_SER_EXTENTSEQ:
10866                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10867                         return (ctl_extent_check_seq(ooa_io, pending_io));
10868                 return (CTL_ACTION_PASS);
10869         case CTL_SER_PASS:
10870                 return (CTL_ACTION_PASS);
10871         case CTL_SER_BLOCKOPT:
10872                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10873                     SCP_QUEUE_ALG_UNRESTRICTED)
10874                         return (CTL_ACTION_BLOCK);
10875                 return (CTL_ACTION_PASS);
10876         case CTL_SER_SKIP:
10877                 return (CTL_ACTION_SKIP);
10878         default:
10879                 panic("%s: Invalid serialization value %d for %d => %d",
10880                     __func__, serialize_row[pending_entry->seridx],
10881                     pending_entry->seridx, ooa_entry->seridx);
10882         }
10883
10884         return (CTL_ACTION_ERROR);
10885 }
10886
10887 /*
10888  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10889  * Assumptions:
10890  * - pending_io is generally either incoming, or on the blocked queue
10891  * - starting I/O is the I/O we want to start the check with.
10892  */
10893 static ctl_action
10894 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10895               union ctl_io *starting_io)
10896 {
10897         union ctl_io *ooa_io;
10898         ctl_action action;
10899
10900         mtx_assert(&lun->lun_lock, MA_OWNED);
10901
10902         /*
10903          * Run back along the OOA queue, starting with the current
10904          * blocked I/O and going through every I/O before it on the
10905          * queue.  If starting_io is NULL, we'll just end up returning
10906          * CTL_ACTION_PASS.
10907          */
10908         for (ooa_io = starting_io; ooa_io != NULL;
10909              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10910              ooa_links)){
10911
10912                 /*
10913                  * This routine just checks to see whether
10914                  * cur_blocked is blocked by ooa_io, which is ahead
10915                  * of it in the queue.  It doesn't queue/dequeue
10916                  * cur_blocked.
10917                  */
10918                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10919                 switch (action) {
10920                 case CTL_ACTION_BLOCK:
10921                 case CTL_ACTION_OVERLAP:
10922                 case CTL_ACTION_OVERLAP_TAG:
10923                 case CTL_ACTION_SKIP:
10924                 case CTL_ACTION_ERROR:
10925                         return (action);
10926                         break; /* NOTREACHED */
10927                 case CTL_ACTION_PASS:
10928                         break;
10929                 default:
10930                         panic("%s: Invalid action %d\n", __func__, action);
10931                 }
10932         }
10933
10934         return (CTL_ACTION_PASS);
10935 }
10936
10937 /*
10938  * Assumptions:
10939  * - An I/O has just completed, and has been removed from the per-LUN OOA
10940  *   queue, so some items on the blocked queue may now be unblocked.
10941  */
10942 static int
10943 ctl_check_blocked(struct ctl_lun *lun)
10944 {
10945         struct ctl_softc *softc = lun->ctl_softc;
10946         union ctl_io *cur_blocked, *next_blocked;
10947
10948         mtx_assert(&lun->lun_lock, MA_OWNED);
10949
10950         /*
10951          * Run forward from the head of the blocked queue, checking each
10952          * entry against the I/Os prior to it on the OOA queue to see if
10953          * there is still any blockage.
10954          *
10955          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10956          * with our removing a variable on it while it is traversing the
10957          * list.
10958          */
10959         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10960              cur_blocked != NULL; cur_blocked = next_blocked) {
10961                 union ctl_io *prev_ooa;
10962                 ctl_action action;
10963
10964                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10965                                                           blocked_links);
10966
10967                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10968                                                       ctl_ooaq, ooa_links);
10969
10970                 /*
10971                  * If cur_blocked happens to be the first item in the OOA
10972                  * queue now, prev_ooa will be NULL, and the action
10973                  * returned will just be CTL_ACTION_PASS.
10974                  */
10975                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10976
10977                 switch (action) {
10978                 case CTL_ACTION_BLOCK:
10979                         /* Nothing to do here, still blocked */
10980                         break;
10981                 case CTL_ACTION_OVERLAP:
10982                 case CTL_ACTION_OVERLAP_TAG:
10983                         /*
10984                          * This shouldn't happen!  In theory we've already
10985                          * checked this command for overlap...
10986                          */
10987                         break;
10988                 case CTL_ACTION_PASS:
10989                 case CTL_ACTION_SKIP: {
10990                         const struct ctl_cmd_entry *entry;
10991
10992                         /*
10993                          * The skip case shouldn't happen, this transaction
10994                          * should have never made it onto the blocked queue.
10995                          */
10996                         /*
10997                          * This I/O is no longer blocked, we can remove it
10998                          * from the blocked queue.  Since this is a TAILQ
10999                          * (doubly linked list), we can do O(1) removals
11000                          * from any place on the list.
11001                          */
11002                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11003                                      blocked_links);
11004                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11005
11006                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11007                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11008                                 /*
11009                                  * Need to send IO back to original side to
11010                                  * run
11011                                  */
11012                                 union ctl_ha_msg msg_info;
11013
11014                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11015                                 msg_info.hdr.original_sc =
11016                                         cur_blocked->io_hdr.original_sc;
11017                                 msg_info.hdr.serializing_sc = cur_blocked;
11018                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11019                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11020                                     sizeof(msg_info.hdr), M_NOWAIT);
11021                                 break;
11022                         }
11023                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11024
11025                         /*
11026                          * Check this I/O for LUN state changes that may
11027                          * have happened while this command was blocked.
11028                          * The LUN state may have been changed by a command
11029                          * ahead of us in the queue, so we need to re-check
11030                          * for any states that can be caused by SCSI
11031                          * commands.
11032                          */
11033                         if (ctl_scsiio_lun_check(lun, entry,
11034                                                  &cur_blocked->scsiio) == 0) {
11035                                 cur_blocked->io_hdr.flags |=
11036                                                       CTL_FLAG_IS_WAS_ON_RTR;
11037                                 ctl_enqueue_rtr(cur_blocked);
11038                         } else
11039                                 ctl_done(cur_blocked);
11040                         break;
11041                 }
11042                 default:
11043                         /*
11044                          * This probably shouldn't happen -- we shouldn't
11045                          * get CTL_ACTION_ERROR, or anything else.
11046                          */
11047                         break;
11048                 }
11049         }
11050
11051         return (CTL_RETVAL_COMPLETE);
11052 }
11053
11054 /*
11055  * This routine (with one exception) checks LUN flags that can be set by
11056  * commands ahead of us in the OOA queue.  These flags have to be checked
11057  * when a command initially comes in, and when we pull a command off the
11058  * blocked queue and are preparing to execute it.  The reason we have to
11059  * check these flags for commands on the blocked queue is that the LUN
11060  * state may have been changed by a command ahead of us while we're on the
11061  * blocked queue.
11062  *
11063  * Ordering is somewhat important with these checks, so please pay
11064  * careful attention to the placement of any new checks.
11065  */
11066 static int
11067 ctl_scsiio_lun_check(struct ctl_lun *lun,
11068     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11069 {
11070         struct ctl_softc *softc = lun->ctl_softc;
11071         int retval;
11072         uint32_t residx;
11073
11074         retval = 0;
11075
11076         mtx_assert(&lun->lun_lock, MA_OWNED);
11077
11078         /*
11079          * If this shelf is a secondary shelf controller, we may have to
11080          * reject some commands disallowed by HA mode and link state.
11081          */
11082         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11083                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11084                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11085                         ctl_set_lun_unavail(ctsio);
11086                         retval = 1;
11087                         goto bailout;
11088                 }
11089                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11090                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11091                         ctl_set_lun_transit(ctsio);
11092                         retval = 1;
11093                         goto bailout;
11094                 }
11095                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11096                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11097                         ctl_set_lun_standby(ctsio);
11098                         retval = 1;
11099                         goto bailout;
11100                 }
11101
11102                 /* The rest of checks are only done on executing side */
11103                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11104                         goto bailout;
11105         }
11106
11107         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11108                 if (lun->be_lun &&
11109                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11110                         ctl_set_hw_write_protected(ctsio);
11111                         retval = 1;
11112                         goto bailout;
11113                 }
11114                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11115                         ctl_set_sense(ctsio, /*current_error*/ 1,
11116                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11117                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11118                         retval = 1;
11119                         goto bailout;
11120                 }
11121         }
11122
11123         /*
11124          * Check for a reservation conflict.  If this command isn't allowed
11125          * even on reserved LUNs, and if this initiator isn't the one who
11126          * reserved us, reject the command with a reservation conflict.
11127          */
11128         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11129         if ((lun->flags & CTL_LUN_RESERVED)
11130          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11131                 if (lun->res_idx != residx) {
11132                         ctl_set_reservation_conflict(ctsio);
11133                         retval = 1;
11134                         goto bailout;
11135                 }
11136         }
11137
11138         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11139             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11140                 /* No reservation or command is allowed. */;
11141         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11142             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11143              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11144              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11145                 /* The command is allowed for Write Exclusive resv. */;
11146         } else {
11147                 /*
11148                  * if we aren't registered or it's a res holder type
11149                  * reservation and this isn't the res holder then set a
11150                  * conflict.
11151                  */
11152                 if (ctl_get_prkey(lun, residx) == 0 ||
11153                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11154                         ctl_set_reservation_conflict(ctsio);
11155                         retval = 1;
11156                         goto bailout;
11157                 }
11158         }
11159
11160         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11161                 if (lun->flags & CTL_LUN_EJECTED)
11162                         ctl_set_lun_ejected(ctsio);
11163                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11164                         if (lun->flags & CTL_LUN_REMOVABLE)
11165                                 ctl_set_lun_no_media(ctsio);
11166                         else
11167                                 ctl_set_lun_int_reqd(ctsio);
11168                 } else if (lun->flags & CTL_LUN_STOPPED)
11169                         ctl_set_lun_stopped(ctsio);
11170                 else
11171                         goto bailout;
11172                 retval = 1;
11173                 goto bailout;
11174         }
11175
11176 bailout:
11177         return (retval);
11178 }
11179
11180 static void
11181 ctl_failover_io(union ctl_io *io, int have_lock)
11182 {
11183         ctl_set_busy(&io->scsiio);
11184         ctl_done(io);
11185 }
11186
11187 static void
11188 ctl_failover_lun(union ctl_io *rio)
11189 {
11190         struct ctl_softc *softc = CTL_SOFTC(rio);
11191         struct ctl_lun *lun;
11192         struct ctl_io_hdr *io, *next_io;
11193         uint32_t targ_lun;
11194
11195         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11196         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11197
11198         /* Find and lock the LUN. */
11199         mtx_lock(&softc->ctl_lock);
11200         if (targ_lun > CTL_MAX_LUNS ||
11201             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11202                 mtx_unlock(&softc->ctl_lock);
11203                 return;
11204         }
11205         mtx_lock(&lun->lun_lock);
11206         mtx_unlock(&softc->ctl_lock);
11207         if (lun->flags & CTL_LUN_DISABLED) {
11208                 mtx_unlock(&lun->lun_lock);
11209                 return;
11210         }
11211
11212         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11213                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11214                         /* We are master */
11215                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11216                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11217                                         io->flags |= CTL_FLAG_ABORT;
11218                                         io->flags |= CTL_FLAG_FAILOVER;
11219                                 } else { /* This can be only due to DATAMOVE */
11220                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11221                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11222                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11223                                         io->port_status = 31340;
11224                                         ctl_enqueue_isc((union ctl_io *)io);
11225                                 }
11226                         }
11227                         /* We are slave */
11228                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11229                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11230                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11231                                         io->flags |= CTL_FLAG_FAILOVER;
11232                                 } else {
11233                                         ctl_set_busy(&((union ctl_io *)io)->
11234                                             scsiio);
11235                                         ctl_done((union ctl_io *)io);
11236                                 }
11237                         }
11238                 }
11239         } else { /* SERIALIZE modes */
11240                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11241                     next_io) {
11242                         /* We are master */
11243                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11244                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11245                                     blocked_links);
11246                                 io->flags &= ~CTL_FLAG_BLOCKED;
11247                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11248                                 ctl_free_io((union ctl_io *)io);
11249                         }
11250                 }
11251                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11252                         /* We are master */
11253                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11254                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11255                                 ctl_free_io((union ctl_io *)io);
11256                         }
11257                         /* We are slave */
11258                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11259                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11260                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11261                                         ctl_set_busy(&((union ctl_io *)io)->
11262                                             scsiio);
11263                                         ctl_done((union ctl_io *)io);
11264                                 }
11265                         }
11266                 }
11267                 ctl_check_blocked(lun);
11268         }
11269         mtx_unlock(&lun->lun_lock);
11270 }
11271
11272 static int
11273 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11274 {
11275         struct ctl_lun *lun;
11276         const struct ctl_cmd_entry *entry;
11277         uint32_t initidx, targ_lun;
11278         int retval = 0;
11279
11280         lun = NULL;
11281         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11282         if (targ_lun < CTL_MAX_LUNS)
11283                 lun = softc->ctl_luns[targ_lun];
11284         if (lun) {
11285                 /*
11286                  * If the LUN is invalid, pretend that it doesn't exist.
11287                  * It will go away as soon as all pending I/O has been
11288                  * completed.
11289                  */
11290                 mtx_lock(&lun->lun_lock);
11291                 if (lun->flags & CTL_LUN_DISABLED) {
11292                         mtx_unlock(&lun->lun_lock);
11293                         lun = NULL;
11294                 }
11295         }
11296         CTL_LUN(ctsio) = lun;
11297         if (lun) {
11298                 CTL_BACKEND_LUN(ctsio) = lun->be_lun;
11299
11300                 /*
11301                  * Every I/O goes into the OOA queue for a particular LUN,
11302                  * and stays there until completion.
11303                  */
11304 #ifdef CTL_TIME_IO
11305                 if (TAILQ_EMPTY(&lun->ooa_queue))
11306                         lun->idle_time += getsbinuptime() - lun->last_busy;
11307 #endif
11308                 TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11309         }
11310
11311         /* Get command entry and return error if it is unsuppotyed. */
11312         entry = ctl_validate_command(ctsio);
11313         if (entry == NULL) {
11314                 if (lun)
11315                         mtx_unlock(&lun->lun_lock);
11316                 return (retval);
11317         }
11318
11319         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11320         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11321
11322         /*
11323          * Check to see whether we can send this command to LUNs that don't
11324          * exist.  This should pretty much only be the case for inquiry
11325          * and request sense.  Further checks, below, really require having
11326          * a LUN, so we can't really check the command anymore.  Just put
11327          * it on the rtr queue.
11328          */
11329         if (lun == NULL) {
11330                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11331                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11332                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11333                         return (retval);
11334                 }
11335
11336                 ctl_set_unsupported_lun(ctsio);
11337                 ctl_done((union ctl_io *)ctsio);
11338                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11339                 return (retval);
11340         } else {
11341                 /*
11342                  * Make sure we support this particular command on this LUN.
11343                  * e.g., we don't support writes to the control LUN.
11344                  */
11345                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11346                         mtx_unlock(&lun->lun_lock);
11347                         ctl_set_invalid_opcode(ctsio);
11348                         ctl_done((union ctl_io *)ctsio);
11349                         return (retval);
11350                 }
11351         }
11352
11353         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11354
11355         /*
11356          * If we've got a request sense, it'll clear the contingent
11357          * allegiance condition.  Otherwise, if we have a CA condition for
11358          * this initiator, clear it, because it sent down a command other
11359          * than request sense.
11360          */
11361         if (ctsio->cdb[0] != REQUEST_SENSE) {
11362                 struct scsi_sense_data *ps;
11363
11364                 ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
11365                 if (ps != NULL)
11366                         ps[initidx % CTL_MAX_INIT_PER_PORT].error_code = 0;
11367         }
11368
11369         /*
11370          * If the command has this flag set, it handles its own unit
11371          * attention reporting, we shouldn't do anything.  Otherwise we
11372          * check for any pending unit attentions, and send them back to the
11373          * initiator.  We only do this when a command initially comes in,
11374          * not when we pull it off the blocked queue.
11375          *
11376          * According to SAM-3, section 5.3.2, the order that things get
11377          * presented back to the host is basically unit attentions caused
11378          * by some sort of reset event, busy status, reservation conflicts
11379          * or task set full, and finally any other status.
11380          *
11381          * One issue here is that some of the unit attentions we report
11382          * don't fall into the "reset" category (e.g. "reported luns data
11383          * has changed").  So reporting it here, before the reservation
11384          * check, may be technically wrong.  I guess the only thing to do
11385          * would be to check for and report the reset events here, and then
11386          * check for the other unit attention types after we check for a
11387          * reservation conflict.
11388          *
11389          * XXX KDM need to fix this
11390          */
11391         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11392                 ctl_ua_type ua_type;
11393                 u_int sense_len = 0;
11394
11395                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11396                     &sense_len, SSD_TYPE_NONE);
11397                 if (ua_type != CTL_UA_NONE) {
11398                         mtx_unlock(&lun->lun_lock);
11399                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11400                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11401                         ctsio->sense_len = sense_len;
11402                         ctl_done((union ctl_io *)ctsio);
11403                         return (retval);
11404                 }
11405         }
11406
11407
11408         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11409                 mtx_unlock(&lun->lun_lock);
11410                 ctl_done((union ctl_io *)ctsio);
11411                 return (retval);
11412         }
11413
11414         /*
11415          * XXX CHD this is where we want to send IO to other side if
11416          * this LUN is secondary on this SC. We will need to make a copy
11417          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11418          * the copy we send as FROM_OTHER.
11419          * We also need to stuff the address of the original IO so we can
11420          * find it easily. Something similar will need be done on the other
11421          * side so when we are done we can find the copy.
11422          */
11423         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11424             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11425             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11426                 union ctl_ha_msg msg_info;
11427                 int isc_retval;
11428
11429                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11430                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11431                 mtx_unlock(&lun->lun_lock);
11432
11433                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11434                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11435                 msg_info.hdr.serializing_sc = NULL;
11436                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11437                 msg_info.scsi.tag_num = ctsio->tag_num;
11438                 msg_info.scsi.tag_type = ctsio->tag_type;
11439                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11440                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11441
11442                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11443                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11444                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11445                         ctl_set_busy(ctsio);
11446                         ctl_done((union ctl_io *)ctsio);
11447                         return (retval);
11448                 }
11449                 return (retval);
11450         }
11451
11452         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11453                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11454                               ctl_ooaq, ooa_links))) {
11455         case CTL_ACTION_BLOCK:
11456                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11457                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11458                                   blocked_links);
11459                 mtx_unlock(&lun->lun_lock);
11460                 return (retval);
11461         case CTL_ACTION_PASS:
11462         case CTL_ACTION_SKIP:
11463                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11464                 mtx_unlock(&lun->lun_lock);
11465                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11466                 break;
11467         case CTL_ACTION_OVERLAP:
11468                 mtx_unlock(&lun->lun_lock);
11469                 ctl_set_overlapped_cmd(ctsio);
11470                 ctl_done((union ctl_io *)ctsio);
11471                 break;
11472         case CTL_ACTION_OVERLAP_TAG:
11473                 mtx_unlock(&lun->lun_lock);
11474                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11475                 ctl_done((union ctl_io *)ctsio);
11476                 break;
11477         case CTL_ACTION_ERROR:
11478         default:
11479                 mtx_unlock(&lun->lun_lock);
11480                 ctl_set_internal_failure(ctsio,
11481                                          /*sks_valid*/ 0,
11482                                          /*retry_count*/ 0);
11483                 ctl_done((union ctl_io *)ctsio);
11484                 break;
11485         }
11486         return (retval);
11487 }
11488
11489 const struct ctl_cmd_entry *
11490 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11491 {
11492         const struct ctl_cmd_entry *entry;
11493         int service_action;
11494
11495         entry = &ctl_cmd_table[ctsio->cdb[0]];
11496         if (sa)
11497                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11498         if (entry->flags & CTL_CMD_FLAG_SA5) {
11499                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11500                 entry = &((const struct ctl_cmd_entry *)
11501                     entry->execute)[service_action];
11502         }
11503         return (entry);
11504 }
11505
11506 const struct ctl_cmd_entry *
11507 ctl_validate_command(struct ctl_scsiio *ctsio)
11508 {
11509         const struct ctl_cmd_entry *entry;
11510         int i, sa;
11511         uint8_t diff;
11512
11513         entry = ctl_get_cmd_entry(ctsio, &sa);
11514         if (entry->execute == NULL) {
11515                 if (sa)
11516                         ctl_set_invalid_field(ctsio,
11517                                               /*sks_valid*/ 1,
11518                                               /*command*/ 1,
11519                                               /*field*/ 1,
11520                                               /*bit_valid*/ 1,
11521                                               /*bit*/ 4);
11522                 else
11523                         ctl_set_invalid_opcode(ctsio);
11524                 ctl_done((union ctl_io *)ctsio);
11525                 return (NULL);
11526         }
11527         KASSERT(entry->length > 0,
11528             ("Not defined length for command 0x%02x/0x%02x",
11529              ctsio->cdb[0], ctsio->cdb[1]));
11530         for (i = 1; i < entry->length; i++) {
11531                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11532                 if (diff == 0)
11533                         continue;
11534                 ctl_set_invalid_field(ctsio,
11535                                       /*sks_valid*/ 1,
11536                                       /*command*/ 1,
11537                                       /*field*/ i,
11538                                       /*bit_valid*/ 1,
11539                                       /*bit*/ fls(diff) - 1);
11540                 ctl_done((union ctl_io *)ctsio);
11541                 return (NULL);
11542         }
11543         return (entry);
11544 }
11545
11546 static int
11547 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11548 {
11549
11550         switch (lun_type) {
11551         case T_DIRECT:
11552                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11553                         return (0);
11554                 break;
11555         case T_PROCESSOR:
11556                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11557                         return (0);
11558                 break;
11559         case T_CDROM:
11560                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11561                         return (0);
11562                 break;
11563         default:
11564                 return (0);
11565         }
11566         return (1);
11567 }
11568
11569 static int
11570 ctl_scsiio(struct ctl_scsiio *ctsio)
11571 {
11572         int retval;
11573         const struct ctl_cmd_entry *entry;
11574
11575         retval = CTL_RETVAL_COMPLETE;
11576
11577         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11578
11579         entry = ctl_get_cmd_entry(ctsio, NULL);
11580
11581         /*
11582          * If this I/O has been aborted, just send it straight to
11583          * ctl_done() without executing it.
11584          */
11585         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11586                 ctl_done((union ctl_io *)ctsio);
11587                 goto bailout;
11588         }
11589
11590         /*
11591          * All the checks should have been handled by ctl_scsiio_precheck().
11592          * We should be clear now to just execute the I/O.
11593          */
11594         retval = entry->execute(ctsio);
11595
11596 bailout:
11597         return (retval);
11598 }
11599
11600 static int
11601 ctl_target_reset(union ctl_io *io)
11602 {
11603         struct ctl_softc *softc = CTL_SOFTC(io);
11604         struct ctl_port *port = CTL_PORT(io);
11605         struct ctl_lun *lun;
11606         uint32_t initidx;
11607         ctl_ua_type ua_type;
11608
11609         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11610                 union ctl_ha_msg msg_info;
11611
11612                 msg_info.hdr.nexus = io->io_hdr.nexus;
11613                 msg_info.task.task_action = io->taskio.task_action;
11614                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11615                 msg_info.hdr.original_sc = NULL;
11616                 msg_info.hdr.serializing_sc = NULL;
11617                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11618                     sizeof(msg_info.task), M_WAITOK);
11619         }
11620
11621         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11622         if (io->taskio.task_action == CTL_TASK_TARGET_RESET)
11623                 ua_type = CTL_UA_TARG_RESET;
11624         else
11625                 ua_type = CTL_UA_BUS_RESET;
11626         mtx_lock(&softc->ctl_lock);
11627         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11628                 if (port != NULL &&
11629                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
11630                         continue;
11631                 ctl_do_lun_reset(lun, initidx, ua_type);
11632         }
11633         mtx_unlock(&softc->ctl_lock);
11634         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11635         return (0);
11636 }
11637
11638 /*
11639  * The LUN should always be set.  The I/O is optional, and is used to
11640  * distinguish between I/Os sent by this initiator, and by other
11641  * initiators.  We set unit attention for initiators other than this one.
11642  * SAM-3 is vague on this point.  It does say that a unit attention should
11643  * be established for other initiators when a LUN is reset (see section
11644  * 5.7.3), but it doesn't specifically say that the unit attention should
11645  * be established for this particular initiator when a LUN is reset.  Here
11646  * is the relevant text, from SAM-3 rev 8:
11647  *
11648  * 5.7.2 When a SCSI initiator port aborts its own tasks
11649  *
11650  * When a SCSI initiator port causes its own task(s) to be aborted, no
11651  * notification that the task(s) have been aborted shall be returned to
11652  * the SCSI initiator port other than the completion response for the
11653  * command or task management function action that caused the task(s) to
11654  * be aborted and notification(s) associated with related effects of the
11655  * action (e.g., a reset unit attention condition).
11656  *
11657  * XXX KDM for now, we're setting unit attention for all initiators.
11658  */
11659 static void
11660 ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua_type)
11661 {
11662         union ctl_io *xio;
11663         int i;
11664
11665         mtx_lock(&lun->lun_lock);
11666         /* Abort tasks. */
11667         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11668              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11669                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11670         }
11671         /* Clear CA. */
11672         for (i = 0; i < CTL_MAX_PORTS; i++) {
11673                 free(lun->pending_sense[i], M_CTL);
11674                 lun->pending_sense[i] = NULL;
11675         }
11676         /* Clear reservation. */
11677         lun->flags &= ~CTL_LUN_RESERVED;
11678         /* Clear prevent media removal. */
11679         if (lun->prevent) {
11680                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
11681                         ctl_clear_mask(lun->prevent, i);
11682                 lun->prevent_count = 0;
11683         }
11684         /* Clear TPC status */
11685         ctl_tpc_lun_clear(lun, -1);
11686         /* Establish UA. */
11687 #if 0
11688         ctl_est_ua_all(lun, initidx, ua_type);
11689 #else
11690         ctl_est_ua_all(lun, -1, ua_type);
11691 #endif
11692         mtx_unlock(&lun->lun_lock);
11693 }
11694
11695 static int
11696 ctl_lun_reset(union ctl_io *io)
11697 {
11698         struct ctl_softc *softc = CTL_SOFTC(io);
11699         struct ctl_lun *lun;
11700         uint32_t targ_lun, initidx;
11701
11702         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11703         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11704         mtx_lock(&softc->ctl_lock);
11705         if (targ_lun >= CTL_MAX_LUNS ||
11706             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11707                 mtx_unlock(&softc->ctl_lock);
11708                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11709                 return (1);
11710         }
11711         ctl_do_lun_reset(lun, initidx, CTL_UA_LUN_RESET);
11712         mtx_unlock(&softc->ctl_lock);
11713         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11714
11715         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11716                 union ctl_ha_msg msg_info;
11717
11718                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11719                 msg_info.hdr.nexus = io->io_hdr.nexus;
11720                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11721                 msg_info.hdr.original_sc = NULL;
11722                 msg_info.hdr.serializing_sc = NULL;
11723                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11724                     sizeof(msg_info.task), M_WAITOK);
11725         }
11726         return (0);
11727 }
11728
11729 static void
11730 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11731     int other_sc)
11732 {
11733         union ctl_io *xio;
11734
11735         mtx_assert(&lun->lun_lock, MA_OWNED);
11736
11737         /*
11738          * Run through the OOA queue and attempt to find the given I/O.
11739          * The target port, initiator ID, tag type and tag number have to
11740          * match the values that we got from the initiator.  If we have an
11741          * untagged command to abort, simply abort the first untagged command
11742          * we come to.  We only allow one untagged command at a time of course.
11743          */
11744         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11745              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11746
11747                 if ((targ_port == UINT32_MAX ||
11748                      targ_port == xio->io_hdr.nexus.targ_port) &&
11749                     (init_id == UINT32_MAX ||
11750                      init_id == xio->io_hdr.nexus.initid)) {
11751                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11752                             init_id != xio->io_hdr.nexus.initid)
11753                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11754                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11755                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11756                                 union ctl_ha_msg msg_info;
11757
11758                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11759                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11760                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11761                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11762                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11763                                 msg_info.hdr.original_sc = NULL;
11764                                 msg_info.hdr.serializing_sc = NULL;
11765                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11766                                     sizeof(msg_info.task), M_NOWAIT);
11767                         }
11768                 }
11769         }
11770 }
11771
11772 static int
11773 ctl_abort_task_set(union ctl_io *io)
11774 {
11775         struct ctl_softc *softc = CTL_SOFTC(io);
11776         struct ctl_lun *lun;
11777         uint32_t targ_lun;
11778
11779         /*
11780          * Look up the LUN.
11781          */
11782         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11783         mtx_lock(&softc->ctl_lock);
11784         if (targ_lun >= CTL_MAX_LUNS ||
11785             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11786                 mtx_unlock(&softc->ctl_lock);
11787                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11788                 return (1);
11789         }
11790
11791         mtx_lock(&lun->lun_lock);
11792         mtx_unlock(&softc->ctl_lock);
11793         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11794                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11795                     io->io_hdr.nexus.initid,
11796                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11797         } else { /* CTL_TASK_CLEAR_TASK_SET */
11798                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11799                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11800         }
11801         mtx_unlock(&lun->lun_lock);
11802         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11803         return (0);
11804 }
11805
11806 static void
11807 ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
11808     ctl_ua_type ua_type)
11809 {
11810         struct ctl_lun *lun;
11811         struct scsi_sense_data *ps;
11812         uint32_t p, i;
11813
11814         p = initidx / CTL_MAX_INIT_PER_PORT;
11815         i = initidx % CTL_MAX_INIT_PER_PORT;
11816         mtx_lock(&softc->ctl_lock);
11817         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11818                 mtx_lock(&lun->lun_lock);
11819                 /* Abort tasks. */
11820                 ctl_abort_tasks_lun(lun, p, i, 1);
11821                 /* Clear CA. */
11822                 ps = lun->pending_sense[p];
11823                 if (ps != NULL)
11824                         ps[i].error_code = 0;
11825                 /* Clear reservation. */
11826                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11827                         lun->flags &= ~CTL_LUN_RESERVED;
11828                 /* Clear prevent media removal. */
11829                 if (lun->prevent && ctl_is_set(lun->prevent, initidx)) {
11830                         ctl_clear_mask(lun->prevent, initidx);
11831                         lun->prevent_count--;
11832                 }
11833                 /* Clear TPC status */
11834                 ctl_tpc_lun_clear(lun, initidx);
11835                 /* Establish UA. */
11836                 ctl_est_ua(lun, initidx, ua_type);
11837                 mtx_unlock(&lun->lun_lock);
11838         }
11839         mtx_unlock(&softc->ctl_lock);
11840 }
11841
11842 static int
11843 ctl_i_t_nexus_reset(union ctl_io *io)
11844 {
11845         struct ctl_softc *softc = CTL_SOFTC(io);
11846         uint32_t initidx;
11847
11848         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11849                 union ctl_ha_msg msg_info;
11850
11851                 msg_info.hdr.nexus = io->io_hdr.nexus;
11852                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
11853                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11854                 msg_info.hdr.original_sc = NULL;
11855                 msg_info.hdr.serializing_sc = NULL;
11856                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11857                     sizeof(msg_info.task), M_WAITOK);
11858         }
11859
11860         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11861         ctl_i_t_nexus_loss(softc, initidx, CTL_UA_I_T_NEXUS_LOSS);
11862         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11863         return (0);
11864 }
11865
11866 static int
11867 ctl_abort_task(union ctl_io *io)
11868 {
11869         struct ctl_softc *softc = CTL_SOFTC(io);
11870         union ctl_io *xio;
11871         struct ctl_lun *lun;
11872 #if 0
11873         struct sbuf sb;
11874         char printbuf[128];
11875 #endif
11876         int found;
11877         uint32_t targ_lun;
11878
11879         found = 0;
11880
11881         /*
11882          * Look up the LUN.
11883          */
11884         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11885         mtx_lock(&softc->ctl_lock);
11886         if (targ_lun >= CTL_MAX_LUNS ||
11887             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11888                 mtx_unlock(&softc->ctl_lock);
11889                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11890                 return (1);
11891         }
11892
11893 #if 0
11894         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11895                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11896 #endif
11897
11898         mtx_lock(&lun->lun_lock);
11899         mtx_unlock(&softc->ctl_lock);
11900         /*
11901          * Run through the OOA queue and attempt to find the given I/O.
11902          * The target port, initiator ID, tag type and tag number have to
11903          * match the values that we got from the initiator.  If we have an
11904          * untagged command to abort, simply abort the first untagged command
11905          * we come to.  We only allow one untagged command at a time of course.
11906          */
11907         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11908              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11909 #if 0
11910                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11911
11912                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11913                             lun->lun, xio->scsiio.tag_num,
11914                             xio->scsiio.tag_type,
11915                             (xio->io_hdr.blocked_links.tqe_prev
11916                             == NULL) ? "" : " BLOCKED",
11917                             (xio->io_hdr.flags &
11918                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11919                             (xio->io_hdr.flags &
11920                             CTL_FLAG_ABORT) ? " ABORT" : "",
11921                             (xio->io_hdr.flags &
11922                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11923                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11924                 sbuf_finish(&sb);
11925                 printf("%s\n", sbuf_data(&sb));
11926 #endif
11927
11928                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11929                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11930                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11931                         continue;
11932
11933                 /*
11934                  * If the abort says that the task is untagged, the
11935                  * task in the queue must be untagged.  Otherwise,
11936                  * we just check to see whether the tag numbers
11937                  * match.  This is because the QLogic firmware
11938                  * doesn't pass back the tag type in an abort
11939                  * request.
11940                  */
11941 #if 0
11942                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11943                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11944                  || (xio->scsiio.tag_num == io->taskio.tag_num))
11945 #endif
11946                 /*
11947                  * XXX KDM we've got problems with FC, because it
11948                  * doesn't send down a tag type with aborts.  So we
11949                  * can only really go by the tag number...
11950                  * This may cause problems with parallel SCSI.
11951                  * Need to figure that out!!
11952                  */
11953                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
11954                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11955                         found = 1;
11956                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
11957                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11958                                 union ctl_ha_msg msg_info;
11959
11960                                 msg_info.hdr.nexus = io->io_hdr.nexus;
11961                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11962                                 msg_info.task.tag_num = io->taskio.tag_num;
11963                                 msg_info.task.tag_type = io->taskio.tag_type;
11964                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11965                                 msg_info.hdr.original_sc = NULL;
11966                                 msg_info.hdr.serializing_sc = NULL;
11967 #if 0
11968                                 printf("Sent Abort to other side\n");
11969 #endif
11970                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11971                                     sizeof(msg_info.task), M_NOWAIT);
11972                         }
11973 #if 0
11974                         printf("ctl_abort_task: found I/O to abort\n");
11975 #endif
11976                 }
11977         }
11978         mtx_unlock(&lun->lun_lock);
11979
11980         if (found == 0) {
11981                 /*
11982                  * This isn't really an error.  It's entirely possible for
11983                  * the abort and command completion to cross on the wire.
11984                  * This is more of an informative/diagnostic error.
11985                  */
11986 #if 0
11987                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11988                        "%u:%u:%u tag %d type %d\n",
11989                        io->io_hdr.nexus.initid,
11990                        io->io_hdr.nexus.targ_port,
11991                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11992                        io->taskio.tag_type);
11993 #endif
11994         }
11995         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11996         return (0);
11997 }
11998
11999 static int
12000 ctl_query_task(union ctl_io *io, int task_set)
12001 {
12002         struct ctl_softc *softc = CTL_SOFTC(io);
12003         union ctl_io *xio;
12004         struct ctl_lun *lun;
12005         int found = 0;
12006         uint32_t targ_lun;
12007
12008         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12009         mtx_lock(&softc->ctl_lock);
12010         if (targ_lun >= CTL_MAX_LUNS ||
12011             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12012                 mtx_unlock(&softc->ctl_lock);
12013                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12014                 return (1);
12015         }
12016         mtx_lock(&lun->lun_lock);
12017         mtx_unlock(&softc->ctl_lock);
12018         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12019              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12020
12021                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12022                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12023                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12024                         continue;
12025
12026                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12027                         found = 1;
12028                         break;
12029                 }
12030         }
12031         mtx_unlock(&lun->lun_lock);
12032         if (found)
12033                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12034         else
12035                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12036         return (0);
12037 }
12038
12039 static int
12040 ctl_query_async_event(union ctl_io *io)
12041 {
12042         struct ctl_softc *softc = CTL_SOFTC(io);
12043         struct ctl_lun *lun;
12044         ctl_ua_type ua;
12045         uint32_t targ_lun, initidx;
12046
12047         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12048         mtx_lock(&softc->ctl_lock);
12049         if (targ_lun >= CTL_MAX_LUNS ||
12050             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12051                 mtx_unlock(&softc->ctl_lock);
12052                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12053                 return (1);
12054         }
12055         mtx_lock(&lun->lun_lock);
12056         mtx_unlock(&softc->ctl_lock);
12057         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12058         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12059         mtx_unlock(&lun->lun_lock);
12060         if (ua != CTL_UA_NONE)
12061                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12062         else
12063                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12064         return (0);
12065 }
12066
12067 static void
12068 ctl_run_task(union ctl_io *io)
12069 {
12070         int retval = 1;
12071
12072         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12073         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12074             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12075         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12076         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12077         switch (io->taskio.task_action) {
12078         case CTL_TASK_ABORT_TASK:
12079                 retval = ctl_abort_task(io);
12080                 break;
12081         case CTL_TASK_ABORT_TASK_SET:
12082         case CTL_TASK_CLEAR_TASK_SET:
12083                 retval = ctl_abort_task_set(io);
12084                 break;
12085         case CTL_TASK_CLEAR_ACA:
12086                 break;
12087         case CTL_TASK_I_T_NEXUS_RESET:
12088                 retval = ctl_i_t_nexus_reset(io);
12089                 break;
12090         case CTL_TASK_LUN_RESET:
12091                 retval = ctl_lun_reset(io);
12092                 break;
12093         case CTL_TASK_TARGET_RESET:
12094         case CTL_TASK_BUS_RESET:
12095                 retval = ctl_target_reset(io);
12096                 break;
12097         case CTL_TASK_PORT_LOGIN:
12098                 break;
12099         case CTL_TASK_PORT_LOGOUT:
12100                 break;
12101         case CTL_TASK_QUERY_TASK:
12102                 retval = ctl_query_task(io, 0);
12103                 break;
12104         case CTL_TASK_QUERY_TASK_SET:
12105                 retval = ctl_query_task(io, 1);
12106                 break;
12107         case CTL_TASK_QUERY_ASYNC_EVENT:
12108                 retval = ctl_query_async_event(io);
12109                 break;
12110         default:
12111                 printf("%s: got unknown task management event %d\n",
12112                        __func__, io->taskio.task_action);
12113                 break;
12114         }
12115         if (retval == 0)
12116                 io->io_hdr.status = CTL_SUCCESS;
12117         else
12118                 io->io_hdr.status = CTL_ERROR;
12119         ctl_done(io);
12120 }
12121
12122 /*
12123  * For HA operation.  Handle commands that come in from the other
12124  * controller.
12125  */
12126 static void
12127 ctl_handle_isc(union ctl_io *io)
12128 {
12129         struct ctl_softc *softc = CTL_SOFTC(io);
12130         struct ctl_lun *lun;
12131         const struct ctl_cmd_entry *entry;
12132         uint32_t targ_lun;
12133
12134         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12135         switch (io->io_hdr.msg_type) {
12136         case CTL_MSG_SERIALIZE:
12137                 ctl_serialize_other_sc_cmd(&io->scsiio);
12138                 break;
12139         case CTL_MSG_R2R:               /* Only used in SER_ONLY mode. */
12140                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12141                 if (targ_lun >= CTL_MAX_LUNS ||
12142                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12143                         ctl_done(io);
12144                         break;
12145                 }
12146                 mtx_lock(&lun->lun_lock);
12147                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
12148                         mtx_unlock(&lun->lun_lock);
12149                         ctl_done(io);
12150                         break;
12151                 }
12152                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12153                 mtx_unlock(&lun->lun_lock);
12154                 ctl_enqueue_rtr(io);
12155                 break;
12156         case CTL_MSG_FINISH_IO:
12157                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12158                         ctl_done(io);
12159                         break;
12160                 }
12161                 if (targ_lun >= CTL_MAX_LUNS ||
12162                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12163                         ctl_free_io(io);
12164                         break;
12165                 }
12166                 mtx_lock(&lun->lun_lock);
12167                 TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12168                 ctl_check_blocked(lun);
12169                 mtx_unlock(&lun->lun_lock);
12170                 ctl_free_io(io);
12171                 break;
12172         case CTL_MSG_PERS_ACTION:
12173                 ctl_hndl_per_res_out_on_other_sc(io);
12174                 ctl_free_io(io);
12175                 break;
12176         case CTL_MSG_BAD_JUJU:
12177                 ctl_done(io);
12178                 break;
12179         case CTL_MSG_DATAMOVE:          /* Only used in XFER mode */
12180                 ctl_datamove_remote(io);
12181                 break;
12182         case CTL_MSG_DATAMOVE_DONE:     /* Only used in XFER mode */
12183                 io->scsiio.be_move_done(io);
12184                 break;
12185         case CTL_MSG_FAILOVER:
12186                 ctl_failover_lun(io);
12187                 ctl_free_io(io);
12188                 break;
12189         default:
12190                 printf("%s: Invalid message type %d\n",
12191                        __func__, io->io_hdr.msg_type);
12192                 ctl_free_io(io);
12193                 break;
12194         }
12195
12196 }
12197
12198
12199 /*
12200  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12201  * there is no match.
12202  */
12203 static ctl_lun_error_pattern
12204 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12205 {
12206         const struct ctl_cmd_entry *entry;
12207         ctl_lun_error_pattern filtered_pattern, pattern;
12208
12209         pattern = desc->error_pattern;
12210
12211         /*
12212          * XXX KDM we need more data passed into this function to match a
12213          * custom pattern, and we actually need to implement custom pattern
12214          * matching.
12215          */
12216         if (pattern & CTL_LUN_PAT_CMD)
12217                 return (CTL_LUN_PAT_CMD);
12218
12219         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12220                 return (CTL_LUN_PAT_ANY);
12221
12222         entry = ctl_get_cmd_entry(ctsio, NULL);
12223
12224         filtered_pattern = entry->pattern & pattern;
12225
12226         /*
12227          * If the user requested specific flags in the pattern (e.g.
12228          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12229          * flags.
12230          *
12231          * If the user did not specify any flags, it doesn't matter whether
12232          * or not the command supports the flags.
12233          */
12234         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12235              (pattern & ~CTL_LUN_PAT_MASK))
12236                 return (CTL_LUN_PAT_NONE);
12237
12238         /*
12239          * If the user asked for a range check, see if the requested LBA
12240          * range overlaps with this command's LBA range.
12241          */
12242         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12243                 uint64_t lba1;
12244                 uint64_t len1;
12245                 ctl_action action;
12246                 int retval;
12247
12248                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12249                 if (retval != 0)
12250                         return (CTL_LUN_PAT_NONE);
12251
12252                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12253                                               desc->lba_range.len, FALSE);
12254                 /*
12255                  * A "pass" means that the LBA ranges don't overlap, so
12256                  * this doesn't match the user's range criteria.
12257                  */
12258                 if (action == CTL_ACTION_PASS)
12259                         return (CTL_LUN_PAT_NONE);
12260         }
12261
12262         return (filtered_pattern);
12263 }
12264
12265 static void
12266 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12267 {
12268         struct ctl_error_desc *desc, *desc2;
12269
12270         mtx_assert(&lun->lun_lock, MA_OWNED);
12271
12272         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12273                 ctl_lun_error_pattern pattern;
12274                 /*
12275                  * Check to see whether this particular command matches
12276                  * the pattern in the descriptor.
12277                  */
12278                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12279                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12280                         continue;
12281
12282                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12283                 case CTL_LUN_INJ_ABORTED:
12284                         ctl_set_aborted(&io->scsiio);
12285                         break;
12286                 case CTL_LUN_INJ_MEDIUM_ERR:
12287                         ctl_set_medium_error(&io->scsiio,
12288                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12289                              CTL_FLAG_DATA_OUT);
12290                         break;
12291                 case CTL_LUN_INJ_UA:
12292                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12293                          * OCCURRED */
12294                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12295                         break;
12296                 case CTL_LUN_INJ_CUSTOM:
12297                         /*
12298                          * We're assuming the user knows what he is doing.
12299                          * Just copy the sense information without doing
12300                          * checks.
12301                          */
12302                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12303                               MIN(sizeof(desc->custom_sense),
12304                                   sizeof(io->scsiio.sense_data)));
12305                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12306                         io->scsiio.sense_len = SSD_FULL_SIZE;
12307                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12308                         break;
12309                 case CTL_LUN_INJ_NONE:
12310                 default:
12311                         /*
12312                          * If this is an error injection type we don't know
12313                          * about, clear the continuous flag (if it is set)
12314                          * so it will get deleted below.
12315                          */
12316                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12317                         break;
12318                 }
12319                 /*
12320                  * By default, each error injection action is a one-shot
12321                  */
12322                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12323                         continue;
12324
12325                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12326
12327                 free(desc, M_CTL);
12328         }
12329 }
12330
12331 #ifdef CTL_IO_DELAY
12332 static void
12333 ctl_datamove_timer_wakeup(void *arg)
12334 {
12335         union ctl_io *io;
12336
12337         io = (union ctl_io *)arg;
12338
12339         ctl_datamove(io);
12340 }
12341 #endif /* CTL_IO_DELAY */
12342
12343 void
12344 ctl_datamove(union ctl_io *io)
12345 {
12346         void (*fe_datamove)(union ctl_io *io);
12347
12348         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12349
12350         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12351
12352         /* No data transferred yet.  Frontend must update this when done. */
12353         io->scsiio.kern_data_resid = io->scsiio.kern_data_len;
12354
12355 #ifdef CTL_TIME_IO
12356         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12357                 char str[256];
12358                 char path_str[64];
12359                 struct sbuf sb;
12360
12361                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12362                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12363
12364                 sbuf_cat(&sb, path_str);
12365                 switch (io->io_hdr.io_type) {
12366                 case CTL_IO_SCSI:
12367                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12368                         sbuf_printf(&sb, "\n");
12369                         sbuf_cat(&sb, path_str);
12370                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12371                                     io->scsiio.tag_num, io->scsiio.tag_type);
12372                         break;
12373                 case CTL_IO_TASK:
12374                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12375                                     "Tag Type: %d\n", io->taskio.task_action,
12376                                     io->taskio.tag_num, io->taskio.tag_type);
12377                         break;
12378                 default:
12379                         panic("%s: Invalid CTL I/O type %d\n",
12380                             __func__, io->io_hdr.io_type);
12381                 }
12382                 sbuf_cat(&sb, path_str);
12383                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12384                             (intmax_t)time_uptime - io->io_hdr.start_time);
12385                 sbuf_finish(&sb);
12386                 printf("%s", sbuf_data(&sb));
12387         }
12388 #endif /* CTL_TIME_IO */
12389
12390 #ifdef CTL_IO_DELAY
12391         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12392                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12393         } else {
12394                 struct ctl_lun *lun;
12395
12396                 lun = CTL_LUN(io);
12397                 if ((lun != NULL)
12398                  && (lun->delay_info.datamove_delay > 0)) {
12399
12400                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12401                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12402                         callout_reset(&io->io_hdr.delay_callout,
12403                                       lun->delay_info.datamove_delay * hz,
12404                                       ctl_datamove_timer_wakeup, io);
12405                         if (lun->delay_info.datamove_type ==
12406                             CTL_DELAY_TYPE_ONESHOT)
12407                                 lun->delay_info.datamove_delay = 0;
12408                         return;
12409                 }
12410         }
12411 #endif
12412
12413         /*
12414          * This command has been aborted.  Set the port status, so we fail
12415          * the data move.
12416          */
12417         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12418                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12419                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12420                        io->io_hdr.nexus.targ_port,
12421                        io->io_hdr.nexus.targ_lun);
12422                 io->io_hdr.port_status = 31337;
12423                 /*
12424                  * Note that the backend, in this case, will get the
12425                  * callback in its context.  In other cases it may get
12426                  * called in the frontend's interrupt thread context.
12427                  */
12428                 io->scsiio.be_move_done(io);
12429                 return;
12430         }
12431
12432         /* Don't confuse frontend with zero length data move. */
12433         if (io->scsiio.kern_data_len == 0) {
12434                 io->scsiio.be_move_done(io);
12435                 return;
12436         }
12437
12438         fe_datamove = CTL_PORT(io)->fe_datamove;
12439         fe_datamove(io);
12440 }
12441
12442 static void
12443 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12444 {
12445         union ctl_ha_msg msg;
12446 #ifdef CTL_TIME_IO
12447         struct bintime cur_bt;
12448 #endif
12449
12450         memset(&msg, 0, sizeof(msg));
12451         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12452         msg.hdr.original_sc = io;
12453         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12454         msg.hdr.nexus = io->io_hdr.nexus;
12455         msg.hdr.status = io->io_hdr.status;
12456         msg.scsi.kern_data_resid = io->scsiio.kern_data_resid;
12457         msg.scsi.tag_num = io->scsiio.tag_num;
12458         msg.scsi.tag_type = io->scsiio.tag_type;
12459         msg.scsi.scsi_status = io->scsiio.scsi_status;
12460         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12461                io->scsiio.sense_len);
12462         msg.scsi.sense_len = io->scsiio.sense_len;
12463         msg.scsi.port_status = io->io_hdr.port_status;
12464         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12465         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12466                 ctl_failover_io(io, /*have_lock*/ have_lock);
12467                 return;
12468         }
12469         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12470             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12471             msg.scsi.sense_len, M_WAITOK);
12472
12473 #ifdef CTL_TIME_IO
12474         getbinuptime(&cur_bt);
12475         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12476         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12477 #endif
12478         io->io_hdr.num_dmas++;
12479 }
12480
12481 /*
12482  * The DMA to the remote side is done, now we need to tell the other side
12483  * we're done so it can continue with its data movement.
12484  */
12485 static void
12486 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12487 {
12488         union ctl_io *io;
12489         uint32_t i;
12490
12491         io = rq->context;
12492
12493         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12494                 printf("%s: ISC DMA write failed with error %d", __func__,
12495                        rq->ret);
12496                 ctl_set_internal_failure(&io->scsiio,
12497                                          /*sks_valid*/ 1,
12498                                          /*retry_count*/ rq->ret);
12499         }
12500
12501         ctl_dt_req_free(rq);
12502
12503         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12504                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12505         free(io->io_hdr.remote_sglist, M_CTL);
12506         io->io_hdr.remote_sglist = NULL;
12507         io->io_hdr.local_sglist = NULL;
12508
12509         /*
12510          * The data is in local and remote memory, so now we need to send
12511          * status (good or back) back to the other side.
12512          */
12513         ctl_send_datamove_done(io, /*have_lock*/ 0);
12514 }
12515
12516 /*
12517  * We've moved the data from the host/controller into local memory.  Now we
12518  * need to push it over to the remote controller's memory.
12519  */
12520 static int
12521 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12522 {
12523         int retval;
12524
12525         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12526                                           ctl_datamove_remote_write_cb);
12527         return (retval);
12528 }
12529
12530 static void
12531 ctl_datamove_remote_write(union ctl_io *io)
12532 {
12533         int retval;
12534         void (*fe_datamove)(union ctl_io *io);
12535
12536         /*
12537          * - Get the data from the host/HBA into local memory.
12538          * - DMA memory from the local controller to the remote controller.
12539          * - Send status back to the remote controller.
12540          */
12541
12542         retval = ctl_datamove_remote_sgl_setup(io);
12543         if (retval != 0)
12544                 return;
12545
12546         /* Switch the pointer over so the FETD knows what to do */
12547         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12548
12549         /*
12550          * Use a custom move done callback, since we need to send completion
12551          * back to the other controller, not to the backend on this side.
12552          */
12553         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12554
12555         fe_datamove = CTL_PORT(io)->fe_datamove;
12556         fe_datamove(io);
12557 }
12558
12559 static int
12560 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12561 {
12562 #if 0
12563         char str[256];
12564         char path_str[64];
12565         struct sbuf sb;
12566 #endif
12567         uint32_t i;
12568
12569         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12570                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12571         free(io->io_hdr.remote_sglist, M_CTL);
12572         io->io_hdr.remote_sglist = NULL;
12573         io->io_hdr.local_sglist = NULL;
12574
12575 #if 0
12576         scsi_path_string(io, path_str, sizeof(path_str));
12577         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12578         sbuf_cat(&sb, path_str);
12579         scsi_command_string(&io->scsiio, NULL, &sb);
12580         sbuf_printf(&sb, "\n");
12581         sbuf_cat(&sb, path_str);
12582         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12583                     io->scsiio.tag_num, io->scsiio.tag_type);
12584         sbuf_cat(&sb, path_str);
12585         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12586                     io->io_hdr.flags, io->io_hdr.status);
12587         sbuf_finish(&sb);
12588         printk("%s", sbuf_data(&sb));
12589 #endif
12590
12591
12592         /*
12593          * The read is done, now we need to send status (good or bad) back
12594          * to the other side.
12595          */
12596         ctl_send_datamove_done(io, /*have_lock*/ 0);
12597
12598         return (0);
12599 }
12600
12601 static void
12602 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12603 {
12604         union ctl_io *io;
12605         void (*fe_datamove)(union ctl_io *io);
12606
12607         io = rq->context;
12608
12609         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12610                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12611                        rq->ret);
12612                 ctl_set_internal_failure(&io->scsiio,
12613                                          /*sks_valid*/ 1,
12614                                          /*retry_count*/ rq->ret);
12615         }
12616
12617         ctl_dt_req_free(rq);
12618
12619         /* Switch the pointer over so the FETD knows what to do */
12620         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12621
12622         /*
12623          * Use a custom move done callback, since we need to send completion
12624          * back to the other controller, not to the backend on this side.
12625          */
12626         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12627
12628         /* XXX KDM add checks like the ones in ctl_datamove? */
12629
12630         fe_datamove = CTL_PORT(io)->fe_datamove;
12631         fe_datamove(io);
12632 }
12633
12634 static int
12635 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12636 {
12637         struct ctl_sg_entry *local_sglist;
12638         uint32_t len_to_go;
12639         int retval;
12640         int i;
12641
12642         retval = 0;
12643         local_sglist = io->io_hdr.local_sglist;
12644         len_to_go = io->scsiio.kern_data_len;
12645
12646         /*
12647          * The difficult thing here is that the size of the various
12648          * S/G segments may be different than the size from the
12649          * remote controller.  That'll make it harder when DMAing
12650          * the data back to the other side.
12651          */
12652         for (i = 0; len_to_go > 0; i++) {
12653                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12654                 local_sglist[i].addr =
12655                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12656
12657                 len_to_go -= local_sglist[i].len;
12658         }
12659         /*
12660          * Reset the number of S/G entries accordingly.  The original
12661          * number of S/G entries is available in rem_sg_entries.
12662          */
12663         io->scsiio.kern_sg_entries = i;
12664
12665 #if 0
12666         printf("%s: kern_sg_entries = %d\n", __func__,
12667                io->scsiio.kern_sg_entries);
12668         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12669                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12670                        local_sglist[i].addr, local_sglist[i].len);
12671 #endif
12672
12673         return (retval);
12674 }
12675
12676 static int
12677 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12678                          ctl_ha_dt_cb callback)
12679 {
12680         struct ctl_ha_dt_req *rq;
12681         struct ctl_sg_entry *remote_sglist, *local_sglist;
12682         uint32_t local_used, remote_used, total_used;
12683         int i, j, isc_ret;
12684
12685         rq = ctl_dt_req_alloc();
12686
12687         /*
12688          * If we failed to allocate the request, and if the DMA didn't fail
12689          * anyway, set busy status.  This is just a resource allocation
12690          * failure.
12691          */
12692         if ((rq == NULL)
12693          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12694              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12695                 ctl_set_busy(&io->scsiio);
12696
12697         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12698             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12699
12700                 if (rq != NULL)
12701                         ctl_dt_req_free(rq);
12702
12703                 /*
12704                  * The data move failed.  We need to return status back
12705                  * to the other controller.  No point in trying to DMA
12706                  * data to the remote controller.
12707                  */
12708
12709                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12710
12711                 return (1);
12712         }
12713
12714         local_sglist = io->io_hdr.local_sglist;
12715         remote_sglist = io->io_hdr.remote_sglist;
12716         local_used = 0;
12717         remote_used = 0;
12718         total_used = 0;
12719
12720         /*
12721          * Pull/push the data over the wire from/to the other controller.
12722          * This takes into account the possibility that the local and
12723          * remote sglists may not be identical in terms of the size of
12724          * the elements and the number of elements.
12725          *
12726          * One fundamental assumption here is that the length allocated for
12727          * both the local and remote sglists is identical.  Otherwise, we've
12728          * essentially got a coding error of some sort.
12729          */
12730         isc_ret = CTL_HA_STATUS_SUCCESS;
12731         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12732                 uint32_t cur_len;
12733                 uint8_t *tmp_ptr;
12734
12735                 rq->command = command;
12736                 rq->context = io;
12737
12738                 /*
12739                  * Both pointers should be aligned.  But it is possible
12740                  * that the allocation length is not.  They should both
12741                  * also have enough slack left over at the end, though,
12742                  * to round up to the next 8 byte boundary.
12743                  */
12744                 cur_len = MIN(local_sglist[i].len - local_used,
12745                               remote_sglist[j].len - remote_used);
12746                 rq->size = cur_len;
12747
12748                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12749                 tmp_ptr += local_used;
12750
12751 #if 0
12752                 /* Use physical addresses when talking to ISC hardware */
12753                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12754                         /* XXX KDM use busdma */
12755                         rq->local = vtophys(tmp_ptr);
12756                 } else
12757                         rq->local = tmp_ptr;
12758 #else
12759                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12760                     ("HA does not support BUS_ADDR"));
12761                 rq->local = tmp_ptr;
12762 #endif
12763
12764                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12765                 tmp_ptr += remote_used;
12766                 rq->remote = tmp_ptr;
12767
12768                 rq->callback = NULL;
12769
12770                 local_used += cur_len;
12771                 if (local_used >= local_sglist[i].len) {
12772                         i++;
12773                         local_used = 0;
12774                 }
12775
12776                 remote_used += cur_len;
12777                 if (remote_used >= remote_sglist[j].len) {
12778                         j++;
12779                         remote_used = 0;
12780                 }
12781                 total_used += cur_len;
12782
12783                 if (total_used >= io->scsiio.kern_data_len)
12784                         rq->callback = callback;
12785
12786 #if 0
12787                 printf("%s: %s: local %p remote %p size %d\n", __func__,
12788                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12789                        rq->local, rq->remote, rq->size);
12790 #endif
12791
12792                 isc_ret = ctl_dt_single(rq);
12793                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
12794                         break;
12795         }
12796         if (isc_ret != CTL_HA_STATUS_WAIT) {
12797                 rq->ret = isc_ret;
12798                 callback(rq);
12799         }
12800
12801         return (0);
12802 }
12803
12804 static void
12805 ctl_datamove_remote_read(union ctl_io *io)
12806 {
12807         int retval;
12808         uint32_t i;
12809
12810         /*
12811          * This will send an error to the other controller in the case of a
12812          * failure.
12813          */
12814         retval = ctl_datamove_remote_sgl_setup(io);
12815         if (retval != 0)
12816                 return;
12817
12818         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12819                                           ctl_datamove_remote_read_cb);
12820         if (retval != 0) {
12821                 /*
12822                  * Make sure we free memory if there was an error..  The
12823                  * ctl_datamove_remote_xfer() function will send the
12824                  * datamove done message, or call the callback with an
12825                  * error if there is a problem.
12826                  */
12827                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12828                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12829                 free(io->io_hdr.remote_sglist, M_CTL);
12830                 io->io_hdr.remote_sglist = NULL;
12831                 io->io_hdr.local_sglist = NULL;
12832         }
12833 }
12834
12835 /*
12836  * Process a datamove request from the other controller.  This is used for
12837  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12838  * first.  Once that is complete, the data gets DMAed into the remote
12839  * controller's memory.  For reads, we DMA from the remote controller's
12840  * memory into our memory first, and then move it out to the FETD.
12841  */
12842 static void
12843 ctl_datamove_remote(union ctl_io *io)
12844 {
12845
12846         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12847
12848         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12849                 ctl_failover_io(io, /*have_lock*/ 0);
12850                 return;
12851         }
12852
12853         /*
12854          * Note that we look for an aborted I/O here, but don't do some of
12855          * the other checks that ctl_datamove() normally does.
12856          * We don't need to run the datamove delay code, since that should
12857          * have been done if need be on the other controller.
12858          */
12859         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12860                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
12861                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12862                        io->io_hdr.nexus.targ_port,
12863                        io->io_hdr.nexus.targ_lun);
12864                 io->io_hdr.port_status = 31338;
12865                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12866                 return;
12867         }
12868
12869         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
12870                 ctl_datamove_remote_write(io);
12871         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
12872                 ctl_datamove_remote_read(io);
12873         else {
12874                 io->io_hdr.port_status = 31339;
12875                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12876         }
12877 }
12878
12879 static void
12880 ctl_process_done(union ctl_io *io)
12881 {
12882         struct ctl_softc *softc = CTL_SOFTC(io);
12883         struct ctl_port *port = CTL_PORT(io);
12884         struct ctl_lun *lun = CTL_LUN(io);
12885         void (*fe_done)(union ctl_io *io);
12886         union ctl_ha_msg msg;
12887
12888         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12889         fe_done = port->fe_done;
12890
12891 #ifdef CTL_TIME_IO
12892         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12893                 char str[256];
12894                 char path_str[64];
12895                 struct sbuf sb;
12896
12897                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12898                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12899
12900                 sbuf_cat(&sb, path_str);
12901                 switch (io->io_hdr.io_type) {
12902                 case CTL_IO_SCSI:
12903                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12904                         sbuf_printf(&sb, "\n");
12905                         sbuf_cat(&sb, path_str);
12906                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12907                                     io->scsiio.tag_num, io->scsiio.tag_type);
12908                         break;
12909                 case CTL_IO_TASK:
12910                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12911                                     "Tag Type: %d\n", io->taskio.task_action,
12912                                     io->taskio.tag_num, io->taskio.tag_type);
12913                         break;
12914                 default:
12915                         panic("%s: Invalid CTL I/O type %d\n",
12916                             __func__, io->io_hdr.io_type);
12917                 }
12918                 sbuf_cat(&sb, path_str);
12919                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12920                             (intmax_t)time_uptime - io->io_hdr.start_time);
12921                 sbuf_finish(&sb);
12922                 printf("%s", sbuf_data(&sb));
12923         }
12924 #endif /* CTL_TIME_IO */
12925
12926         switch (io->io_hdr.io_type) {
12927         case CTL_IO_SCSI:
12928                 break;
12929         case CTL_IO_TASK:
12930                 if (ctl_debug & CTL_DEBUG_INFO)
12931                         ctl_io_error_print(io, NULL);
12932                 fe_done(io);
12933                 return;
12934         default:
12935                 panic("%s: Invalid CTL I/O type %d\n",
12936                     __func__, io->io_hdr.io_type);
12937         }
12938
12939         if (lun == NULL) {
12940                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12941                                  io->io_hdr.nexus.targ_mapped_lun));
12942                 goto bailout;
12943         }
12944
12945         mtx_lock(&lun->lun_lock);
12946
12947         /*
12948          * Check to see if we have any informational exception and status
12949          * of this command can be modified to report it in form of either
12950          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
12951          */
12952         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
12953             io->io_hdr.status == CTL_SUCCESS &&
12954             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
12955                 uint8_t mrie = lun->MODE_IE.mrie;
12956                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
12957                     (lun->MODE_VER.byte3 & SMS_VER_PER));
12958                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
12959                      mrie == SIEP_MRIE_REC_UNCOND ||
12960                      mrie == SIEP_MRIE_NO_SENSE) &&
12961                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
12962                      CTL_CMD_FLAG_NO_SENSE) == 0) {
12963                         ctl_set_sense(&io->scsiio,
12964                               /*current_error*/ 1,
12965                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
12966                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
12967                               /*asc*/ lun->ie_asc,
12968                               /*ascq*/ lun->ie_ascq,
12969                               SSD_ELEM_NONE);
12970                         lun->ie_reported = 1;
12971                 }
12972         } else if (lun->ie_reported < 0)
12973                 lun->ie_reported = 0;
12974
12975         /*
12976          * Check to see if we have any errors to inject here.  We only
12977          * inject errors for commands that don't already have errors set.
12978          */
12979         if (!STAILQ_EMPTY(&lun->error_list) &&
12980             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
12981             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
12982                 ctl_inject_error(lun, io);
12983
12984         /*
12985          * XXX KDM how do we treat commands that aren't completed
12986          * successfully?
12987          *
12988          * XXX KDM should we also track I/O latency?
12989          */
12990         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
12991             io->io_hdr.io_type == CTL_IO_SCSI) {
12992                 int type;
12993 #ifdef CTL_TIME_IO
12994                 struct bintime bt;
12995
12996                 getbinuptime(&bt);
12997                 bintime_sub(&bt, &io->io_hdr.start_bt);
12998 #endif
12999                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13000                     CTL_FLAG_DATA_IN)
13001                         type = CTL_STATS_READ;
13002                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13003                     CTL_FLAG_DATA_OUT)
13004                         type = CTL_STATS_WRITE;
13005                 else
13006                         type = CTL_STATS_NO_IO;
13007
13008 #ifdef CTL_LEGACY_STATS
13009                 uint32_t targ_port = port->targ_port;
13010                 lun->legacy_stats.ports[targ_port].bytes[type] +=
13011                     io->scsiio.kern_total_len;
13012                 lun->legacy_stats.ports[targ_port].operations[type] ++;
13013                 lun->legacy_stats.ports[targ_port].num_dmas[type] +=
13014                     io->io_hdr.num_dmas;
13015 #ifdef CTL_TIME_IO
13016                 bintime_add(&lun->legacy_stats.ports[targ_port].dma_time[type],
13017                    &io->io_hdr.dma_bt);
13018                 bintime_add(&lun->legacy_stats.ports[targ_port].time[type],
13019                     &bt);
13020 #endif
13021 #endif /* CTL_LEGACY_STATS */
13022
13023                 lun->stats.bytes[type] += io->scsiio.kern_total_len;
13024                 lun->stats.operations[type] ++;
13025                 lun->stats.dmas[type] += io->io_hdr.num_dmas;
13026 #ifdef CTL_TIME_IO
13027                 bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt);
13028                 bintime_add(&lun->stats.time[type], &bt);
13029 #endif
13030
13031                 mtx_lock(&port->port_lock);
13032                 port->stats.bytes[type] += io->scsiio.kern_total_len;
13033                 port->stats.operations[type] ++;
13034                 port->stats.dmas[type] += io->io_hdr.num_dmas;
13035 #ifdef CTL_TIME_IO
13036                 bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt);
13037                 bintime_add(&port->stats.time[type], &bt);
13038 #endif
13039                 mtx_unlock(&port->port_lock);
13040         }
13041
13042         /*
13043          * Remove this from the OOA queue.
13044          */
13045         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13046 #ifdef CTL_TIME_IO
13047         if (TAILQ_EMPTY(&lun->ooa_queue))
13048                 lun->last_busy = getsbinuptime();
13049 #endif
13050
13051         /*
13052          * Run through the blocked queue on this LUN and see if anything
13053          * has become unblocked, now that this transaction is done.
13054          */
13055         ctl_check_blocked(lun);
13056
13057         /*
13058          * If the LUN has been invalidated, free it if there is nothing
13059          * left on its OOA queue.
13060          */
13061         if ((lun->flags & CTL_LUN_INVALID)
13062          && TAILQ_EMPTY(&lun->ooa_queue)) {
13063                 mtx_unlock(&lun->lun_lock);
13064                 ctl_free_lun(lun);
13065         } else
13066                 mtx_unlock(&lun->lun_lock);
13067
13068 bailout:
13069
13070         /*
13071          * If this command has been aborted, make sure we set the status
13072          * properly.  The FETD is responsible for freeing the I/O and doing
13073          * whatever it needs to do to clean up its state.
13074          */
13075         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13076                 ctl_set_task_aborted(&io->scsiio);
13077
13078         /*
13079          * If enabled, print command error status.
13080          */
13081         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13082             (ctl_debug & CTL_DEBUG_INFO) != 0)
13083                 ctl_io_error_print(io, NULL);
13084
13085         /*
13086          * Tell the FETD or the other shelf controller we're done with this
13087          * command.  Note that only SCSI commands get to this point.  Task
13088          * management commands are completed above.
13089          */
13090         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13091             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13092                 memset(&msg, 0, sizeof(msg));
13093                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13094                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13095                 msg.hdr.nexus = io->io_hdr.nexus;
13096                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13097                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13098                     M_WAITOK);
13099         }
13100
13101         fe_done(io);
13102 }
13103
13104 /*
13105  * Front end should call this if it doesn't do autosense.  When the request
13106  * sense comes back in from the initiator, we'll dequeue this and send it.
13107  */
13108 int
13109 ctl_queue_sense(union ctl_io *io)
13110 {
13111         struct ctl_softc *softc = CTL_SOFTC(io);
13112         struct ctl_port *port = CTL_PORT(io);
13113         struct ctl_lun *lun;
13114         struct scsi_sense_data *ps;
13115         uint32_t initidx, p, targ_lun;
13116
13117         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13118
13119         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13120
13121         /*
13122          * LUN lookup will likely move to the ctl_work_thread() once we
13123          * have our new queueing infrastructure (that doesn't put things on
13124          * a per-LUN queue initially).  That is so that we can handle
13125          * things like an INQUIRY to a LUN that we don't have enabled.  We
13126          * can't deal with that right now.
13127          * If we don't have a LUN for this, just toss the sense information.
13128          */
13129         mtx_lock(&softc->ctl_lock);
13130         if (targ_lun >= CTL_MAX_LUNS ||
13131             (lun = softc->ctl_luns[targ_lun]) == NULL) {
13132                 mtx_unlock(&softc->ctl_lock);
13133                 goto bailout;
13134         }
13135         mtx_lock(&lun->lun_lock);
13136         mtx_unlock(&softc->ctl_lock);
13137
13138         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13139         p = initidx / CTL_MAX_INIT_PER_PORT;
13140         if (lun->pending_sense[p] == NULL) {
13141                 lun->pending_sense[p] = malloc(sizeof(*ps) * CTL_MAX_INIT_PER_PORT,
13142                     M_CTL, M_NOWAIT | M_ZERO);
13143         }
13144         if ((ps = lun->pending_sense[p]) != NULL) {
13145                 ps += initidx % CTL_MAX_INIT_PER_PORT;
13146                 memset(ps, 0, sizeof(*ps));
13147                 memcpy(ps, &io->scsiio.sense_data, io->scsiio.sense_len);
13148         }
13149         mtx_unlock(&lun->lun_lock);
13150
13151 bailout:
13152         ctl_free_io(io);
13153         return (CTL_RETVAL_COMPLETE);
13154 }
13155
13156 /*
13157  * Primary command inlet from frontend ports.  All SCSI and task I/O
13158  * requests must go through this function.
13159  */
13160 int
13161 ctl_queue(union ctl_io *io)
13162 {
13163         struct ctl_port *port = CTL_PORT(io);
13164
13165         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13166
13167 #ifdef CTL_TIME_IO
13168         io->io_hdr.start_time = time_uptime;
13169         getbinuptime(&io->io_hdr.start_bt);
13170 #endif /* CTL_TIME_IO */
13171
13172         /* Map FE-specific LUN ID into global one. */
13173         io->io_hdr.nexus.targ_mapped_lun =
13174             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13175
13176         switch (io->io_hdr.io_type) {
13177         case CTL_IO_SCSI:
13178         case CTL_IO_TASK:
13179                 if (ctl_debug & CTL_DEBUG_CDB)
13180                         ctl_io_print(io);
13181                 ctl_enqueue_incoming(io);
13182                 break;
13183         default:
13184                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13185                 return (EINVAL);
13186         }
13187
13188         return (CTL_RETVAL_COMPLETE);
13189 }
13190
13191 #ifdef CTL_IO_DELAY
13192 static void
13193 ctl_done_timer_wakeup(void *arg)
13194 {
13195         union ctl_io *io;
13196
13197         io = (union ctl_io *)arg;
13198         ctl_done(io);
13199 }
13200 #endif /* CTL_IO_DELAY */
13201
13202 void
13203 ctl_serseq_done(union ctl_io *io)
13204 {
13205         struct ctl_lun *lun = CTL_LUN(io);;
13206
13207         if (lun->be_lun == NULL ||
13208             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13209                 return;
13210         mtx_lock(&lun->lun_lock);
13211         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13212         ctl_check_blocked(lun);
13213         mtx_unlock(&lun->lun_lock);
13214 }
13215
13216 void
13217 ctl_done(union ctl_io *io)
13218 {
13219
13220         /*
13221          * Enable this to catch duplicate completion issues.
13222          */
13223 #if 0
13224         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13225                 printf("%s: type %d msg %d cdb %x iptl: "
13226                        "%u:%u:%u tag 0x%04x "
13227                        "flag %#x status %x\n",
13228                         __func__,
13229                         io->io_hdr.io_type,
13230                         io->io_hdr.msg_type,
13231                         io->scsiio.cdb[0],
13232                         io->io_hdr.nexus.initid,
13233                         io->io_hdr.nexus.targ_port,
13234                         io->io_hdr.nexus.targ_lun,
13235                         (io->io_hdr.io_type ==
13236                         CTL_IO_TASK) ?
13237                         io->taskio.tag_num :
13238                         io->scsiio.tag_num,
13239                         io->io_hdr.flags,
13240                         io->io_hdr.status);
13241         } else
13242                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13243 #endif
13244
13245         /*
13246          * This is an internal copy of an I/O, and should not go through
13247          * the normal done processing logic.
13248          */
13249         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13250                 return;
13251
13252 #ifdef CTL_IO_DELAY
13253         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13254                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13255         } else {
13256                 struct ctl_lun *lun = CTL_LUN(io);
13257
13258                 if ((lun != NULL)
13259                  && (lun->delay_info.done_delay > 0)) {
13260
13261                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13262                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13263                         callout_reset(&io->io_hdr.delay_callout,
13264                                       lun->delay_info.done_delay * hz,
13265                                       ctl_done_timer_wakeup, io);
13266                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13267                                 lun->delay_info.done_delay = 0;
13268                         return;
13269                 }
13270         }
13271 #endif /* CTL_IO_DELAY */
13272
13273         ctl_enqueue_done(io);
13274 }
13275
13276 static void
13277 ctl_work_thread(void *arg)
13278 {
13279         struct ctl_thread *thr = (struct ctl_thread *)arg;
13280         struct ctl_softc *softc = thr->ctl_softc;
13281         union ctl_io *io;
13282         int retval;
13283
13284         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13285
13286         while (!softc->shutdown) {
13287                 /*
13288                  * We handle the queues in this order:
13289                  * - ISC
13290                  * - done queue (to free up resources, unblock other commands)
13291                  * - RtR queue
13292                  * - incoming queue
13293                  *
13294                  * If those queues are empty, we break out of the loop and
13295                  * go to sleep.
13296                  */
13297                 mtx_lock(&thr->queue_lock);
13298                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13299                 if (io != NULL) {
13300                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13301                         mtx_unlock(&thr->queue_lock);
13302                         ctl_handle_isc(io);
13303                         continue;
13304                 }
13305                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13306                 if (io != NULL) {
13307                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13308                         /* clear any blocked commands, call fe_done */
13309                         mtx_unlock(&thr->queue_lock);
13310                         ctl_process_done(io);
13311                         continue;
13312                 }
13313                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13314                 if (io != NULL) {
13315                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13316                         mtx_unlock(&thr->queue_lock);
13317                         if (io->io_hdr.io_type == CTL_IO_TASK)
13318                                 ctl_run_task(io);
13319                         else
13320                                 ctl_scsiio_precheck(softc, &io->scsiio);
13321                         continue;
13322                 }
13323                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13324                 if (io != NULL) {
13325                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13326                         mtx_unlock(&thr->queue_lock);
13327                         retval = ctl_scsiio(&io->scsiio);
13328                         if (retval != CTL_RETVAL_COMPLETE)
13329                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13330                         continue;
13331                 }
13332
13333                 /* Sleep until we have something to do. */
13334                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13335         }
13336         thr->thread = NULL;
13337         kthread_exit();
13338 }
13339
13340 static void
13341 ctl_lun_thread(void *arg)
13342 {
13343         struct ctl_softc *softc = (struct ctl_softc *)arg;
13344         struct ctl_be_lun *be_lun;
13345
13346         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13347
13348         while (!softc->shutdown) {
13349                 mtx_lock(&softc->ctl_lock);
13350                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13351                 if (be_lun != NULL) {
13352                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13353                         mtx_unlock(&softc->ctl_lock);
13354                         ctl_create_lun(be_lun);
13355                         continue;
13356                 }
13357
13358                 /* Sleep until we have something to do. */
13359                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13360                     PDROP | PRIBIO, "-", 0);
13361         }
13362         softc->lun_thread = NULL;
13363         kthread_exit();
13364 }
13365
13366 static void
13367 ctl_thresh_thread(void *arg)
13368 {
13369         struct ctl_softc *softc = (struct ctl_softc *)arg;
13370         struct ctl_lun *lun;
13371         struct ctl_logical_block_provisioning_page *page;
13372         const char *attr;
13373         union ctl_ha_msg msg;
13374         uint64_t thres, val;
13375         int i, e, set;
13376
13377         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13378
13379         while (!softc->shutdown) {
13380                 mtx_lock(&softc->ctl_lock);
13381                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13382                         if ((lun->flags & CTL_LUN_DISABLED) ||
13383                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13384                             lun->backend->lun_attr == NULL)
13385                                 continue;
13386                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13387                             softc->ha_mode == CTL_HA_MODE_XFER)
13388                                 continue;
13389                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13390                                 continue;
13391                         e = 0;
13392                         page = &lun->MODE_LBP;
13393                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13394                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13395                                         continue;
13396                                 thres = scsi_4btoul(page->descr[i].count);
13397                                 thres <<= CTL_LBP_EXPONENT;
13398                                 switch (page->descr[i].resource) {
13399                                 case 0x01:
13400                                         attr = "blocksavail";
13401                                         break;
13402                                 case 0x02:
13403                                         attr = "blocksused";
13404                                         break;
13405                                 case 0xf1:
13406                                         attr = "poolblocksavail";
13407                                         break;
13408                                 case 0xf2:
13409                                         attr = "poolblocksused";
13410                                         break;
13411                                 default:
13412                                         continue;
13413                                 }
13414                                 mtx_unlock(&softc->ctl_lock); // XXX
13415                                 val = lun->backend->lun_attr(
13416                                     lun->be_lun->be_lun, attr);
13417                                 mtx_lock(&softc->ctl_lock);
13418                                 if (val == UINT64_MAX)
13419                                         continue;
13420                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13421                                     == SLBPPD_ARMING_INC)
13422                                         e = (val >= thres);
13423                                 else
13424                                         e = (val <= thres);
13425                                 if (e)
13426                                         break;
13427                         }
13428                         mtx_lock(&lun->lun_lock);
13429                         if (e) {
13430                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13431                                     (uint8_t *)page, lun->ua_tpt_info);
13432                                 if (lun->lasttpt == 0 ||
13433                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13434                                         lun->lasttpt = time_uptime;
13435                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13436                                         set = 1;
13437                                 } else
13438                                         set = 0;
13439                         } else {
13440                                 lun->lasttpt = 0;
13441                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13442                                 set = -1;
13443                         }
13444                         mtx_unlock(&lun->lun_lock);
13445                         if (set != 0 &&
13446                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13447                                 /* Send msg to other side. */
13448                                 bzero(&msg.ua, sizeof(msg.ua));
13449                                 msg.hdr.msg_type = CTL_MSG_UA;
13450                                 msg.hdr.nexus.initid = -1;
13451                                 msg.hdr.nexus.targ_port = -1;
13452                                 msg.hdr.nexus.targ_lun = lun->lun;
13453                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13454                                 msg.ua.ua_all = 1;
13455                                 msg.ua.ua_set = (set > 0);
13456                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13457                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13458                                 mtx_unlock(&softc->ctl_lock); // XXX
13459                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13460                                     sizeof(msg.ua), M_WAITOK);
13461                                 mtx_lock(&softc->ctl_lock);
13462                         }
13463                 }
13464                 mtx_sleep(&softc->thresh_thread, &softc->ctl_lock,
13465                     PDROP | PRIBIO, "-", CTL_LBP_PERIOD * hz);
13466         }
13467         softc->thresh_thread = NULL;
13468         kthread_exit();
13469 }
13470
13471 static void
13472 ctl_enqueue_incoming(union ctl_io *io)
13473 {
13474         struct ctl_softc *softc = CTL_SOFTC(io);
13475         struct ctl_thread *thr;
13476         u_int idx;
13477
13478         idx = (io->io_hdr.nexus.targ_port * 127 +
13479                io->io_hdr.nexus.initid) % worker_threads;
13480         thr = &softc->threads[idx];
13481         mtx_lock(&thr->queue_lock);
13482         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13483         mtx_unlock(&thr->queue_lock);
13484         wakeup(thr);
13485 }
13486
13487 static void
13488 ctl_enqueue_rtr(union ctl_io *io)
13489 {
13490         struct ctl_softc *softc = CTL_SOFTC(io);
13491         struct ctl_thread *thr;
13492
13493         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13494         mtx_lock(&thr->queue_lock);
13495         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13496         mtx_unlock(&thr->queue_lock);
13497         wakeup(thr);
13498 }
13499
13500 static void
13501 ctl_enqueue_done(union ctl_io *io)
13502 {
13503         struct ctl_softc *softc = CTL_SOFTC(io);
13504         struct ctl_thread *thr;
13505
13506         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13507         mtx_lock(&thr->queue_lock);
13508         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13509         mtx_unlock(&thr->queue_lock);
13510         wakeup(thr);
13511 }
13512
13513 static void
13514 ctl_enqueue_isc(union ctl_io *io)
13515 {
13516         struct ctl_softc *softc = CTL_SOFTC(io);
13517         struct ctl_thread *thr;
13518
13519         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13520         mtx_lock(&thr->queue_lock);
13521         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13522         mtx_unlock(&thr->queue_lock);
13523         wakeup(thr);
13524 }
13525
13526 /*
13527  *  vim: ts=8
13528  */