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_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 482 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
 483                             ctl_ua_type ua_type);
 484 static int ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io,
 485                          ctl_ua_type ua_type);
 486 static int ctl_lun_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 487 static int ctl_abort_task(union ctl_io *io);
 488 static int ctl_abort_task_set(union ctl_io *io);
 489 static int ctl_query_task(union ctl_io *io, int task_set);
 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         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1292         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1293         free(port->wwpn_iid[iid].name, M_CTL);
1294         if (msg->iid.name_len) {
1295                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1296                     msg->iid.name_len, M_CTL);
1297         } else
1298                 port->wwpn_iid[iid].name = NULL;
1299 }
1300
1301 static void
1302 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1303 {
1304
1305         if (msg->login.version != CTL_HA_VERSION) {
1306                 printf("CTL HA peers have different versions %d != %d\n",
1307                     msg->login.version, CTL_HA_VERSION);
1308                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1309                 return;
1310         }
1311         if (msg->login.ha_mode != softc->ha_mode) {
1312                 printf("CTL HA peers have different ha_mode %d != %d\n",
1313                     msg->login.ha_mode, softc->ha_mode);
1314                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1315                 return;
1316         }
1317         if (msg->login.ha_id == softc->ha_id) {
1318                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1319                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1320                 return;
1321         }
1322         if (msg->login.max_luns != CTL_MAX_LUNS ||
1323             msg->login.max_ports != CTL_MAX_PORTS ||
1324             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1325                 printf("CTL HA peers have different limits\n");
1326                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1327                 return;
1328         }
1329 }
1330
1331 static void
1332 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1333 {
1334         struct ctl_lun *lun;
1335         u_int i;
1336         uint32_t initidx, targ_lun;
1337
1338         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1339         mtx_lock(&softc->ctl_lock);
1340         if (targ_lun >= CTL_MAX_LUNS ||
1341             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1342                 mtx_unlock(&softc->ctl_lock);
1343                 return;
1344         }
1345         mtx_lock(&lun->lun_lock);
1346         mtx_unlock(&softc->ctl_lock);
1347         if (lun->flags & CTL_LUN_DISABLED) {
1348                 mtx_unlock(&lun->lun_lock);
1349                 return;
1350         }
1351         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1352                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1353                     msg->mode.page_code &&
1354                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1355                         break;
1356         }
1357         if (i == CTL_NUM_MODE_PAGES) {
1358                 mtx_unlock(&lun->lun_lock);
1359                 return;
1360         }
1361         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1362             lun->mode_pages.index[i].page_len);
1363         initidx = ctl_get_initindex(&msg->hdr.nexus);
1364         if (initidx != -1)
1365                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1366         mtx_unlock(&lun->lun_lock);
1367 }
1368
1369 /*
1370  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1371  * subsystem come in here.
1372  */
1373 static void
1374 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1375 {
1376         struct ctl_softc *softc = control_softc;
1377         union ctl_io *io;
1378         struct ctl_prio *presio;
1379         ctl_ha_status isc_status;
1380
1381         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1382         if (event == CTL_HA_EVT_MSG_RECV) {
1383                 union ctl_ha_msg *msg, msgbuf;
1384
1385                 if (param > sizeof(msgbuf))
1386                         msg = malloc(param, M_CTL, M_WAITOK);
1387                 else
1388                         msg = &msgbuf;
1389                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1390                     M_WAITOK);
1391                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1392                         printf("%s: Error receiving message: %d\n",
1393                             __func__, isc_status);
1394                         if (msg != &msgbuf)
1395                                 free(msg, M_CTL);
1396                         return;
1397                 }
1398
1399                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1400                 switch (msg->hdr.msg_type) {
1401                 case CTL_MSG_SERIALIZE:
1402                         io = ctl_alloc_io(softc->othersc_pool);
1403                         ctl_zero_io(io);
1404                         // populate ctsio from msg
1405                         io->io_hdr.io_type = CTL_IO_SCSI;
1406                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1407                         io->io_hdr.original_sc = msg->hdr.original_sc;
1408                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1409                                             CTL_FLAG_IO_ACTIVE;
1410                         /*
1411                          * If we're in serialization-only mode, we don't
1412                          * want to go through full done processing.  Thus
1413                          * the COPY flag.
1414                          *
1415                          * XXX KDM add another flag that is more specific.
1416                          */
1417                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1418                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1419                         io->io_hdr.nexus = msg->hdr.nexus;
1420 #if 0
1421                         printf("port %u, iid %u, lun %u\n",
1422                                io->io_hdr.nexus.targ_port,
1423                                io->io_hdr.nexus.initid,
1424                                io->io_hdr.nexus.targ_lun);
1425 #endif
1426                         io->scsiio.tag_num = msg->scsi.tag_num;
1427                         io->scsiio.tag_type = msg->scsi.tag_type;
1428 #ifdef CTL_TIME_IO
1429                         io->io_hdr.start_time = time_uptime;
1430                         getbinuptime(&io->io_hdr.start_bt);
1431 #endif /* CTL_TIME_IO */
1432                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1433                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1434                                CTL_MAX_CDBLEN);
1435                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1436                                 const struct ctl_cmd_entry *entry;
1437
1438                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1439                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1440                                 io->io_hdr.flags |=
1441                                         entry->flags & CTL_FLAG_DATA_MASK;
1442                         }
1443                         ctl_enqueue_isc(io);
1444                         break;
1445
1446                 /* Performed on the Originating SC, XFER mode only */
1447                 case CTL_MSG_DATAMOVE: {
1448                         struct ctl_sg_entry *sgl;
1449                         int i, j;
1450
1451                         io = msg->hdr.original_sc;
1452                         if (io == NULL) {
1453                                 printf("%s: original_sc == NULL!\n", __func__);
1454                                 /* XXX KDM do something here */
1455                                 break;
1456                         }
1457                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1458                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1459                         /*
1460                          * Keep track of this, we need to send it back over
1461                          * when the datamove is complete.
1462                          */
1463                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1464                         if (msg->hdr.status == CTL_SUCCESS)
1465                                 io->io_hdr.status = msg->hdr.status;
1466
1467                         if (msg->dt.sg_sequence == 0) {
1468 #ifdef CTL_TIME_IO
1469                                 getbinuptime(&io->io_hdr.dma_start_bt);
1470 #endif
1471                                 i = msg->dt.kern_sg_entries +
1472                                     msg->dt.kern_data_len /
1473                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1474                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1475                                     M_WAITOK | M_ZERO);
1476                                 io->io_hdr.remote_sglist = sgl;
1477                                 io->io_hdr.local_sglist =
1478                                     &sgl[msg->dt.kern_sg_entries];
1479
1480                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1481
1482                                 io->scsiio.kern_sg_entries =
1483                                         msg->dt.kern_sg_entries;
1484                                 io->scsiio.rem_sg_entries =
1485                                         msg->dt.kern_sg_entries;
1486                                 io->scsiio.kern_data_len =
1487                                         msg->dt.kern_data_len;
1488                                 io->scsiio.kern_total_len =
1489                                         msg->dt.kern_total_len;
1490                                 io->scsiio.kern_data_resid =
1491                                         msg->dt.kern_data_resid;
1492                                 io->scsiio.kern_rel_offset =
1493                                         msg->dt.kern_rel_offset;
1494                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1495                                 io->io_hdr.flags |= msg->dt.flags &
1496                                     CTL_FLAG_BUS_ADDR;
1497                         } else
1498                                 sgl = (struct ctl_sg_entry *)
1499                                         io->scsiio.kern_data_ptr;
1500
1501                         for (i = msg->dt.sent_sg_entries, j = 0;
1502                              i < (msg->dt.sent_sg_entries +
1503                              msg->dt.cur_sg_entries); i++, j++) {
1504                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1505                                 sgl[i].len = msg->dt.sg_list[j].len;
1506
1507 #if 0
1508                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1509                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1510 #endif
1511                         }
1512
1513                         /*
1514                          * If this is the last piece of the I/O, we've got
1515                          * the full S/G list.  Queue processing in the thread.
1516                          * Otherwise wait for the next piece.
1517                          */
1518                         if (msg->dt.sg_last != 0)
1519                                 ctl_enqueue_isc(io);
1520                         break;
1521                 }
1522                 /* Performed on the Serializing (primary) SC, XFER mode only */
1523                 case CTL_MSG_DATAMOVE_DONE: {
1524                         if (msg->hdr.serializing_sc == NULL) {
1525                                 printf("%s: serializing_sc == NULL!\n",
1526                                        __func__);
1527                                 /* XXX KDM now what? */
1528                                 break;
1529                         }
1530                         /*
1531                          * We grab the sense information here in case
1532                          * there was a failure, so we can return status
1533                          * back to the initiator.
1534                          */
1535                         io = msg->hdr.serializing_sc;
1536                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1537                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1538                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1539                         io->io_hdr.port_status = msg->scsi.port_status;
1540                         io->scsiio.kern_data_resid = msg->scsi.kern_data_resid;
1541                         if (msg->hdr.status != CTL_STATUS_NONE) {
1542                                 io->io_hdr.status = msg->hdr.status;
1543                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1544                                 io->scsiio.sense_len = msg->scsi.sense_len;
1545                                 memcpy(&io->scsiio.sense_data,
1546                                     &msg->scsi.sense_data,
1547                                     msg->scsi.sense_len);
1548                                 if (msg->hdr.status == CTL_SUCCESS)
1549                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1550                         }
1551                         ctl_enqueue_isc(io);
1552                         break;
1553                 }
1554
1555                 /* Preformed on Originating SC, SER_ONLY mode */
1556                 case CTL_MSG_R2R:
1557                         io = msg->hdr.original_sc;
1558                         if (io == NULL) {
1559                                 printf("%s: original_sc == NULL!\n",
1560                                     __func__);
1561                                 break;
1562                         }
1563                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1564                         io->io_hdr.msg_type = CTL_MSG_R2R;
1565                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1566                         ctl_enqueue_isc(io);
1567                         break;
1568
1569                 /*
1570                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1571                  * mode.
1572                  * Performed on the Originating (i.e. secondary) SC in XFER
1573                  * mode
1574                  */
1575                 case CTL_MSG_FINISH_IO:
1576                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1577                                 ctl_isc_handler_finish_xfer(softc, msg);
1578                         else
1579                                 ctl_isc_handler_finish_ser_only(softc, msg);
1580                         break;
1581
1582                 /* Preformed on Originating SC */
1583                 case CTL_MSG_BAD_JUJU:
1584                         io = msg->hdr.original_sc;
1585                         if (io == NULL) {
1586                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1587                                        __func__);
1588                                 break;
1589                         }
1590                         ctl_copy_sense_data(msg, io);
1591                         /*
1592                          * IO should have already been cleaned up on other
1593                          * SC so clear this flag so we won't send a message
1594                          * back to finish the IO there.
1595                          */
1596                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1597                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1598
1599                         /* io = msg->hdr.serializing_sc; */
1600                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1601                         ctl_enqueue_isc(io);
1602                         break;
1603
1604                 /* Handle resets sent from the other side */
1605                 case CTL_MSG_MANAGE_TASKS: {
1606                         struct ctl_taskio *taskio;
1607                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1608                             softc->othersc_pool);
1609                         ctl_zero_io((union ctl_io *)taskio);
1610                         taskio->io_hdr.io_type = CTL_IO_TASK;
1611                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1612                         taskio->io_hdr.nexus = msg->hdr.nexus;
1613                         taskio->task_action = msg->task.task_action;
1614                         taskio->tag_num = msg->task.tag_num;
1615                         taskio->tag_type = msg->task.tag_type;
1616 #ifdef CTL_TIME_IO
1617                         taskio->io_hdr.start_time = time_uptime;
1618                         getbinuptime(&taskio->io_hdr.start_bt);
1619 #endif /* CTL_TIME_IO */
1620                         ctl_run_task((union ctl_io *)taskio);
1621                         break;
1622                 }
1623                 /* Persistent Reserve action which needs attention */
1624                 case CTL_MSG_PERS_ACTION:
1625                         presio = (struct ctl_prio *)ctl_alloc_io(
1626                             softc->othersc_pool);
1627                         ctl_zero_io((union ctl_io *)presio);
1628                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1629                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1630                         presio->io_hdr.nexus = msg->hdr.nexus;
1631                         presio->pr_msg = msg->pr;
1632                         ctl_enqueue_isc((union ctl_io *)presio);
1633                         break;
1634                 case CTL_MSG_UA:
1635                         ctl_isc_ua(softc, msg, param);
1636                         break;
1637                 case CTL_MSG_PORT_SYNC:
1638                         ctl_isc_port_sync(softc, msg, param);
1639                         break;
1640                 case CTL_MSG_LUN_SYNC:
1641                         ctl_isc_lun_sync(softc, msg, param);
1642                         break;
1643                 case CTL_MSG_IID_SYNC:
1644                         ctl_isc_iid_sync(softc, msg, param);
1645                         break;
1646                 case CTL_MSG_LOGIN:
1647                         ctl_isc_login(softc, msg, param);
1648                         break;
1649                 case CTL_MSG_MODE_SYNC:
1650                         ctl_isc_mode_sync(softc, msg, param);
1651                         break;
1652                 default:
1653                         printf("Received HA message of unknown type %d\n",
1654                             msg->hdr.msg_type);
1655                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1656                         break;
1657                 }
1658                 if (msg != &msgbuf)
1659                         free(msg, M_CTL);
1660         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1661                 printf("CTL: HA link status changed from %d to %d\n",
1662                     softc->ha_link, param);
1663                 if (param == softc->ha_link)
1664                         return;
1665                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1666                         softc->ha_link = param;
1667                         ctl_isc_ha_link_down(softc);
1668                 } else {
1669                         softc->ha_link = param;
1670                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1671                                 ctl_isc_ha_link_up(softc);
1672                 }
1673                 return;
1674         } else {
1675                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1676                 return;
1677         }
1678 }
1679
1680 static void
1681 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1682 {
1683
1684         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1685             src->scsi.sense_len);
1686         dest->scsiio.scsi_status = src->scsi.scsi_status;
1687         dest->scsiio.sense_len = src->scsi.sense_len;
1688         dest->io_hdr.status = src->hdr.status;
1689 }
1690
1691 static void
1692 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1693 {
1694
1695         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1696             src->scsiio.sense_len);
1697         dest->scsi.scsi_status = src->scsiio.scsi_status;
1698         dest->scsi.sense_len = src->scsiio.sense_len;
1699         dest->hdr.status = src->io_hdr.status;
1700 }
1701
1702 void
1703 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1704 {
1705         struct ctl_softc *softc = lun->ctl_softc;
1706         ctl_ua_type *pu;
1707
1708         if (initidx < softc->init_min || initidx >= softc->init_max)
1709                 return;
1710         mtx_assert(&lun->lun_lock, MA_OWNED);
1711         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1712         if (pu == NULL)
1713                 return;
1714         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1715 }
1716
1717 void
1718 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1719 {
1720         int i;
1721
1722         mtx_assert(&lun->lun_lock, MA_OWNED);
1723         if (lun->pending_ua[port] == NULL)
1724                 return;
1725         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1726                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1727                         continue;
1728                 lun->pending_ua[port][i] |= ua;
1729         }
1730 }
1731
1732 void
1733 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1734 {
1735         struct ctl_softc *softc = lun->ctl_softc;
1736         int i;
1737
1738         mtx_assert(&lun->lun_lock, MA_OWNED);
1739         for (i = softc->port_min; i < softc->port_max; i++)
1740                 ctl_est_ua_port(lun, i, except, ua);
1741 }
1742
1743 void
1744 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1745 {
1746         struct ctl_softc *softc = lun->ctl_softc;
1747         ctl_ua_type *pu;
1748
1749         if (initidx < softc->init_min || initidx >= softc->init_max)
1750                 return;
1751         mtx_assert(&lun->lun_lock, MA_OWNED);
1752         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1753         if (pu == NULL)
1754                 return;
1755         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1756 }
1757
1758 void
1759 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1760 {
1761         struct ctl_softc *softc = lun->ctl_softc;
1762         int i, j;
1763
1764         mtx_assert(&lun->lun_lock, MA_OWNED);
1765         for (i = softc->port_min; i < softc->port_max; i++) {
1766                 if (lun->pending_ua[i] == NULL)
1767                         continue;
1768                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1769                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1770                                 continue;
1771                         lun->pending_ua[i][j] &= ~ua;
1772                 }
1773         }
1774 }
1775
1776 void
1777 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1778     ctl_ua_type ua_type)
1779 {
1780         struct ctl_lun *lun;
1781
1782         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1783         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1784                 mtx_lock(&lun->lun_lock);
1785                 ctl_clr_ua(lun, initidx, ua_type);
1786                 mtx_unlock(&lun->lun_lock);
1787         }
1788 }
1789
1790 static int
1791 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1792 {
1793         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1794         struct ctl_lun *lun;
1795         struct ctl_lun_req ireq;
1796         int error, value;
1797
1798         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1799         error = sysctl_handle_int(oidp, &value, 0, req);
1800         if ((error != 0) || (req->newptr == NULL))
1801                 return (error);
1802
1803         mtx_lock(&softc->ctl_lock);
1804         if (value == 0)
1805                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1806         else
1807                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1808         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1809                 mtx_unlock(&softc->ctl_lock);
1810                 bzero(&ireq, sizeof(ireq));
1811                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1812                 ireq.reqdata.modify.lun_id = lun->lun;
1813                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1814                     curthread);
1815                 if (ireq.status != CTL_LUN_OK) {
1816                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1817                             __func__, ireq.status, ireq.error_str);
1818                 }
1819                 mtx_lock(&softc->ctl_lock);
1820         }
1821         mtx_unlock(&softc->ctl_lock);
1822         return (0);
1823 }
1824
1825 static int
1826 ctl_init(void)
1827 {
1828         struct make_dev_args args;
1829         struct ctl_softc *softc;
1830         int i, error;
1831
1832         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1833                                M_WAITOK | M_ZERO);
1834
1835         make_dev_args_init(&args);
1836         args.mda_devsw = &ctl_cdevsw;
1837         args.mda_uid = UID_ROOT;
1838         args.mda_gid = GID_OPERATOR;
1839         args.mda_mode = 0600;
1840         args.mda_si_drv1 = softc;
1841         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1842         if (error != 0) {
1843                 free(softc, M_DEVBUF);
1844                 control_softc = NULL;
1845                 return (error);
1846         }
1847
1848         sysctl_ctx_init(&softc->sysctl_ctx);
1849         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1850                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1851                 CTLFLAG_RD, 0, "CAM Target Layer");
1852
1853         if (softc->sysctl_tree == NULL) {
1854                 printf("%s: unable to allocate sysctl tree\n", __func__);
1855                 destroy_dev(softc->dev);
1856                 free(softc, M_DEVBUF);
1857                 control_softc = NULL;
1858                 return (ENOMEM);
1859         }
1860
1861         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1862         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1863             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1864         softc->flags = 0;
1865
1866         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1867             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1868             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1869
1870         /*
1871          * In Copan's HA scheme, the "master" and "slave" roles are
1872          * figured out through the slot the controller is in.  Although it
1873          * is an active/active system, someone has to be in charge.
1874          */
1875         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1876             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1877             "HA head ID (0 - no HA)");
1878         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1879                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1880                 softc->is_single = 1;
1881                 softc->port_cnt = CTL_MAX_PORTS;
1882                 softc->port_min = 0;
1883         } else {
1884                 softc->port_cnt = CTL_MAX_PORTS / NUM_HA_SHELVES;
1885                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1886         }
1887         softc->port_max = softc->port_min + softc->port_cnt;
1888         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1889         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1890
1891         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1892             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1893             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1894
1895         STAILQ_INIT(&softc->lun_list);
1896         STAILQ_INIT(&softc->pending_lun_queue);
1897         STAILQ_INIT(&softc->fe_list);
1898         STAILQ_INIT(&softc->port_list);
1899         STAILQ_INIT(&softc->be_list);
1900         ctl_tpc_init(softc);
1901
1902         if (worker_threads <= 0)
1903                 worker_threads = max(1, mp_ncpus / 4);
1904         if (worker_threads > CTL_MAX_THREADS)
1905                 worker_threads = CTL_MAX_THREADS;
1906
1907         for (i = 0; i < worker_threads; i++) {
1908                 struct ctl_thread *thr = &softc->threads[i];
1909
1910                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1911                 thr->ctl_softc = softc;
1912                 STAILQ_INIT(&thr->incoming_queue);
1913                 STAILQ_INIT(&thr->rtr_queue);
1914                 STAILQ_INIT(&thr->done_queue);
1915                 STAILQ_INIT(&thr->isc_queue);
1916
1917                 error = kproc_kthread_add(ctl_work_thread, thr,
1918                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1919                 if (error != 0) {
1920                         printf("error creating CTL work thread!\n");
1921                         return (error);
1922                 }
1923         }
1924         error = kproc_kthread_add(ctl_lun_thread, softc,
1925             &softc->ctl_proc, &softc->lun_thread, 0, 0, "ctl", "lun");
1926         if (error != 0) {
1927                 printf("error creating CTL lun thread!\n");
1928                 return (error);
1929         }
1930         error = kproc_kthread_add(ctl_thresh_thread, softc,
1931             &softc->ctl_proc, &softc->thresh_thread, 0, 0, "ctl", "thresh");
1932         if (error != 0) {
1933                 printf("error creating CTL threshold thread!\n");
1934                 return (error);
1935         }
1936
1937         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1938             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1939             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1940
1941         if (softc->is_single == 0) {
1942                 if (ctl_frontend_register(&ha_frontend) != 0)
1943                         softc->is_single = 1;
1944         }
1945         return (0);
1946 }
1947
1948 static int
1949 ctl_shutdown(void)
1950 {
1951         struct ctl_softc *softc = control_softc;
1952         int i;
1953
1954         if (softc->is_single == 0)
1955                 ctl_frontend_deregister(&ha_frontend);
1956
1957         destroy_dev(softc->dev);
1958
1959         /* Shutdown CTL threads. */
1960         softc->shutdown = 1;
1961         for (i = 0; i < worker_threads; i++) {
1962                 struct ctl_thread *thr = &softc->threads[i];
1963                 while (thr->thread != NULL) {
1964                         wakeup(thr);
1965                         if (thr->thread != NULL)
1966                                 pause("CTL thr shutdown", 1);
1967                 }
1968                 mtx_destroy(&thr->queue_lock);
1969         }
1970         while (softc->lun_thread != NULL) {
1971                 wakeup(&softc->pending_lun_queue);
1972                 if (softc->lun_thread != NULL)
1973                         pause("CTL thr shutdown", 1);
1974         }
1975         while (softc->thresh_thread != NULL) {
1976                 wakeup(softc->thresh_thread);
1977                 if (softc->thresh_thread != NULL)
1978                         pause("CTL thr shutdown", 1);
1979         }
1980
1981         ctl_tpc_shutdown(softc);
1982         uma_zdestroy(softc->io_zone);
1983         mtx_destroy(&softc->ctl_lock);
1984
1985         sysctl_ctx_free(&softc->sysctl_ctx);
1986
1987         free(softc, M_DEVBUF);
1988         control_softc = NULL;
1989         return (0);
1990 }
1991
1992 static int
1993 ctl_module_event_handler(module_t mod, int what, void *arg)
1994 {
1995
1996         switch (what) {
1997         case MOD_LOAD:
1998                 return (ctl_init());
1999         case MOD_UNLOAD:
2000                 return (ctl_shutdown());
2001         default:
2002                 return (EOPNOTSUPP);
2003         }
2004 }
2005
2006 /*
2007  * XXX KDM should we do some access checks here?  Bump a reference count to
2008  * prevent a CTL module from being unloaded while someone has it open?
2009  */
2010 static int
2011 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
2012 {
2013         return (0);
2014 }
2015
2016 static int
2017 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2018 {
2019         return (0);
2020 }
2021
2022 /*
2023  * Remove an initiator by port number and initiator ID.
2024  * Returns 0 for success, -1 for failure.
2025  */
2026 int
2027 ctl_remove_initiator(struct ctl_port *port, int iid)
2028 {
2029         struct ctl_softc *softc = port->ctl_softc;
2030
2031         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2032
2033         if (iid > CTL_MAX_INIT_PER_PORT) {
2034                 printf("%s: initiator ID %u > maximun %u!\n",
2035                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2036                 return (-1);
2037         }
2038
2039         mtx_lock(&softc->ctl_lock);
2040         port->wwpn_iid[iid].in_use--;
2041         port->wwpn_iid[iid].last_use = time_uptime;
2042         mtx_unlock(&softc->ctl_lock);
2043         ctl_isc_announce_iid(port, iid);
2044
2045         return (0);
2046 }
2047
2048 /*
2049  * Add an initiator to the initiator map.
2050  * Returns iid for success, < 0 for failure.
2051  */
2052 int
2053 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2054 {
2055         struct ctl_softc *softc = port->ctl_softc;
2056         time_t best_time;
2057         int i, best;
2058
2059         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2060
2061         if (iid >= CTL_MAX_INIT_PER_PORT) {
2062                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2063                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2064                 free(name, M_CTL);
2065                 return (-1);
2066         }
2067
2068         mtx_lock(&softc->ctl_lock);
2069
2070         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2071                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2072                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2073                                 iid = i;
2074                                 break;
2075                         }
2076                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2077                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2078                                 iid = i;
2079                                 break;
2080                         }
2081                 }
2082         }
2083
2084         if (iid < 0) {
2085                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2086                         if (port->wwpn_iid[i].in_use == 0 &&
2087                             port->wwpn_iid[i].wwpn == 0 &&
2088                             port->wwpn_iid[i].name == NULL) {
2089                                 iid = i;
2090                                 break;
2091                         }
2092                 }
2093         }
2094
2095         if (iid < 0) {
2096                 best = -1;
2097                 best_time = INT32_MAX;
2098                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2099                         if (port->wwpn_iid[i].in_use == 0) {
2100                                 if (port->wwpn_iid[i].last_use < best_time) {
2101                                         best = i;
2102                                         best_time = port->wwpn_iid[i].last_use;
2103                                 }
2104                         }
2105                 }
2106                 iid = best;
2107         }
2108
2109         if (iid < 0) {
2110                 mtx_unlock(&softc->ctl_lock);
2111                 free(name, M_CTL);
2112                 return (-2);
2113         }
2114
2115         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2116                 /*
2117                  * This is not an error yet.
2118                  */
2119                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2120 #if 0
2121                         printf("%s: port %d iid %u WWPN %#jx arrived"
2122                             " again\n", __func__, port->targ_port,
2123                             iid, (uintmax_t)wwpn);
2124 #endif
2125                         goto take;
2126                 }
2127                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2128                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2129 #if 0
2130                         printf("%s: port %d iid %u name '%s' arrived"
2131                             " again\n", __func__, port->targ_port,
2132                             iid, name);
2133 #endif
2134                         goto take;
2135                 }
2136
2137                 /*
2138                  * This is an error, but what do we do about it?  The
2139                  * driver is telling us we have a new WWPN for this
2140                  * initiator ID, so we pretty much need to use it.
2141                  */
2142                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2143                     " but WWPN %#jx '%s' is still at that address\n",
2144                     __func__, port->targ_port, iid, wwpn, name,
2145                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2146                     port->wwpn_iid[iid].name);
2147
2148                 /*
2149                  * XXX KDM clear have_ca and ua_pending on each LUN for
2150                  * this initiator.
2151                  */
2152         }
2153 take:
2154         free(port->wwpn_iid[iid].name, M_CTL);
2155         port->wwpn_iid[iid].name = name;
2156         port->wwpn_iid[iid].wwpn = wwpn;
2157         port->wwpn_iid[iid].in_use++;
2158         mtx_unlock(&softc->ctl_lock);
2159         ctl_isc_announce_iid(port, iid);
2160
2161         return (iid);
2162 }
2163
2164 static int
2165 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2166 {
2167         int len;
2168
2169         switch (port->port_type) {
2170         case CTL_PORT_FC:
2171         {
2172                 struct scsi_transportid_fcp *id =
2173                     (struct scsi_transportid_fcp *)buf;
2174                 if (port->wwpn_iid[iid].wwpn == 0)
2175                         return (0);
2176                 memset(id, 0, sizeof(*id));
2177                 id->format_protocol = SCSI_PROTO_FC;
2178                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2179                 return (sizeof(*id));
2180         }
2181         case CTL_PORT_ISCSI:
2182         {
2183                 struct scsi_transportid_iscsi_port *id =
2184                     (struct scsi_transportid_iscsi_port *)buf;
2185                 if (port->wwpn_iid[iid].name == NULL)
2186                         return (0);
2187                 memset(id, 0, 256);
2188                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2189                     SCSI_PROTO_ISCSI;
2190                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2191                 len = roundup2(min(len, 252), 4);
2192                 scsi_ulto2b(len, id->additional_length);
2193                 return (sizeof(*id) + len);
2194         }
2195         case CTL_PORT_SAS:
2196         {
2197                 struct scsi_transportid_sas *id =
2198                     (struct scsi_transportid_sas *)buf;
2199                 if (port->wwpn_iid[iid].wwpn == 0)
2200                         return (0);
2201                 memset(id, 0, sizeof(*id));
2202                 id->format_protocol = SCSI_PROTO_SAS;
2203                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2204                 return (sizeof(*id));
2205         }
2206         default:
2207         {
2208                 struct scsi_transportid_spi *id =
2209                     (struct scsi_transportid_spi *)buf;
2210                 memset(id, 0, sizeof(*id));
2211                 id->format_protocol = SCSI_PROTO_SPI;
2212                 scsi_ulto2b(iid, id->scsi_addr);
2213                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2214                 return (sizeof(*id));
2215         }
2216         }
2217 }
2218
2219 /*
2220  * Serialize a command that went down the "wrong" side, and so was sent to
2221  * this controller for execution.  The logic is a little different than the
2222  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2223  * sent back to the other side, but in the success case, we execute the
2224  * command on this side (XFER mode) or tell the other side to execute it
2225  * (SER_ONLY mode).
2226  */
2227 static void
2228 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2229 {
2230         struct ctl_softc *softc = CTL_SOFTC(ctsio);
2231         struct ctl_port *port = CTL_PORT(ctsio);
2232         union ctl_ha_msg msg_info;
2233         struct ctl_lun *lun;
2234         const struct ctl_cmd_entry *entry;
2235         uint32_t targ_lun;
2236
2237         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2238
2239         /* Make sure that we know about this port. */
2240         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2241                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2242                                          /*retry_count*/ 1);
2243                 goto badjuju;
2244         }
2245
2246         /* Make sure that we know about this LUN. */
2247         mtx_lock(&softc->ctl_lock);
2248         if (targ_lun >= CTL_MAX_LUNS ||
2249             (lun = softc->ctl_luns[targ_lun]) == NULL) {
2250                 mtx_unlock(&softc->ctl_lock);
2251
2252                 /*
2253                  * The other node would not send this request to us unless
2254                  * received announce that we are primary node for this LUN.
2255                  * If this LUN does not exist now, it is probably result of
2256                  * a race, so respond to initiator in the most opaque way.
2257                  */
2258                 ctl_set_busy(ctsio);
2259                 goto badjuju;
2260         }
2261         mtx_lock(&lun->lun_lock);
2262         mtx_unlock(&softc->ctl_lock);
2263
2264         /*
2265          * If the LUN is invalid, pretend that it doesn't exist.
2266          * It will go away as soon as all pending I/Os completed.
2267          */
2268         if (lun->flags & CTL_LUN_DISABLED) {
2269                 mtx_unlock(&lun->lun_lock);
2270                 ctl_set_busy(ctsio);
2271                 goto badjuju;
2272         }
2273
2274         entry = ctl_get_cmd_entry(ctsio, NULL);
2275         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2276                 mtx_unlock(&lun->lun_lock);
2277                 goto badjuju;
2278         }
2279
2280         CTL_LUN(ctsio) = lun;
2281         CTL_BACKEND_LUN(ctsio) = lun->be_lun;
2282
2283         /*
2284          * Every I/O goes into the OOA queue for a
2285          * particular LUN, and stays there until completion.
2286          */
2287 #ifdef CTL_TIME_IO
2288         if (TAILQ_EMPTY(&lun->ooa_queue))
2289                 lun->idle_time += getsbinuptime() - lun->last_busy;
2290 #endif
2291         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2292
2293         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2294                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2295                  ooa_links))) {
2296         case CTL_ACTION_BLOCK:
2297                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2298                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2299                                   blocked_links);
2300                 mtx_unlock(&lun->lun_lock);
2301                 break;
2302         case CTL_ACTION_PASS:
2303         case CTL_ACTION_SKIP:
2304                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2305                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2306                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2307                         mtx_unlock(&lun->lun_lock);
2308                 } else {
2309                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2310                         mtx_unlock(&lun->lun_lock);
2311
2312                         /* send msg back to other side */
2313                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2314                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2315                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2316                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2317                             sizeof(msg_info.hdr), M_WAITOK);
2318                 }
2319                 break;
2320         case CTL_ACTION_OVERLAP:
2321                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2322                 mtx_unlock(&lun->lun_lock);
2323                 ctl_set_overlapped_cmd(ctsio);
2324                 goto badjuju;
2325         case CTL_ACTION_OVERLAP_TAG:
2326                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2327                 mtx_unlock(&lun->lun_lock);
2328                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2329                 goto badjuju;
2330         case CTL_ACTION_ERROR:
2331         default:
2332                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2333                 mtx_unlock(&lun->lun_lock);
2334
2335                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2336                                          /*retry_count*/ 0);
2337 badjuju:
2338                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2339                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2340                 msg_info.hdr.serializing_sc = NULL;
2341                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2342                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2343                     sizeof(msg_info.scsi), M_WAITOK);
2344                 ctl_free_io((union ctl_io *)ctsio);
2345                 break;
2346         }
2347 }
2348
2349 /*
2350  * Returns 0 for success, errno for failure.
2351  */
2352 static void
2353 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2354                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2355 {
2356         union ctl_io *io;
2357
2358         mtx_lock(&lun->lun_lock);
2359         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2360              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2361              ooa_links)) {
2362                 struct ctl_ooa_entry *entry;
2363
2364                 /*
2365                  * If we've got more than we can fit, just count the
2366                  * remaining entries.
2367                  */
2368                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2369                         continue;
2370
2371                 entry = &kern_entries[*cur_fill_num];
2372
2373                 entry->tag_num = io->scsiio.tag_num;
2374                 entry->lun_num = lun->lun;
2375 #ifdef CTL_TIME_IO
2376                 entry->start_bt = io->io_hdr.start_bt;
2377 #endif
2378                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2379                 entry->cdb_len = io->scsiio.cdb_len;
2380                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2381                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2382
2383                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2384                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2385
2386                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2387                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2388
2389                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2390                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2391
2392                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2393                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2394         }
2395         mtx_unlock(&lun->lun_lock);
2396 }
2397
2398 static void *
2399 ctl_copyin_alloc(void *user_addr, unsigned int len, char *error_str,
2400                  size_t error_str_len)
2401 {
2402         void *kptr;
2403
2404         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2405
2406         if (copyin(user_addr, kptr, len) != 0) {
2407                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2408                          "from user address %p to kernel address %p", len,
2409                          user_addr, kptr);
2410                 free(kptr, M_CTL);
2411                 return (NULL);
2412         }
2413
2414         return (kptr);
2415 }
2416
2417 static void
2418 ctl_free_args(int num_args, struct ctl_be_arg *args)
2419 {
2420         int i;
2421
2422         if (args == NULL)
2423                 return;
2424
2425         for (i = 0; i < num_args; i++) {
2426                 free(args[i].kname, M_CTL);
2427                 free(args[i].kvalue, M_CTL);
2428         }
2429
2430         free(args, M_CTL);
2431 }
2432
2433 static struct ctl_be_arg *
2434 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2435                 char *error_str, size_t error_str_len)
2436 {
2437         struct ctl_be_arg *args;
2438         int i;
2439
2440         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2441                                 error_str, error_str_len);
2442
2443         if (args == NULL)
2444                 goto bailout;
2445
2446         for (i = 0; i < num_args; i++) {
2447                 args[i].kname = NULL;
2448                 args[i].kvalue = NULL;
2449         }
2450
2451         for (i = 0; i < num_args; i++) {
2452                 uint8_t *tmpptr;
2453
2454                 if (args[i].namelen == 0) {
2455                         snprintf(error_str, error_str_len, "Argument %d "
2456                                  "name length is zero", i);
2457                         goto bailout;
2458                 }
2459
2460                 args[i].kname = ctl_copyin_alloc(args[i].name,
2461                         args[i].namelen, error_str, error_str_len);
2462                 if (args[i].kname == NULL)
2463                         goto bailout;
2464
2465                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2466                         snprintf(error_str, error_str_len, "Argument %d "
2467                                  "name is not NUL-terminated", i);
2468                         goto bailout;
2469                 }
2470
2471                 if (args[i].flags & CTL_BEARG_RD) {
2472                         if (args[i].vallen == 0) {
2473                                 snprintf(error_str, error_str_len, "Argument %d "
2474                                          "value length is zero", i);
2475                                 goto bailout;
2476                         }
2477
2478                         tmpptr = ctl_copyin_alloc(args[i].value,
2479                                 args[i].vallen, error_str, error_str_len);
2480                         if (tmpptr == NULL)
2481                                 goto bailout;
2482
2483                         if ((args[i].flags & CTL_BEARG_ASCII)
2484                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2485                                 snprintf(error_str, error_str_len, "Argument "
2486                                     "%d value is not NUL-terminated", i);
2487                                 free(tmpptr, M_CTL);
2488                                 goto bailout;
2489                         }
2490                         args[i].kvalue = tmpptr;
2491                 } else {
2492                         args[i].kvalue = malloc(args[i].vallen,
2493                             M_CTL, M_WAITOK | M_ZERO);
2494                 }
2495         }
2496
2497         return (args);
2498 bailout:
2499
2500         ctl_free_args(num_args, args);
2501
2502         return (NULL);
2503 }
2504
2505 static void
2506 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2507 {
2508         int i;
2509
2510         for (i = 0; i < num_args; i++) {
2511                 if (args[i].flags & CTL_BEARG_WR)
2512                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2513         }
2514 }
2515
2516 /*
2517  * Escape characters that are illegal or not recommended in XML.
2518  */
2519 int
2520 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2521 {
2522         char *end = str + size;
2523         int retval;
2524
2525         retval = 0;
2526
2527         for (; *str && str < end; str++) {
2528                 switch (*str) {
2529                 case '&':
2530                         retval = sbuf_printf(sb, "&amp;");
2531                         break;
2532                 case '>':
2533                         retval = sbuf_printf(sb, "&gt;");
2534                         break;
2535                 case '<':
2536                         retval = sbuf_printf(sb, "&lt;");
2537                         break;
2538                 default:
2539                         retval = sbuf_putc(sb, *str);
2540                         break;
2541                 }
2542
2543                 if (retval != 0)
2544                         break;
2545
2546         }
2547
2548         return (retval);
2549 }
2550
2551 static void
2552 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2553 {
2554         struct scsi_vpd_id_descriptor *desc;
2555         int i;
2556
2557         if (id == NULL || id->len < 4)
2558                 return;
2559         desc = (struct scsi_vpd_id_descriptor *)id->data;
2560         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2561         case SVPD_ID_TYPE_T10:
2562                 sbuf_printf(sb, "t10.");
2563                 break;
2564         case SVPD_ID_TYPE_EUI64:
2565                 sbuf_printf(sb, "eui.");
2566                 break;
2567         case SVPD_ID_TYPE_NAA:
2568                 sbuf_printf(sb, "naa.");
2569                 break;
2570         case SVPD_ID_TYPE_SCSI_NAME:
2571                 break;
2572         }
2573         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2574         case SVPD_ID_CODESET_BINARY:
2575                 for (i = 0; i < desc->length; i++)
2576                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2577                 break;
2578         case SVPD_ID_CODESET_ASCII:
2579                 sbuf_printf(sb, "%.*s", (int)desc->length,
2580                     (char *)desc->identifier);
2581                 break;
2582         case SVPD_ID_CODESET_UTF8:
2583                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2584                 break;
2585         }
2586 }
2587
2588 static int
2589 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2590           struct thread *td)
2591 {
2592         struct ctl_softc *softc = dev->si_drv1;
2593         struct ctl_port *port;
2594         struct ctl_lun *lun;
2595         int retval;
2596
2597         retval = 0;
2598
2599         switch (cmd) {
2600         case CTL_IO:
2601                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2602                 break;
2603         case CTL_ENABLE_PORT:
2604         case CTL_DISABLE_PORT:
2605         case CTL_SET_PORT_WWNS: {
2606                 struct ctl_port *port;
2607                 struct ctl_port_entry *entry;
2608
2609                 entry = (struct ctl_port_entry *)addr;
2610
2611                 mtx_lock(&softc->ctl_lock);
2612                 STAILQ_FOREACH(port, &softc->port_list, links) {
2613                         int action, done;
2614
2615                         if (port->targ_port < softc->port_min ||
2616                             port->targ_port >= softc->port_max)
2617                                 continue;
2618
2619                         action = 0;
2620                         done = 0;
2621                         if ((entry->port_type == CTL_PORT_NONE)
2622                          && (entry->targ_port == port->targ_port)) {
2623                                 /*
2624                                  * If the user only wants to enable or
2625                                  * disable or set WWNs on a specific port,
2626                                  * do the operation and we're done.
2627                                  */
2628                                 action = 1;
2629                                 done = 1;
2630                         } else if (entry->port_type & port->port_type) {
2631                                 /*
2632                                  * Compare the user's type mask with the
2633                                  * particular frontend type to see if we
2634                                  * have a match.
2635                                  */
2636                                 action = 1;
2637                                 done = 0;
2638
2639                                 /*
2640                                  * Make sure the user isn't trying to set
2641                                  * WWNs on multiple ports at the same time.
2642                                  */
2643                                 if (cmd == CTL_SET_PORT_WWNS) {
2644                                         printf("%s: Can't set WWNs on "
2645                                                "multiple ports\n", __func__);
2646                                         retval = EINVAL;
2647                                         break;
2648                                 }
2649                         }
2650                         if (action == 0)
2651                                 continue;
2652
2653                         /*
2654                          * XXX KDM we have to drop the lock here, because
2655                          * the online/offline operations can potentially
2656                          * block.  We need to reference count the frontends
2657                          * so they can't go away,
2658                          */
2659                         if (cmd == CTL_ENABLE_PORT) {
2660                                 mtx_unlock(&softc->ctl_lock);
2661                                 ctl_port_online(port);
2662                                 mtx_lock(&softc->ctl_lock);
2663                         } else if (cmd == CTL_DISABLE_PORT) {
2664                                 mtx_unlock(&softc->ctl_lock);
2665                                 ctl_port_offline(port);
2666                                 mtx_lock(&softc->ctl_lock);
2667                         } else if (cmd == CTL_SET_PORT_WWNS) {
2668                                 ctl_port_set_wwns(port,
2669                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2670                                     1 : 0, entry->wwnn,
2671                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2672                                     1 : 0, entry->wwpn);
2673                         }
2674                         if (done != 0)
2675                                 break;
2676                 }
2677                 mtx_unlock(&softc->ctl_lock);
2678                 break;
2679         }
2680         case CTL_GET_OOA: {
2681                 struct ctl_ooa *ooa_hdr;
2682                 struct ctl_ooa_entry *entries;
2683                 uint32_t cur_fill_num;
2684
2685                 ooa_hdr = (struct ctl_ooa *)addr;
2686
2687                 if ((ooa_hdr->alloc_len == 0)
2688                  || (ooa_hdr->alloc_num == 0)) {
2689                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2690                                "must be non-zero\n", __func__,
2691                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2692                         retval = EINVAL;
2693                         break;
2694                 }
2695
2696                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2697                     sizeof(struct ctl_ooa_entry))) {
2698                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2699                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2700                                __func__, ooa_hdr->alloc_len,
2701                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2702                         retval = EINVAL;
2703                         break;
2704                 }
2705
2706                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2707                 if (entries == NULL) {
2708                         printf("%s: could not allocate %d bytes for OOA "
2709                                "dump\n", __func__, ooa_hdr->alloc_len);
2710                         retval = ENOMEM;
2711                         break;
2712                 }
2713
2714                 mtx_lock(&softc->ctl_lock);
2715                 if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 &&
2716                     (ooa_hdr->lun_num >= CTL_MAX_LUNS ||
2717                      softc->ctl_luns[ooa_hdr->lun_num] == NULL)) {
2718                         mtx_unlock(&softc->ctl_lock);
2719                         free(entries, M_CTL);
2720                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2721                                __func__, (uintmax_t)ooa_hdr->lun_num);
2722                         retval = EINVAL;
2723                         break;
2724                 }
2725
2726                 cur_fill_num = 0;
2727
2728                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2729                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2730                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2731                                     ooa_hdr, entries);
2732                         }
2733                 } else {
2734                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2735                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2736                             entries);
2737                 }
2738                 mtx_unlock(&softc->ctl_lock);
2739
2740                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2741                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2742                         sizeof(struct ctl_ooa_entry);
2743                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2744                 if (retval != 0) {
2745                         printf("%s: error copying out %d bytes for OOA dump\n",
2746                                __func__, ooa_hdr->fill_len);
2747                 }
2748
2749                 getbinuptime(&ooa_hdr->cur_bt);
2750
2751                 if (cur_fill_num > ooa_hdr->alloc_num) {
2752                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2753                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2754                 } else {
2755                         ooa_hdr->dropped_num = 0;
2756                         ooa_hdr->status = CTL_OOA_OK;
2757                 }
2758
2759                 free(entries, M_CTL);
2760                 break;
2761         }
2762         case CTL_DELAY_IO: {
2763                 struct ctl_io_delay_info *delay_info;
2764
2765                 delay_info = (struct ctl_io_delay_info *)addr;
2766
2767 #ifdef CTL_IO_DELAY
2768                 mtx_lock(&softc->ctl_lock);
2769                 if (delay_info->lun_id >= CTL_MAX_LUNS ||
2770                     (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) {
2771                         mtx_unlock(&softc->ctl_lock);
2772                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2773                         break;
2774                 }
2775                 mtx_lock(&lun->lun_lock);
2776                 mtx_unlock(&softc->ctl_lock);
2777                 delay_info->status = CTL_DELAY_STATUS_OK;
2778                 switch (delay_info->delay_type) {
2779                 case CTL_DELAY_TYPE_CONT:
2780                 case CTL_DELAY_TYPE_ONESHOT:
2781                         break;
2782                 default:
2783                         delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE;
2784                         break;
2785                 }
2786                 switch (delay_info->delay_loc) {
2787                 case CTL_DELAY_LOC_DATAMOVE:
2788                         lun->delay_info.datamove_type = delay_info->delay_type;
2789                         lun->delay_info.datamove_delay = delay_info->delay_secs;
2790                         break;
2791                 case CTL_DELAY_LOC_DONE:
2792                         lun->delay_info.done_type = delay_info->delay_type;
2793                         lun->delay_info.done_delay = delay_info->delay_secs;
2794                         break;
2795                 default:
2796                         delay_info->status = CTL_DELAY_STATUS_INVALID_LOC;
2797                         break;
2798                 }
2799                 mtx_unlock(&lun->lun_lock);
2800 #else
2801                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2802 #endif /* CTL_IO_DELAY */
2803                 break;
2804         }
2805 #ifdef CTL_LEGACY_STATS
2806         case CTL_GETSTATS: {
2807                 struct ctl_stats *stats = (struct ctl_stats *)addr;
2808                 int i;
2809
2810                 /*
2811                  * XXX KDM no locking here.  If the LUN list changes,
2812                  * things can blow up.
2813                  */
2814                 i = 0;
2815                 stats->status = CTL_SS_OK;
2816                 stats->fill_len = 0;
2817                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2818                         if (stats->fill_len + sizeof(lun->legacy_stats) >
2819                             stats->alloc_len) {
2820                                 stats->status = CTL_SS_NEED_MORE_SPACE;
2821                                 break;
2822                         }
2823                         retval = copyout(&lun->legacy_stats, &stats->lun_stats[i++],
2824                                          sizeof(lun->legacy_stats));
2825                         if (retval != 0)
2826                                 break;
2827                         stats->fill_len += sizeof(lun->legacy_stats);
2828                 }
2829                 stats->num_luns = softc->num_luns;
2830                 stats->flags = CTL_STATS_FLAG_NONE;
2831 #ifdef CTL_TIME_IO
2832                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
2833 #endif
2834                 getnanouptime(&stats->timestamp);
2835                 break;
2836         }
2837 #endif /* CTL_LEGACY_STATS */
2838         case CTL_ERROR_INJECT: {
2839                 struct ctl_error_desc *err_desc, *new_err_desc;
2840
2841                 err_desc = (struct ctl_error_desc *)addr;
2842
2843                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2844                                       M_WAITOK | M_ZERO);
2845                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2846
2847                 mtx_lock(&softc->ctl_lock);
2848                 if (err_desc->lun_id >= CTL_MAX_LUNS ||
2849                     (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) {
2850                         mtx_unlock(&softc->ctl_lock);
2851                         free(new_err_desc, M_CTL);
2852                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2853                                __func__, (uintmax_t)err_desc->lun_id);
2854                         retval = EINVAL;
2855                         break;
2856                 }
2857                 mtx_lock(&lun->lun_lock);
2858                 mtx_unlock(&softc->ctl_lock);
2859
2860                 /*
2861                  * We could do some checking here to verify the validity
2862                  * of the request, but given the complexity of error
2863                  * injection requests, the checking logic would be fairly
2864                  * complex.
2865                  *
2866                  * For now, if the request is invalid, it just won't get
2867                  * executed and might get deleted.
2868                  */
2869                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2870
2871                 /*
2872                  * XXX KDM check to make sure the serial number is unique,
2873                  * in case we somehow manage to wrap.  That shouldn't
2874                  * happen for a very long time, but it's the right thing to
2875                  * do.
2876                  */
2877                 new_err_desc->serial = lun->error_serial;
2878                 err_desc->serial = lun->error_serial;
2879                 lun->error_serial++;
2880
2881                 mtx_unlock(&lun->lun_lock);
2882                 break;
2883         }
2884         case CTL_ERROR_INJECT_DELETE: {
2885                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2886                 int delete_done;
2887
2888                 delete_desc = (struct ctl_error_desc *)addr;
2889                 delete_done = 0;
2890
2891                 mtx_lock(&softc->ctl_lock);
2892                 if (delete_desc->lun_id >= CTL_MAX_LUNS ||
2893                     (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) {
2894                         mtx_unlock(&softc->ctl_lock);
2895                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2896                                __func__, (uintmax_t)delete_desc->lun_id);
2897                         retval = EINVAL;
2898                         break;
2899                 }
2900                 mtx_lock(&lun->lun_lock);
2901                 mtx_unlock(&softc->ctl_lock);
2902                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2903                         if (desc->serial != delete_desc->serial)
2904                                 continue;
2905
2906                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2907                                       links);
2908                         free(desc, M_CTL);
2909                         delete_done = 1;
2910                 }
2911                 mtx_unlock(&lun->lun_lock);
2912                 if (delete_done == 0) {
2913                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2914                                "error serial %ju on LUN %u\n", __func__,
2915                                delete_desc->serial, delete_desc->lun_id);
2916                         retval = EINVAL;
2917                         break;
2918                 }
2919                 break;
2920         }
2921         case CTL_DUMP_STRUCTS: {
2922                 int j, k;
2923                 struct ctl_port *port;
2924                 struct ctl_frontend *fe;
2925
2926                 mtx_lock(&softc->ctl_lock);
2927                 printf("CTL Persistent Reservation information start:\n");
2928                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2929                         mtx_lock(&lun->lun_lock);
2930                         if ((lun->flags & CTL_LUN_DISABLED) != 0) {
2931                                 mtx_unlock(&lun->lun_lock);
2932                                 continue;
2933                         }
2934
2935                         for (j = 0; j < CTL_MAX_PORTS; j++) {
2936                                 if (lun->pr_keys[j] == NULL)
2937                                         continue;
2938                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2939                                         if (lun->pr_keys[j][k] == 0)
2940                                                 continue;
2941                                         printf("  LUN %ju port %d iid %d key "
2942                                                "%#jx\n", lun->lun, j, k,
2943                                                (uintmax_t)lun->pr_keys[j][k]);
2944                                 }
2945                         }
2946                         mtx_unlock(&lun->lun_lock);
2947                 }
2948                 printf("CTL Persistent Reservation information end\n");
2949                 printf("CTL Ports:\n");
2950                 STAILQ_FOREACH(port, &softc->port_list, links) {
2951                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2952                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2953                                port->frontend->name, port->port_type,
2954                                port->physical_port, port->virtual_port,
2955                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2956                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2957                                 if (port->wwpn_iid[j].in_use == 0 &&
2958                                     port->wwpn_iid[j].wwpn == 0 &&
2959                                     port->wwpn_iid[j].name == NULL)
2960                                         continue;
2961
2962                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2963                                     j, port->wwpn_iid[j].in_use,
2964                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2965                                     port->wwpn_iid[j].name);
2966                         }
2967                 }
2968                 printf("CTL Port information end\n");
2969                 mtx_unlock(&softc->ctl_lock);
2970                 /*
2971                  * XXX KDM calling this without a lock.  We'd likely want
2972                  * to drop the lock before calling the frontend's dump
2973                  * routine anyway.
2974                  */
2975                 printf("CTL Frontends:\n");
2976                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2977                         printf("  Frontend '%s'\n", fe->name);
2978                         if (fe->fe_dump != NULL)
2979                                 fe->fe_dump();
2980                 }
2981                 printf("CTL Frontend information end\n");
2982                 break;
2983         }
2984         case CTL_LUN_REQ: {
2985                 struct ctl_lun_req *lun_req;
2986                 struct ctl_backend_driver *backend;
2987
2988                 lun_req = (struct ctl_lun_req *)addr;
2989
2990                 backend = ctl_backend_find(lun_req->backend);
2991                 if (backend == NULL) {
2992                         lun_req->status = CTL_LUN_ERROR;
2993                         snprintf(lun_req->error_str,
2994                                  sizeof(lun_req->error_str),
2995                                  "Backend \"%s\" not found.",
2996                                  lun_req->backend);
2997                         break;
2998                 }
2999                 if (lun_req->num_be_args > 0) {
3000                         lun_req->kern_be_args = ctl_copyin_args(
3001                                 lun_req->num_be_args,
3002                                 lun_req->be_args,
3003                                 lun_req->error_str,
3004                                 sizeof(lun_req->error_str));
3005                         if (lun_req->kern_be_args == NULL) {
3006                                 lun_req->status = CTL_LUN_ERROR;
3007                                 break;
3008                         }
3009                 }
3010
3011                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3012
3013                 if (lun_req->num_be_args > 0) {
3014                         ctl_copyout_args(lun_req->num_be_args,
3015                                       lun_req->kern_be_args);
3016                         ctl_free_args(lun_req->num_be_args,
3017                                       lun_req->kern_be_args);
3018                 }
3019                 break;
3020         }
3021         case CTL_LUN_LIST: {
3022                 struct sbuf *sb;
3023                 struct ctl_lun_list *list;
3024                 struct ctl_option *opt;
3025
3026                 list = (struct ctl_lun_list *)addr;
3027
3028                 /*
3029                  * Allocate a fixed length sbuf here, based on the length
3030                  * of the user's buffer.  We could allocate an auto-extending
3031                  * buffer, and then tell the user how much larger our
3032                  * amount of data is than his buffer, but that presents
3033                  * some problems:
3034                  *
3035                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3036                  *     we can't hold a lock while calling them with an
3037                  *     auto-extending buffer.
3038                  *
3039                  * 2.  There is not currently a LUN reference counting
3040                  *     mechanism, outside of outstanding transactions on
3041                  *     the LUN's OOA queue.  So a LUN could go away on us
3042                  *     while we're getting the LUN number, backend-specific
3043                  *     information, etc.  Thus, given the way things
3044                  *     currently work, we need to hold the CTL lock while
3045                  *     grabbing LUN information.
3046                  *
3047                  * So, from the user's standpoint, the best thing to do is
3048                  * allocate what he thinks is a reasonable buffer length,
3049                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3050                  * double the buffer length and try again.  (And repeat
3051                  * that until he succeeds.)
3052                  */
3053                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3054                 if (sb == NULL) {
3055                         list->status = CTL_LUN_LIST_ERROR;
3056                         snprintf(list->error_str, sizeof(list->error_str),
3057                                  "Unable to allocate %d bytes for LUN list",
3058                                  list->alloc_len);
3059                         break;
3060                 }
3061
3062                 sbuf_printf(sb, "<ctllunlist>\n");
3063
3064                 mtx_lock(&softc->ctl_lock);
3065                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3066                         mtx_lock(&lun->lun_lock);
3067                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3068                                              (uintmax_t)lun->lun);
3069
3070                         /*
3071                          * Bail out as soon as we see that we've overfilled
3072                          * the buffer.
3073                          */
3074                         if (retval != 0)
3075                                 break;
3076
3077                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3078                                              "</backend_type>\n",
3079                                              (lun->backend == NULL) ?  "none" :
3080                                              lun->backend->name);
3081
3082                         if (retval != 0)
3083                                 break;
3084
3085                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3086                                              lun->be_lun->lun_type);
3087
3088                         if (retval != 0)
3089                                 break;
3090
3091                         if (lun->backend == NULL) {
3092                                 retval = sbuf_printf(sb, "</lun>\n");
3093                                 if (retval != 0)
3094                                         break;
3095                                 continue;
3096                         }
3097
3098                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3099                                              (lun->be_lun->maxlba > 0) ?
3100                                              lun->be_lun->maxlba + 1 : 0);
3101
3102                         if (retval != 0)
3103                                 break;
3104
3105                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3106                                              lun->be_lun->blocksize);
3107
3108                         if (retval != 0)
3109                                 break;
3110
3111                         retval = sbuf_printf(sb, "\t<serial_number>");
3112
3113                         if (retval != 0)
3114                                 break;
3115
3116                         retval = ctl_sbuf_printf_esc(sb,
3117                             lun->be_lun->serial_num,
3118                             sizeof(lun->be_lun->serial_num));
3119
3120                         if (retval != 0)
3121                                 break;
3122
3123                         retval = sbuf_printf(sb, "</serial_number>\n");
3124
3125                         if (retval != 0)
3126                                 break;
3127
3128                         retval = sbuf_printf(sb, "\t<device_id>");
3129
3130                         if (retval != 0)
3131                                 break;
3132
3133                         retval = ctl_sbuf_printf_esc(sb,
3134                             lun->be_lun->device_id,
3135                             sizeof(lun->be_lun->device_id));
3136
3137                         if (retval != 0)
3138                                 break;
3139
3140                         retval = sbuf_printf(sb, "</device_id>\n");
3141
3142                         if (retval != 0)
3143                                 break;
3144
3145                         if (lun->backend->lun_info != NULL) {
3146                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3147                                 if (retval != 0)
3148                                         break;
3149                         }
3150                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3151                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3152                                     opt->name, opt->value, opt->name);
3153                                 if (retval != 0)
3154                                         break;
3155                         }
3156
3157                         retval = sbuf_printf(sb, "</lun>\n");
3158
3159                         if (retval != 0)
3160                                 break;
3161                         mtx_unlock(&lun->lun_lock);
3162                 }
3163                 if (lun != NULL)
3164                         mtx_unlock(&lun->lun_lock);
3165                 mtx_unlock(&softc->ctl_lock);
3166
3167                 if ((retval != 0)
3168                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3169                         retval = 0;
3170                         sbuf_delete(sb);
3171                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3172                         snprintf(list->error_str, sizeof(list->error_str),
3173                                  "Out of space, %d bytes is too small",
3174                                  list->alloc_len);
3175                         break;
3176                 }
3177
3178                 sbuf_finish(sb);
3179
3180                 retval = copyout(sbuf_data(sb), list->lun_xml,
3181                                  sbuf_len(sb) + 1);
3182
3183                 list->fill_len = sbuf_len(sb) + 1;
3184                 list->status = CTL_LUN_LIST_OK;
3185                 sbuf_delete(sb);
3186                 break;
3187         }
3188         case CTL_ISCSI: {
3189                 struct ctl_iscsi *ci;
3190                 struct ctl_frontend *fe;
3191
3192                 ci = (struct ctl_iscsi *)addr;
3193
3194                 fe = ctl_frontend_find("iscsi");
3195                 if (fe == NULL) {
3196                         ci->status = CTL_ISCSI_ERROR;
3197                         snprintf(ci->error_str, sizeof(ci->error_str),
3198                             "Frontend \"iscsi\" not found.");
3199                         break;
3200                 }
3201
3202                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3203                 break;
3204         }
3205         case CTL_PORT_REQ: {
3206                 struct ctl_req *req;
3207                 struct ctl_frontend *fe;
3208
3209                 req = (struct ctl_req *)addr;
3210
3211                 fe = ctl_frontend_find(req->driver);
3212                 if (fe == NULL) {
3213                         req->status = CTL_LUN_ERROR;
3214                         snprintf(req->error_str, sizeof(req->error_str),
3215                             "Frontend \"%s\" not found.", req->driver);
3216                         break;
3217                 }
3218                 if (req->num_args > 0) {
3219                         req->kern_args = ctl_copyin_args(req->num_args,
3220                             req->args, req->error_str, sizeof(req->error_str));
3221                         if (req->kern_args == NULL) {
3222                                 req->status = CTL_LUN_ERROR;
3223                                 break;
3224                         }
3225                 }
3226
3227                 if (fe->ioctl)
3228                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3229                 else
3230                         retval = ENODEV;
3231
3232                 if (req->num_args > 0) {
3233                         ctl_copyout_args(req->num_args, req->kern_args);
3234                         ctl_free_args(req->num_args, req->kern_args);
3235                 }
3236                 break;
3237         }
3238         case CTL_PORT_LIST: {
3239                 struct sbuf *sb;
3240                 struct ctl_port *port;
3241                 struct ctl_lun_list *list;
3242                 struct ctl_option *opt;
3243                 int j;
3244                 uint32_t plun;
3245
3246                 list = (struct ctl_lun_list *)addr;
3247
3248                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3249                 if (sb == NULL) {
3250                         list->status = CTL_LUN_LIST_ERROR;
3251                         snprintf(list->error_str, sizeof(list->error_str),
3252                                  "Unable to allocate %d bytes for LUN list",
3253                                  list->alloc_len);
3254                         break;
3255                 }
3256
3257                 sbuf_printf(sb, "<ctlportlist>\n");
3258
3259                 mtx_lock(&softc->ctl_lock);
3260                 STAILQ_FOREACH(port, &softc->port_list, links) {
3261                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3262                                              (uintmax_t)port->targ_port);
3263
3264                         /*
3265                          * Bail out as soon as we see that we've overfilled
3266                          * the buffer.
3267                          */
3268                         if (retval != 0)
3269                                 break;
3270
3271                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3272                             "</frontend_type>\n", port->frontend->name);
3273                         if (retval != 0)
3274                                 break;
3275
3276                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3277                                              port->port_type);
3278                         if (retval != 0)
3279                                 break;
3280
3281                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3282                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3283                         if (retval != 0)
3284                                 break;
3285
3286                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3287                             port->port_name);
3288                         if (retval != 0)
3289                                 break;
3290
3291                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3292                             port->physical_port);
3293                         if (retval != 0)
3294                                 break;
3295
3296                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3297                             port->virtual_port);
3298                         if (retval != 0)
3299                                 break;
3300
3301                         if (port->target_devid != NULL) {
3302                                 sbuf_printf(sb, "\t<target>");
3303                                 ctl_id_sbuf(port->target_devid, sb);
3304                                 sbuf_printf(sb, "</target>\n");
3305                         }
3306
3307                         if (port->port_devid != NULL) {
3308                                 sbuf_printf(sb, "\t<port>");
3309                                 ctl_id_sbuf(port->port_devid, sb);
3310                                 sbuf_printf(sb, "</port>\n");
3311                         }
3312
3313                         if (port->port_info != NULL) {
3314                                 retval = port->port_info(port->onoff_arg, sb);
3315                                 if (retval != 0)
3316                                         break;
3317                         }
3318                         STAILQ_FOREACH(opt, &port->options, links) {
3319                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3320                                     opt->name, opt->value, opt->name);
3321                                 if (retval != 0)
3322                                         break;
3323                         }
3324
3325                         if (port->lun_map != NULL) {
3326                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3327                                 for (j = 0; j < port->lun_map_size; j++) {
3328                                         plun = ctl_lun_map_from_port(port, j);
3329                                         if (plun == UINT32_MAX)
3330                                                 continue;
3331                                         sbuf_printf(sb,
3332                                             "\t<lun id=\"%u\">%u</lun>\n",
3333                                             j, plun);
3334                                 }
3335                         }
3336
3337                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3338                                 if (port->wwpn_iid[j].in_use == 0 ||
3339                                     (port->wwpn_iid[j].wwpn == 0 &&
3340                                      port->wwpn_iid[j].name == NULL))
3341                                         continue;
3342
3343                                 if (port->wwpn_iid[j].name != NULL)
3344                                         retval = sbuf_printf(sb,
3345                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3346                                             j, port->wwpn_iid[j].name);
3347                                 else
3348                                         retval = sbuf_printf(sb,
3349                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3350                                             j, port->wwpn_iid[j].wwpn);
3351                                 if (retval != 0)
3352                                         break;
3353                         }
3354                         if (retval != 0)
3355                                 break;
3356
3357                         retval = sbuf_printf(sb, "</targ_port>\n");
3358                         if (retval != 0)
3359                                 break;
3360                 }
3361                 mtx_unlock(&softc->ctl_lock);
3362
3363                 if ((retval != 0)
3364                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3365                         retval = 0;
3366                         sbuf_delete(sb);
3367                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3368                         snprintf(list->error_str, sizeof(list->error_str),
3369                                  "Out of space, %d bytes is too small",
3370                                  list->alloc_len);
3371                         break;
3372                 }
3373
3374                 sbuf_finish(sb);
3375
3376                 retval = copyout(sbuf_data(sb), list->lun_xml,
3377                                  sbuf_len(sb) + 1);
3378
3379                 list->fill_len = sbuf_len(sb) + 1;
3380                 list->status = CTL_LUN_LIST_OK;
3381                 sbuf_delete(sb);
3382                 break;
3383         }
3384         case CTL_LUN_MAP: {
3385                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3386                 struct ctl_port *port;
3387
3388                 mtx_lock(&softc->ctl_lock);
3389                 if (lm->port < softc->port_min ||
3390                     lm->port >= softc->port_max ||
3391                     (port = softc->ctl_ports[lm->port]) == NULL) {
3392                         mtx_unlock(&softc->ctl_lock);
3393                         return (ENXIO);
3394                 }
3395                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3396                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3397                                 if (ctl_lun_map_to_port(port, lun->lun) ==
3398                                     UINT32_MAX)
3399                                         continue;
3400                                 mtx_lock(&lun->lun_lock);
3401                                 ctl_est_ua_port(lun, lm->port, -1,
3402                                     CTL_UA_LUN_CHANGE);
3403                                 mtx_unlock(&lun->lun_lock);
3404                         }
3405                 }
3406                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3407                 if (lm->plun != UINT32_MAX) {
3408                         if (lm->lun == UINT32_MAX)
3409                                 retval = ctl_lun_map_unset(port, lm->plun);
3410                         else if (lm->lun < CTL_MAX_LUNS &&
3411                             softc->ctl_luns[lm->lun] != NULL)
3412                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3413                         else
3414                                 return (ENXIO);
3415                 } else {
3416                         if (lm->lun == UINT32_MAX)
3417                                 retval = ctl_lun_map_deinit(port);
3418                         else
3419                                 retval = ctl_lun_map_init(port);
3420                 }
3421                 if (port->status & CTL_PORT_STATUS_ONLINE)
3422                         ctl_isc_announce_port(port);
3423                 break;
3424         }
3425         case CTL_GET_LUN_STATS: {
3426                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3427                 int i;
3428
3429                 /*
3430                  * XXX KDM no locking here.  If the LUN list changes,
3431                  * things can blow up.
3432                  */
3433                 i = 0;
3434                 stats->status = CTL_SS_OK;
3435                 stats->fill_len = 0;
3436                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3437                         if (lun->lun < stats->first_item)
3438                                 continue;
3439                         if (stats->fill_len + sizeof(lun->stats) >
3440                             stats->alloc_len) {
3441                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3442                                 break;
3443                         }
3444                         retval = copyout(&lun->stats, &stats->stats[i++],
3445                                          sizeof(lun->stats));
3446                         if (retval != 0)
3447                                 break;
3448                         stats->fill_len += sizeof(lun->stats);
3449                 }
3450                 stats->num_items = softc->num_luns;
3451                 stats->flags = CTL_STATS_FLAG_NONE;
3452 #ifdef CTL_TIME_IO
3453                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3454 #endif
3455                 getnanouptime(&stats->timestamp);
3456                 break;
3457         }
3458         case CTL_GET_PORT_STATS: {
3459                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3460                 int i;
3461
3462                 /*
3463                  * XXX KDM no locking here.  If the LUN list changes,
3464                  * things can blow up.
3465                  */
3466                 i = 0;
3467                 stats->status = CTL_SS_OK;
3468                 stats->fill_len = 0;
3469                 STAILQ_FOREACH(port, &softc->port_list, links) {
3470                         if (port->targ_port < stats->first_item)
3471                                 continue;
3472                         if (stats->fill_len + sizeof(port->stats) >
3473                             stats->alloc_len) {
3474                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3475                                 break;
3476                         }
3477                         retval = copyout(&port->stats, &stats->stats[i++],
3478                                          sizeof(port->stats));
3479                         if (retval != 0)
3480                                 break;
3481                         stats->fill_len += sizeof(port->stats);
3482                 }
3483                 stats->num_items = softc->num_ports;
3484                 stats->flags = CTL_STATS_FLAG_NONE;
3485 #ifdef CTL_TIME_IO
3486                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3487 #endif
3488                 getnanouptime(&stats->timestamp);
3489                 break;
3490         }
3491         default: {
3492                 /* XXX KDM should we fix this? */
3493 #if 0
3494                 struct ctl_backend_driver *backend;
3495                 unsigned int type;
3496                 int found;
3497
3498                 found = 0;
3499
3500                 /*
3501                  * We encode the backend type as the ioctl type for backend
3502                  * ioctls.  So parse it out here, and then search for a
3503                  * backend of this type.
3504                  */
3505                 type = _IOC_TYPE(cmd);
3506
3507                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3508                         if (backend->type == type) {
3509                                 found = 1;
3510                                 break;
3511                         }
3512                 }
3513                 if (found == 0) {
3514                         printf("ctl: unknown ioctl command %#lx or backend "
3515                                "%d\n", cmd, type);
3516                         retval = EINVAL;
3517                         break;
3518                 }
3519                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3520 #endif
3521                 retval = ENOTTY;
3522                 break;
3523         }
3524         }
3525         return (retval);
3526 }
3527
3528 uint32_t
3529 ctl_get_initindex(struct ctl_nexus *nexus)
3530 {
3531         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3532 }
3533
3534 int
3535 ctl_lun_map_init(struct ctl_port *port)
3536 {
3537         struct ctl_softc *softc = port->ctl_softc;
3538         struct ctl_lun *lun;
3539         int size = ctl_lun_map_size;
3540         uint32_t i;
3541
3542         if (port->lun_map == NULL || port->lun_map_size < size) {
3543                 port->lun_map_size = 0;
3544                 free(port->lun_map, M_CTL);
3545                 port->lun_map = malloc(size * sizeof(uint32_t),
3546                     M_CTL, M_NOWAIT);
3547         }
3548         if (port->lun_map == NULL)
3549                 return (ENOMEM);
3550         for (i = 0; i < size; i++)
3551                 port->lun_map[i] = UINT32_MAX;
3552         port->lun_map_size = size;
3553         if (port->status & CTL_PORT_STATUS_ONLINE) {
3554                 if (port->lun_disable != NULL) {
3555                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3556                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3557                 }
3558                 ctl_isc_announce_port(port);
3559         }
3560         return (0);
3561 }
3562
3563 int
3564 ctl_lun_map_deinit(struct ctl_port *port)
3565 {
3566         struct ctl_softc *softc = port->ctl_softc;
3567         struct ctl_lun *lun;
3568
3569         if (port->lun_map == NULL)
3570                 return (0);
3571         port->lun_map_size = 0;
3572         free(port->lun_map, M_CTL);
3573         port->lun_map = NULL;
3574         if (port->status & CTL_PORT_STATUS_ONLINE) {
3575                 if (port->lun_enable != NULL) {
3576                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3577                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3578                 }
3579                 ctl_isc_announce_port(port);
3580         }
3581         return (0);
3582 }
3583
3584 int
3585 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3586 {
3587         int status;
3588         uint32_t old;
3589
3590         if (port->lun_map == NULL) {
3591                 status = ctl_lun_map_init(port);
3592                 if (status != 0)
3593                         return (status);
3594         }
3595         if (plun >= port->lun_map_size)
3596                 return (EINVAL);
3597         old = port->lun_map[plun];
3598         port->lun_map[plun] = glun;
3599         if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) {
3600                 if (port->lun_enable != NULL)
3601                         port->lun_enable(port->targ_lun_arg, plun);
3602                 ctl_isc_announce_port(port);
3603         }
3604         return (0);
3605 }
3606
3607 int
3608 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3609 {
3610         uint32_t old;
3611
3612         if (port->lun_map == NULL || plun >= port->lun_map_size)
3613                 return (0);
3614         old = port->lun_map[plun];
3615         port->lun_map[plun] = UINT32_MAX;
3616         if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) {
3617                 if (port->lun_disable != NULL)
3618                         port->lun_disable(port->targ_lun_arg, plun);
3619                 ctl_isc_announce_port(port);
3620         }
3621         return (0);
3622 }
3623
3624 uint32_t
3625 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3626 {
3627
3628         if (port == NULL)
3629                 return (UINT32_MAX);
3630         if (port->lun_map == NULL)
3631                 return (lun_id);
3632         if (lun_id > port->lun_map_size)
3633                 return (UINT32_MAX);
3634         return (port->lun_map[lun_id]);
3635 }
3636
3637 uint32_t
3638 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3639 {
3640         uint32_t i;
3641
3642         if (port == NULL)
3643                 return (UINT32_MAX);
3644         if (port->lun_map == NULL)
3645                 return (lun_id);
3646         for (i = 0; i < port->lun_map_size; i++) {
3647                 if (port->lun_map[i] == lun_id)
3648                         return (i);
3649         }
3650         return (UINT32_MAX);
3651 }
3652
3653 uint32_t
3654 ctl_decode_lun(uint64_t encoded)
3655 {
3656         uint8_t lun[8];
3657         uint32_t result = 0xffffffff;
3658
3659         be64enc(lun, encoded);
3660         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3661         case RPL_LUNDATA_ATYP_PERIPH:
3662                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3663                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3664                         result = lun[1];
3665                 break;
3666         case RPL_LUNDATA_ATYP_FLAT:
3667                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3668                     lun[6] == 0 && lun[7] == 0)
3669                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3670                 break;
3671         case RPL_LUNDATA_ATYP_EXTLUN:
3672                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3673                 case 0x02:
3674                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3675                         case 0x00:
3676                                 result = lun[1];
3677                                 break;
3678                         case 0x10:
3679                                 result = (lun[1] << 16) + (lun[2] << 8) +
3680                                     lun[3];
3681                                 break;
3682                         case 0x20:
3683                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3684                                         result = (lun[2] << 24) +
3685                                             (lun[3] << 16) + (lun[4] << 8) +
3686                                             lun[5];
3687                                 break;
3688                         }
3689                         break;
3690                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3691                         result = 0xffffffff;
3692                         break;
3693                 }
3694                 break;
3695         }
3696         return (result);
3697 }
3698
3699 uint64_t
3700 ctl_encode_lun(uint32_t decoded)
3701 {
3702         uint64_t l = decoded;
3703
3704         if (l <= 0xff)
3705                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3706         if (l <= 0x3fff)
3707                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3708         if (l <= 0xffffff)
3709                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3710                     (l << 32));
3711         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3712 }
3713
3714 int
3715 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3716 {
3717         int i;
3718
3719         for (i = first; i < last; i++) {
3720                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3721                         return (i);
3722         }
3723         return (-1);
3724 }
3725
3726 int
3727 ctl_set_mask(uint32_t *mask, uint32_t bit)
3728 {
3729         uint32_t chunk, piece;
3730
3731         chunk = bit >> 5;
3732         piece = bit % (sizeof(uint32_t) * 8);
3733
3734         if ((mask[chunk] & (1 << piece)) != 0)
3735                 return (-1);
3736         else
3737                 mask[chunk] |= (1 << piece);
3738
3739         return (0);
3740 }
3741
3742 int
3743 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3744 {
3745         uint32_t chunk, piece;
3746
3747         chunk = bit >> 5;
3748         piece = bit % (sizeof(uint32_t) * 8);
3749
3750         if ((mask[chunk] & (1 << piece)) == 0)
3751                 return (-1);
3752         else
3753                 mask[chunk] &= ~(1 << piece);
3754
3755         return (0);
3756 }
3757
3758 int
3759 ctl_is_set(uint32_t *mask, uint32_t bit)
3760 {
3761         uint32_t chunk, piece;
3762
3763         chunk = bit >> 5;
3764         piece = bit % (sizeof(uint32_t) * 8);
3765
3766         if ((mask[chunk] & (1 << piece)) == 0)
3767                 return (0);
3768         else
3769                 return (1);
3770 }
3771
3772 static uint64_t
3773 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3774 {
3775         uint64_t *t;
3776
3777         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3778         if (t == NULL)
3779                 return (0);
3780         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3781 }
3782
3783 static void
3784 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3785 {
3786         uint64_t *t;
3787
3788         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3789         if (t == NULL)
3790                 return;
3791         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3792 }
3793
3794 static void
3795 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3796 {
3797         uint64_t *p;
3798         u_int i;
3799
3800         i = residx/CTL_MAX_INIT_PER_PORT;
3801         if (lun->pr_keys[i] != NULL)
3802                 return;
3803         mtx_unlock(&lun->lun_lock);
3804         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3805             M_WAITOK | M_ZERO);
3806         mtx_lock(&lun->lun_lock);
3807         if (lun->pr_keys[i] == NULL)
3808                 lun->pr_keys[i] = p;
3809         else
3810                 free(p, M_CTL);
3811 }
3812
3813 static void
3814 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3815 {
3816         uint64_t *t;
3817
3818         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3819         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3820         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3821 }
3822
3823 /*
3824  * ctl_softc, pool_name, total_ctl_io are passed in.
3825  * npool is passed out.
3826  */
3827 int
3828 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3829                 uint32_t total_ctl_io, void **npool)
3830 {
3831         struct ctl_io_pool *pool;
3832
3833         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3834                                             M_NOWAIT | M_ZERO);
3835         if (pool == NULL)
3836                 return (ENOMEM);
3837
3838         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3839         pool->ctl_softc = ctl_softc;
3840 #ifdef IO_POOLS
3841         pool->zone = uma_zsecond_create(pool->name, NULL,
3842             NULL, NULL, NULL, ctl_softc->io_zone);
3843         /* uma_prealloc(pool->zone, total_ctl_io); */
3844 #else
3845         pool->zone = ctl_softc->io_zone;
3846 #endif
3847
3848         *npool = pool;
3849         return (0);
3850 }
3851
3852 void
3853 ctl_pool_free(struct ctl_io_pool *pool)
3854 {
3855
3856         if (pool == NULL)
3857                 return;
3858
3859 #ifdef IO_POOLS
3860         uma_zdestroy(pool->zone);
3861 #endif
3862         free(pool, M_CTL);
3863 }
3864
3865 union ctl_io *
3866 ctl_alloc_io(void *pool_ref)
3867 {
3868         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3869         union ctl_io *io;
3870
3871         io = uma_zalloc(pool->zone, M_WAITOK);
3872         if (io != NULL) {
3873                 io->io_hdr.pool = pool_ref;
3874                 CTL_SOFTC(io) = pool->ctl_softc;
3875         }
3876         return (io);
3877 }
3878
3879 union ctl_io *
3880 ctl_alloc_io_nowait(void *pool_ref)
3881 {
3882         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3883         union ctl_io *io;
3884
3885         io = uma_zalloc(pool->zone, M_NOWAIT);
3886         if (io != NULL) {
3887                 io->io_hdr.pool = pool_ref;
3888                 CTL_SOFTC(io) = pool->ctl_softc;
3889         }
3890         return (io);
3891 }
3892
3893 void
3894 ctl_free_io(union ctl_io *io)
3895 {
3896         struct ctl_io_pool *pool;
3897
3898         if (io == NULL)
3899                 return;
3900
3901         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3902         uma_zfree(pool->zone, io);
3903 }
3904
3905 void
3906 ctl_zero_io(union ctl_io *io)
3907 {
3908         struct ctl_io_pool *pool;
3909
3910         if (io == NULL)
3911                 return;
3912
3913         /*
3914          * May need to preserve linked list pointers at some point too.
3915          */
3916         pool = io->io_hdr.pool;
3917         memset(io, 0, sizeof(*io));
3918         io->io_hdr.pool = pool;
3919         CTL_SOFTC(io) = pool->ctl_softc;
3920 }
3921
3922 int
3923 ctl_expand_number(const char *buf, uint64_t *num)
3924 {
3925         char *endptr;
3926         uint64_t number;
3927         unsigned shift;
3928
3929         number = strtoq(buf, &endptr, 0);
3930
3931         switch (tolower((unsigned char)*endptr)) {
3932         case 'e':
3933                 shift = 60;
3934                 break;
3935         case 'p':
3936                 shift = 50;
3937                 break;
3938         case 't':
3939                 shift = 40;
3940                 break;
3941         case 'g':
3942                 shift = 30;
3943                 break;
3944         case 'm':
3945                 shift = 20;
3946                 break;
3947         case 'k':
3948                 shift = 10;
3949                 break;
3950         case 'b':
3951         case '\0': /* No unit. */
3952                 *num = number;
3953                 return (0);
3954         default:
3955                 /* Unrecognized unit. */
3956                 return (-1);
3957         }
3958
3959         if ((number << shift) >> shift != number) {
3960                 /* Overflow */
3961                 return (-1);
3962         }
3963         *num = number << shift;
3964         return (0);
3965 }
3966
3967
3968 /*
3969  * This routine could be used in the future to load default and/or saved
3970  * mode page parameters for a particuar lun.
3971  */
3972 static int
3973 ctl_init_page_index(struct ctl_lun *lun)
3974 {
3975         int i, page_code;
3976         struct ctl_page_index *page_index;
3977         const char *value;
3978         uint64_t ival;
3979
3980         memcpy(&lun->mode_pages.index, page_index_template,
3981                sizeof(page_index_template));
3982
3983         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3984
3985                 page_index = &lun->mode_pages.index[i];
3986                 if (lun->be_lun->lun_type == T_DIRECT &&
3987                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
3988                         continue;
3989                 if (lun->be_lun->lun_type == T_PROCESSOR &&
3990                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
3991                         continue;
3992                 if (lun->be_lun->lun_type == T_CDROM &&
3993                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
3994                         continue;
3995
3996                 page_code = page_index->page_code & SMPH_PC_MASK;
3997                 switch (page_code) {
3998                 case SMS_RW_ERROR_RECOVERY_PAGE: {
3999                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4000                             ("subpage %#x for page %#x is incorrect!",
4001                             page_index->subpage, page_code));
4002                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4003                                &rw_er_page_default,
4004                                sizeof(rw_er_page_default));
4005                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4006                                &rw_er_page_changeable,
4007                                sizeof(rw_er_page_changeable));
4008                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4009                                &rw_er_page_default,
4010                                sizeof(rw_er_page_default));
4011                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4012                                &rw_er_page_default,
4013                                sizeof(rw_er_page_default));
4014                         page_index->page_data =
4015                                 (uint8_t *)lun->mode_pages.rw_er_page;
4016                         break;
4017                 }
4018                 case SMS_FORMAT_DEVICE_PAGE: {
4019                         struct scsi_format_page *format_page;
4020
4021                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4022                             ("subpage %#x for page %#x is incorrect!",
4023                             page_index->subpage, page_code));
4024
4025                         /*
4026                          * Sectors per track are set above.  Bytes per
4027                          * sector need to be set here on a per-LUN basis.
4028                          */
4029                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4030                                &format_page_default,
4031                                sizeof(format_page_default));
4032                         memcpy(&lun->mode_pages.format_page[
4033                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4034                                sizeof(format_page_changeable));
4035                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4036                                &format_page_default,
4037                                sizeof(format_page_default));
4038                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4039                                &format_page_default,
4040                                sizeof(format_page_default));
4041
4042                         format_page = &lun->mode_pages.format_page[
4043                                 CTL_PAGE_CURRENT];
4044                         scsi_ulto2b(lun->be_lun->blocksize,
4045                                     format_page->bytes_per_sector);
4046
4047                         format_page = &lun->mode_pages.format_page[
4048                                 CTL_PAGE_DEFAULT];
4049                         scsi_ulto2b(lun->be_lun->blocksize,
4050                                     format_page->bytes_per_sector);
4051
4052                         format_page = &lun->mode_pages.format_page[
4053                                 CTL_PAGE_SAVED];
4054                         scsi_ulto2b(lun->be_lun->blocksize,
4055                                     format_page->bytes_per_sector);
4056
4057                         page_index->page_data =
4058                                 (uint8_t *)lun->mode_pages.format_page;
4059                         break;
4060                 }
4061                 case SMS_RIGID_DISK_PAGE: {
4062                         struct scsi_rigid_disk_page *rigid_disk_page;
4063                         uint32_t sectors_per_cylinder;
4064                         uint64_t cylinders;
4065 #ifndef __XSCALE__
4066                         int shift;
4067 #endif /* !__XSCALE__ */
4068
4069                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4070                             ("subpage %#x for page %#x is incorrect!",
4071                             page_index->subpage, page_code));
4072
4073                         /*
4074                          * Rotation rate and sectors per track are set
4075                          * above.  We calculate the cylinders here based on
4076                          * capacity.  Due to the number of heads and
4077                          * sectors per track we're using, smaller arrays
4078                          * may turn out to have 0 cylinders.  Linux and
4079                          * FreeBSD don't pay attention to these mode pages
4080                          * to figure out capacity, but Solaris does.  It
4081                          * seems to deal with 0 cylinders just fine, and
4082                          * works out a fake geometry based on the capacity.
4083                          */
4084                         memcpy(&lun->mode_pages.rigid_disk_page[
4085                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4086                                sizeof(rigid_disk_page_default));
4087                         memcpy(&lun->mode_pages.rigid_disk_page[
4088                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4089                                sizeof(rigid_disk_page_changeable));
4090
4091                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4092                                 CTL_DEFAULT_HEADS;
4093
4094                         /*
4095                          * The divide method here will be more accurate,
4096                          * probably, but results in floating point being
4097                          * used in the kernel on i386 (__udivdi3()).  On the
4098                          * XScale, though, __udivdi3() is implemented in
4099                          * software.
4100                          *
4101                          * The shift method for cylinder calculation is
4102                          * accurate if sectors_per_cylinder is a power of
4103                          * 2.  Otherwise it might be slightly off -- you
4104                          * might have a bit of a truncation problem.
4105                          */
4106 #ifdef  __XSCALE__
4107                         cylinders = (lun->be_lun->maxlba + 1) /
4108                                 sectors_per_cylinder;
4109 #else
4110                         for (shift = 31; shift > 0; shift--) {
4111                                 if (sectors_per_cylinder & (1 << shift))
4112                                         break;
4113                         }
4114                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4115 #endif
4116
4117                         /*
4118                          * We've basically got 3 bytes, or 24 bits for the
4119                          * cylinder size in the mode page.  If we're over,
4120                          * just round down to 2^24.
4121                          */
4122                         if (cylinders > 0xffffff)
4123                                 cylinders = 0xffffff;
4124
4125                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4126                                 CTL_PAGE_DEFAULT];
4127                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4128
4129                         if ((value = ctl_get_opt(&lun->be_lun->options,
4130                             "rpm")) != NULL) {
4131                                 scsi_ulto2b(strtol(value, NULL, 0),
4132                                      rigid_disk_page->rotation_rate);
4133                         }
4134
4135                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4136                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4137                                sizeof(rigid_disk_page_default));
4138                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4139                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4140                                sizeof(rigid_disk_page_default));
4141
4142                         page_index->page_data =
4143                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4144                         break;
4145                 }
4146                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4147                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4148                             ("subpage %#x for page %#x is incorrect!",
4149                             page_index->subpage, page_code));
4150                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4151                                &verify_er_page_default,
4152                                sizeof(verify_er_page_default));
4153                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4154                                &verify_er_page_changeable,
4155                                sizeof(verify_er_page_changeable));
4156                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4157                                &verify_er_page_default,
4158                                sizeof(verify_er_page_default));
4159                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4160                                &verify_er_page_default,
4161                                sizeof(verify_er_page_default));
4162                         page_index->page_data =
4163                                 (uint8_t *)lun->mode_pages.verify_er_page;
4164                         break;
4165                 }
4166                 case SMS_CACHING_PAGE: {
4167                         struct scsi_caching_page *caching_page;
4168
4169                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4170                             ("subpage %#x for page %#x is incorrect!",
4171                             page_index->subpage, page_code));
4172                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4173                                &caching_page_default,
4174                                sizeof(caching_page_default));
4175                         memcpy(&lun->mode_pages.caching_page[
4176                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4177                                sizeof(caching_page_changeable));
4178                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4179                                &caching_page_default,
4180                                sizeof(caching_page_default));
4181                         caching_page = &lun->mode_pages.caching_page[
4182                             CTL_PAGE_SAVED];
4183                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
4184                         if (value != NULL && strcmp(value, "off") == 0)
4185                                 caching_page->flags1 &= ~SCP_WCE;
4186                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4187                         if (value != NULL && strcmp(value, "off") == 0)
4188                                 caching_page->flags1 |= SCP_RCD;
4189                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4190                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4191                                sizeof(caching_page_default));
4192                         page_index->page_data =
4193                                 (uint8_t *)lun->mode_pages.caching_page;
4194                         break;
4195                 }
4196                 case SMS_CONTROL_MODE_PAGE: {
4197                         switch (page_index->subpage) {
4198                         case SMS_SUBPAGE_PAGE_0: {
4199                                 struct scsi_control_page *control_page;
4200
4201                                 memcpy(&lun->mode_pages.control_page[
4202                                     CTL_PAGE_DEFAULT],
4203                                        &control_page_default,
4204                                        sizeof(control_page_default));
4205                                 memcpy(&lun->mode_pages.control_page[
4206                                     CTL_PAGE_CHANGEABLE],
4207                                        &control_page_changeable,
4208                                        sizeof(control_page_changeable));
4209                                 memcpy(&lun->mode_pages.control_page[
4210                                     CTL_PAGE_SAVED],
4211                                        &control_page_default,
4212                                        sizeof(control_page_default));
4213                                 control_page = &lun->mode_pages.control_page[
4214                                     CTL_PAGE_SAVED];
4215                                 value = ctl_get_opt(&lun->be_lun->options,
4216                                     "reordering");
4217                                 if (value != NULL &&
4218                                     strcmp(value, "unrestricted") == 0) {
4219                                         control_page->queue_flags &=
4220                                             ~SCP_QUEUE_ALG_MASK;
4221                                         control_page->queue_flags |=
4222                                             SCP_QUEUE_ALG_UNRESTRICTED;
4223                                 }
4224                                 memcpy(&lun->mode_pages.control_page[
4225                                     CTL_PAGE_CURRENT],
4226                                        &lun->mode_pages.control_page[
4227                                     CTL_PAGE_SAVED],
4228                                        sizeof(control_page_default));
4229                                 page_index->page_data =
4230                                     (uint8_t *)lun->mode_pages.control_page;
4231                                 break;
4232                         }
4233                         case 0x01:
4234                                 memcpy(&lun->mode_pages.control_ext_page[
4235                                     CTL_PAGE_DEFAULT],
4236                                        &control_ext_page_default,
4237                                        sizeof(control_ext_page_default));
4238                                 memcpy(&lun->mode_pages.control_ext_page[
4239                                     CTL_PAGE_CHANGEABLE],
4240                                        &control_ext_page_changeable,
4241                                        sizeof(control_ext_page_changeable));
4242                                 memcpy(&lun->mode_pages.control_ext_page[
4243                                     CTL_PAGE_SAVED],
4244                                        &control_ext_page_default,
4245                                        sizeof(control_ext_page_default));
4246                                 memcpy(&lun->mode_pages.control_ext_page[
4247                                     CTL_PAGE_CURRENT],
4248                                        &lun->mode_pages.control_ext_page[
4249                                     CTL_PAGE_SAVED],
4250                                        sizeof(control_ext_page_default));
4251                                 page_index->page_data =
4252                                     (uint8_t *)lun->mode_pages.control_ext_page;
4253                                 break;
4254                         default:
4255                                 panic("subpage %#x for page %#x is incorrect!",
4256                                       page_index->subpage, page_code);
4257                         }
4258                         break;
4259                 }
4260                 case SMS_INFO_EXCEPTIONS_PAGE: {
4261                         switch (page_index->subpage) {
4262                         case SMS_SUBPAGE_PAGE_0:
4263                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4264                                        &ie_page_default,
4265                                        sizeof(ie_page_default));
4266                                 memcpy(&lun->mode_pages.ie_page[
4267                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4268                                        sizeof(ie_page_changeable));
4269                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4270                                        &ie_page_default,
4271                                        sizeof(ie_page_default));
4272                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4273                                        &ie_page_default,
4274                                        sizeof(ie_page_default));
4275                                 page_index->page_data =
4276                                         (uint8_t *)lun->mode_pages.ie_page;
4277                                 break;
4278                         case 0x02: {
4279                                 struct ctl_logical_block_provisioning_page *page;
4280
4281                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4282                                        &lbp_page_default,
4283                                        sizeof(lbp_page_default));
4284                                 memcpy(&lun->mode_pages.lbp_page[
4285                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4286                                        sizeof(lbp_page_changeable));
4287                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4288                                        &lbp_page_default,
4289                                        sizeof(lbp_page_default));
4290                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4291                                 value = ctl_get_opt(&lun->be_lun->options,
4292                                     "avail-threshold");
4293                                 if (value != NULL &&
4294                                     ctl_expand_number(value, &ival) == 0) {
4295                                         page->descr[0].flags |= SLBPPD_ENABLED |
4296                                             SLBPPD_ARMING_DEC;
4297                                         if (lun->be_lun->blocksize)
4298                                                 ival /= lun->be_lun->blocksize;
4299                                         else
4300                                                 ival /= 512;
4301                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4302                                             page->descr[0].count);
4303                                 }
4304                                 value = ctl_get_opt(&lun->be_lun->options,
4305                                     "used-threshold");
4306                                 if (value != NULL &&
4307                                     ctl_expand_number(value, &ival) == 0) {
4308                                         page->descr[1].flags |= SLBPPD_ENABLED |
4309                                             SLBPPD_ARMING_INC;
4310                                         if (lun->be_lun->blocksize)
4311                                                 ival /= lun->be_lun->blocksize;
4312                                         else
4313                                                 ival /= 512;
4314                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4315                                             page->descr[1].count);
4316                                 }
4317                                 value = ctl_get_opt(&lun->be_lun->options,
4318                                     "pool-avail-threshold");
4319                                 if (value != NULL &&
4320                                     ctl_expand_number(value, &ival) == 0) {
4321                                         page->descr[2].flags |= SLBPPD_ENABLED |
4322                                             SLBPPD_ARMING_DEC;
4323                                         if (lun->be_lun->blocksize)
4324                                                 ival /= lun->be_lun->blocksize;
4325                                         else
4326                                                 ival /= 512;
4327                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4328                                             page->descr[2].count);
4329                                 }
4330                                 value = ctl_get_opt(&lun->be_lun->options,
4331                                     "pool-used-threshold");
4332                                 if (value != NULL &&
4333                                     ctl_expand_number(value, &ival) == 0) {
4334                                         page->descr[3].flags |= SLBPPD_ENABLED |
4335                                             SLBPPD_ARMING_INC;
4336                                         if (lun->be_lun->blocksize)
4337                                                 ival /= lun->be_lun->blocksize;
4338                                         else
4339                                                 ival /= 512;
4340                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4341                                             page->descr[3].count);
4342                                 }
4343                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4344                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4345                                        sizeof(lbp_page_default));
4346                                 page_index->page_data =
4347                                         (uint8_t *)lun->mode_pages.lbp_page;
4348                                 break;
4349                         }
4350                         default:
4351                                 panic("subpage %#x for page %#x is incorrect!",
4352                                       page_index->subpage, page_code);
4353                         }
4354                         break;
4355                 }
4356                 case SMS_CDDVD_CAPS_PAGE:{
4357                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4358                             ("subpage %#x for page %#x is incorrect!",
4359                             page_index->subpage, page_code));
4360                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4361                                &cddvd_page_default,
4362                                sizeof(cddvd_page_default));
4363                         memcpy(&lun->mode_pages.cddvd_page[
4364                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4365                                sizeof(cddvd_page_changeable));
4366                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4367                                &cddvd_page_default,
4368                                sizeof(cddvd_page_default));
4369                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4370                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4371                                sizeof(cddvd_page_default));
4372                         page_index->page_data =
4373                                 (uint8_t *)lun->mode_pages.cddvd_page;
4374                         break;
4375                 }
4376                 default:
4377                         panic("invalid page code value %#x", page_code);
4378                 }
4379         }
4380
4381         return (CTL_RETVAL_COMPLETE);
4382 }
4383
4384 static int
4385 ctl_init_log_page_index(struct ctl_lun *lun)
4386 {
4387         struct ctl_page_index *page_index;
4388         int i, j, k, prev;
4389
4390         memcpy(&lun->log_pages.index, log_page_index_template,
4391                sizeof(log_page_index_template));
4392
4393         prev = -1;
4394         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4395
4396                 page_index = &lun->log_pages.index[i];
4397                 if (lun->be_lun->lun_type == T_DIRECT &&
4398                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4399                         continue;
4400                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4401                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4402                         continue;
4403                 if (lun->be_lun->lun_type == T_CDROM &&
4404                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4405                         continue;
4406
4407                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4408                     lun->backend->lun_attr == NULL)
4409                         continue;
4410
4411                 if (page_index->page_code != prev) {
4412                         lun->log_pages.pages_page[j] = page_index->page_code;
4413                         prev = page_index->page_code;
4414                         j++;
4415                 }
4416                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4417                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4418                 k++;
4419         }
4420         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4421         lun->log_pages.index[0].page_len = j;
4422         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4423         lun->log_pages.index[1].page_len = k * 2;
4424         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4425         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4426         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4427         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4428         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.ie_page;
4429         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.ie_page);
4430
4431         return (CTL_RETVAL_COMPLETE);
4432 }
4433
4434 static int
4435 hex2bin(const char *str, uint8_t *buf, int buf_size)
4436 {
4437         int i;
4438         u_char c;
4439
4440         memset(buf, 0, buf_size);
4441         while (isspace(str[0]))
4442                 str++;
4443         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4444                 str += 2;
4445         buf_size *= 2;
4446         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4447                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4448                         str++;
4449                 c = str[i];
4450                 if (isdigit(c))
4451                         c -= '0';
4452                 else if (isalpha(c))
4453                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4454                 else
4455                         break;
4456                 if (c >= 16)
4457                         break;
4458                 if ((i & 1) == 0)
4459                         buf[i / 2] |= (c << 4);
4460                 else
4461                         buf[i / 2] |= c;
4462         }
4463         return ((i + 1) / 2);
4464 }
4465
4466 /*
4467  * LUN allocation.
4468  *
4469  * Requirements:
4470  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4471  *   wants us to allocate the LUN and he can block.
4472  * - ctl_softc is always set
4473  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4474  *
4475  * Returns 0 for success, non-zero (errno) for failure.
4476  */
4477 static int
4478 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4479               struct ctl_be_lun *const be_lun)
4480 {
4481         struct ctl_lun *nlun, *lun;
4482         struct scsi_vpd_id_descriptor *desc;
4483         struct scsi_vpd_id_t10 *t10id;
4484         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4485         int lun_number, lun_malloced;
4486         int devidlen, idlen1, idlen2 = 0, len;
4487
4488         if (be_lun == NULL)
4489                 return (EINVAL);
4490
4491         /*
4492          * We currently only support Direct Access or Processor LUN types.
4493          */
4494         switch (be_lun->lun_type) {
4495         case T_DIRECT:
4496         case T_PROCESSOR:
4497         case T_CDROM:
4498                 break;
4499         case T_SEQUENTIAL:
4500         case T_CHANGER:
4501         default:
4502                 be_lun->lun_config_status(be_lun->be_lun,
4503                                           CTL_LUN_CONFIG_FAILURE);
4504                 break;
4505         }
4506         if (ctl_lun == NULL) {
4507                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4508                 lun_malloced = 1;
4509         } else {
4510                 lun_malloced = 0;
4511                 lun = ctl_lun;
4512         }
4513
4514         memset(lun, 0, sizeof(*lun));
4515         if (lun_malloced)
4516                 lun->flags = CTL_LUN_MALLOCED;
4517
4518         /* Generate LUN ID. */
4519         devidlen = max(CTL_DEVID_MIN_LEN,
4520             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4521         idlen1 = sizeof(*t10id) + devidlen;
4522         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4523         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4524         if (scsiname != NULL) {
4525                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4526                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4527         }
4528         eui = ctl_get_opt(&be_lun->options, "eui");
4529         if (eui != NULL) {
4530                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4531         }
4532         naa = ctl_get_opt(&be_lun->options, "naa");
4533         if (naa != NULL) {
4534                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4535         }
4536         uuid = ctl_get_opt(&be_lun->options, "uuid");
4537         if (uuid != NULL) {
4538                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4539         }
4540         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4541             M_CTL, M_WAITOK | M_ZERO);
4542         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4543         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4544         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4545         desc->length = idlen1;
4546         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4547         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4548         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4549                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4550         } else {
4551                 strncpy(t10id->vendor, vendor,
4552                     min(sizeof(t10id->vendor), strlen(vendor)));
4553         }
4554         strncpy((char *)t10id->vendor_spec_id,
4555             (char *)be_lun->device_id, devidlen);
4556         if (scsiname != NULL) {
4557                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4558                     desc->length);
4559                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4560                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4561                     SVPD_ID_TYPE_SCSI_NAME;
4562                 desc->length = idlen2;
4563                 strlcpy(desc->identifier, scsiname, idlen2);
4564         }
4565         if (eui != NULL) {
4566                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4567                     desc->length);
4568                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4569                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4570                     SVPD_ID_TYPE_EUI64;
4571                 desc->length = hex2bin(eui, desc->identifier, 16);
4572                 desc->length = desc->length > 12 ? 16 :
4573                     (desc->length > 8 ? 12 : 8);
4574                 len -= 16 - desc->length;
4575         }
4576         if (naa != NULL) {
4577                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4578                     desc->length);
4579                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4580                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4581                     SVPD_ID_TYPE_NAA;
4582                 desc->length = hex2bin(naa, desc->identifier, 16);
4583                 desc->length = desc->length > 8 ? 16 : 8;
4584                 len -= 16 - desc->length;
4585         }
4586         if (uuid != NULL) {
4587                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4588                     desc->length);
4589                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4590                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4591                     SVPD_ID_TYPE_UUID;
4592                 desc->identifier[0] = 0x10;
4593                 hex2bin(uuid, &desc->identifier[2], 16);
4594                 desc->length = 18;
4595         }
4596         lun->lun_devid->len = len;
4597
4598         mtx_lock(&ctl_softc->ctl_lock);
4599         /*
4600          * See if the caller requested a particular LUN number.  If so, see
4601          * if it is available.  Otherwise, allocate the first available LUN.
4602          */
4603         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4604                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4605                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4606                         mtx_unlock(&ctl_softc->ctl_lock);
4607                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4608                                 printf("ctl: requested LUN ID %d is higher "
4609                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4610                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4611                         } else {
4612                                 /*
4613                                  * XXX KDM return an error, or just assign
4614                                  * another LUN ID in this case??
4615                                  */
4616                                 printf("ctl: requested LUN ID %d is already "
4617                                        "in use\n", be_lun->req_lun_id);
4618                         }
4619 fail:
4620                         free(lun->lun_devid, M_CTL);
4621                         if (lun->flags & CTL_LUN_MALLOCED)
4622                                 free(lun, M_CTL);
4623                         be_lun->lun_config_status(be_lun->be_lun,
4624                                                   CTL_LUN_CONFIG_FAILURE);
4625                         return (ENOSPC);
4626                 }
4627                 lun_number = be_lun->req_lun_id;
4628         } else {
4629                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4630                 if (lun_number == -1) {
4631                         mtx_unlock(&ctl_softc->ctl_lock);
4632                         printf("ctl: can't allocate LUN, out of LUNs\n");
4633                         goto fail;
4634                 }
4635         }
4636         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4637         mtx_unlock(&ctl_softc->ctl_lock);
4638
4639         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4640         lun->lun = lun_number;
4641         lun->be_lun = be_lun;
4642         /*
4643          * The processor LUN is always enabled.  Disk LUNs come on line
4644          * disabled, and must be enabled by the backend.
4645          */
4646         lun->flags |= CTL_LUN_DISABLED;
4647         lun->backend = be_lun->be;
4648         be_lun->ctl_lun = lun;
4649         be_lun->lun_id = lun_number;
4650         atomic_add_int(&be_lun->be->num_luns, 1);
4651         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4652                 lun->flags |= CTL_LUN_EJECTED;
4653         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4654                 lun->flags |= CTL_LUN_NO_MEDIA;
4655         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4656                 lun->flags |= CTL_LUN_STOPPED;
4657
4658         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4659                 lun->flags |= CTL_LUN_PRIMARY_SC;
4660
4661         value = ctl_get_opt(&be_lun->options, "removable");
4662         if (value != NULL) {
4663                 if (strcmp(value, "on") == 0)
4664                         lun->flags |= CTL_LUN_REMOVABLE;
4665         } else if (be_lun->lun_type == T_CDROM)
4666                 lun->flags |= CTL_LUN_REMOVABLE;
4667
4668         lun->ctl_softc = ctl_softc;
4669 #ifdef CTL_TIME_IO
4670         lun->last_busy = getsbinuptime();
4671 #endif
4672         TAILQ_INIT(&lun->ooa_queue);
4673         TAILQ_INIT(&lun->blocked_queue);
4674         STAILQ_INIT(&lun->error_list);
4675         lun->ie_reported = 1;
4676         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4677         ctl_tpc_lun_init(lun);
4678         if (lun->flags & CTL_LUN_REMOVABLE) {
4679                 lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4,
4680                     M_CTL, M_WAITOK);
4681         }
4682
4683         /*
4684          * Initialize the mode and log page index.
4685          */
4686         ctl_init_page_index(lun);
4687         ctl_init_log_page_index(lun);
4688
4689         /* Setup statistics gathering */
4690 #ifdef CTL_LEGACY_STATS
4691         lun->legacy_stats.device_type = be_lun->lun_type;
4692         lun->legacy_stats.lun_number = lun_number;
4693         lun->legacy_stats.blocksize = be_lun->blocksize;
4694         if (be_lun->blocksize == 0)
4695                 lun->legacy_stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4696         for (len = 0; len < CTL_MAX_PORTS; len++)
4697                 lun->legacy_stats.ports[len].targ_port = len;
4698 #endif /* CTL_LEGACY_STATS */
4699         lun->stats.item = lun_number;
4700
4701         /*
4702          * Now, before we insert this lun on the lun list, set the lun
4703          * inventory changed UA for all other luns.
4704          */
4705         mtx_lock(&ctl_softc->ctl_lock);
4706         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4707                 mtx_lock(&nlun->lun_lock);
4708                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4709                 mtx_unlock(&nlun->lun_lock);
4710         }
4711         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4712         ctl_softc->ctl_luns[lun_number] = lun;
4713         ctl_softc->num_luns++;
4714         mtx_unlock(&ctl_softc->ctl_lock);
4715
4716         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4717         return (0);
4718 }
4719
4720 /*
4721  * Delete a LUN.
4722  * Assumptions:
4723  * - LUN has already been marked invalid and any pending I/O has been taken
4724  *   care of.
4725  */
4726 static int
4727 ctl_free_lun(struct ctl_lun *lun)
4728 {
4729         struct ctl_softc *softc = lun->ctl_softc;
4730         struct ctl_lun *nlun;
4731         int i;
4732
4733         mtx_assert(&softc->ctl_lock, MA_OWNED);
4734
4735         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4736
4737         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4738
4739         softc->ctl_luns[lun->lun] = NULL;
4740
4741         if (!TAILQ_EMPTY(&lun->ooa_queue))
4742                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4743
4744         softc->num_luns--;
4745
4746         /*
4747          * Tell the backend to free resources, if this LUN has a backend.
4748          */
4749         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4750         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4751
4752         lun->ie_reportcnt = UINT32_MAX;
4753         callout_drain(&lun->ie_callout);
4754
4755         ctl_tpc_lun_shutdown(lun);
4756         mtx_destroy(&lun->lun_lock);
4757         free(lun->lun_devid, M_CTL);
4758         for (i = 0; i < CTL_MAX_PORTS; i++)
4759                 free(lun->pending_ua[i], M_CTL);
4760         for (i = 0; i < CTL_MAX_PORTS; i++)
4761                 free(lun->pr_keys[i], M_CTL);
4762         free(lun->write_buffer, M_CTL);
4763         free(lun->prevent, M_CTL);
4764         if (lun->flags & CTL_LUN_MALLOCED)
4765                 free(lun, M_CTL);
4766
4767         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4768                 mtx_lock(&nlun->lun_lock);
4769                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4770                 mtx_unlock(&nlun->lun_lock);
4771         }
4772
4773         return (0);
4774 }
4775
4776 static void
4777 ctl_create_lun(struct ctl_be_lun *be_lun)
4778 {
4779
4780         /*
4781          * ctl_alloc_lun() should handle all potential failure cases.
4782          */
4783         ctl_alloc_lun(control_softc, NULL, be_lun);
4784 }
4785
4786 int
4787 ctl_add_lun(struct ctl_be_lun *be_lun)
4788 {
4789         struct ctl_softc *softc = control_softc;
4790
4791         mtx_lock(&softc->ctl_lock);
4792         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4793         mtx_unlock(&softc->ctl_lock);
4794         wakeup(&softc->pending_lun_queue);
4795
4796         return (0);
4797 }
4798
4799 int
4800 ctl_enable_lun(struct ctl_be_lun *be_lun)
4801 {
4802         struct ctl_softc *softc;
4803         struct ctl_port *port, *nport;
4804         struct ctl_lun *lun;
4805         int retval;
4806
4807         lun = (struct ctl_lun *)be_lun->ctl_lun;
4808         softc = lun->ctl_softc;
4809
4810         mtx_lock(&softc->ctl_lock);
4811         mtx_lock(&lun->lun_lock);
4812         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4813                 /*
4814                  * eh?  Why did we get called if the LUN is already
4815                  * enabled?
4816                  */
4817                 mtx_unlock(&lun->lun_lock);
4818                 mtx_unlock(&softc->ctl_lock);
4819                 return (0);
4820         }
4821         lun->flags &= ~CTL_LUN_DISABLED;
4822         mtx_unlock(&lun->lun_lock);
4823
4824         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4825                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4826                     port->lun_map != NULL || port->lun_enable == NULL)
4827                         continue;
4828
4829                 /*
4830                  * Drop the lock while we call the FETD's enable routine.
4831                  * This can lead to a callback into CTL (at least in the
4832                  * case of the internal initiator frontend.
4833                  */
4834                 mtx_unlock(&softc->ctl_lock);
4835                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4836                 mtx_lock(&softc->ctl_lock);
4837                 if (retval != 0) {
4838                         printf("%s: FETD %s port %d returned error "
4839                                "%d for lun_enable on lun %jd\n",
4840                                __func__, port->port_name, port->targ_port,
4841                                retval, (intmax_t)lun->lun);
4842                 }
4843         }
4844
4845         mtx_unlock(&softc->ctl_lock);
4846         ctl_isc_announce_lun(lun);
4847
4848         return (0);
4849 }
4850
4851 int
4852 ctl_disable_lun(struct ctl_be_lun *be_lun)
4853 {
4854         struct ctl_softc *softc;
4855         struct ctl_port *port;
4856         struct ctl_lun *lun;
4857         int retval;
4858
4859         lun = (struct ctl_lun *)be_lun->ctl_lun;
4860         softc = lun->ctl_softc;
4861
4862         mtx_lock(&softc->ctl_lock);
4863         mtx_lock(&lun->lun_lock);
4864         if (lun->flags & CTL_LUN_DISABLED) {
4865                 mtx_unlock(&lun->lun_lock);
4866                 mtx_unlock(&softc->ctl_lock);
4867                 return (0);
4868         }
4869         lun->flags |= CTL_LUN_DISABLED;
4870         mtx_unlock(&lun->lun_lock);
4871
4872         STAILQ_FOREACH(port, &softc->port_list, links) {
4873                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4874                     port->lun_map != NULL || port->lun_disable == NULL)
4875                         continue;
4876
4877                 /*
4878                  * Drop the lock before we call the frontend's disable
4879                  * routine, to avoid lock order reversals.
4880                  *
4881                  * XXX KDM what happens if the frontend list changes while
4882                  * we're traversing it?  It's unlikely, but should be handled.
4883                  */
4884                 mtx_unlock(&softc->ctl_lock);
4885                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4886                 mtx_lock(&softc->ctl_lock);
4887                 if (retval != 0) {
4888                         printf("%s: FETD %s port %d returned error "
4889                                "%d for lun_disable on lun %jd\n",
4890                                __func__, port->port_name, port->targ_port,
4891                                retval, (intmax_t)lun->lun);
4892                 }
4893         }
4894
4895         mtx_unlock(&softc->ctl_lock);
4896         ctl_isc_announce_lun(lun);
4897
4898         return (0);
4899 }
4900
4901 int
4902 ctl_start_lun(struct ctl_be_lun *be_lun)
4903 {
4904         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4905
4906         mtx_lock(&lun->lun_lock);
4907         lun->flags &= ~CTL_LUN_STOPPED;
4908         mtx_unlock(&lun->lun_lock);
4909         return (0);
4910 }
4911
4912 int
4913 ctl_stop_lun(struct ctl_be_lun *be_lun)
4914 {
4915         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4916
4917         mtx_lock(&lun->lun_lock);
4918         lun->flags |= CTL_LUN_STOPPED;
4919         mtx_unlock(&lun->lun_lock);
4920         return (0);
4921 }
4922
4923 int
4924 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4925 {
4926         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4927
4928         mtx_lock(&lun->lun_lock);
4929         lun->flags |= CTL_LUN_NO_MEDIA;
4930         mtx_unlock(&lun->lun_lock);
4931         return (0);
4932 }
4933
4934 int
4935 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4936 {
4937         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4938         union ctl_ha_msg msg;
4939
4940         mtx_lock(&lun->lun_lock);
4941         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4942         if (lun->flags & CTL_LUN_REMOVABLE)
4943                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4944         mtx_unlock(&lun->lun_lock);
4945         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4946             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4947                 bzero(&msg.ua, sizeof(msg.ua));
4948                 msg.hdr.msg_type = CTL_MSG_UA;
4949                 msg.hdr.nexus.initid = -1;
4950                 msg.hdr.nexus.targ_port = -1;
4951                 msg.hdr.nexus.targ_lun = lun->lun;
4952                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4953                 msg.ua.ua_all = 1;
4954                 msg.ua.ua_set = 1;
4955                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4956                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4957                     M_WAITOK);
4958         }
4959         return (0);
4960 }
4961
4962 int
4963 ctl_lun_ejected(struct ctl_be_lun *be_lun)
4964 {
4965         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4966
4967         mtx_lock(&lun->lun_lock);
4968         lun->flags |= CTL_LUN_EJECTED;
4969         mtx_unlock(&lun->lun_lock);
4970         return (0);
4971 }
4972
4973 int
4974 ctl_lun_primary(struct ctl_be_lun *be_lun)
4975 {
4976         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4977
4978         mtx_lock(&lun->lun_lock);
4979         lun->flags |= CTL_LUN_PRIMARY_SC;
4980         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4981         mtx_unlock(&lun->lun_lock);
4982         ctl_isc_announce_lun(lun);
4983         return (0);
4984 }
4985
4986 int
4987 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4988 {
4989         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4990
4991         mtx_lock(&lun->lun_lock);
4992         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4993         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4994         mtx_unlock(&lun->lun_lock);
4995         ctl_isc_announce_lun(lun);
4996         return (0);
4997 }
4998
4999 int
5000 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
5001 {
5002         struct ctl_softc *softc;
5003         struct ctl_lun *lun;
5004
5005         lun = (struct ctl_lun *)be_lun->ctl_lun;
5006         softc = lun->ctl_softc;
5007
5008         mtx_lock(&lun->lun_lock);
5009
5010         /*
5011          * The LUN needs to be disabled before it can be marked invalid.
5012          */
5013         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
5014                 mtx_unlock(&lun->lun_lock);
5015                 return (-1);
5016         }
5017         /*
5018          * Mark the LUN invalid.
5019          */
5020         lun->flags |= CTL_LUN_INVALID;
5021
5022         /*
5023          * If there is nothing in the OOA queue, go ahead and free the LUN.
5024          * If we have something in the OOA queue, we'll free it when the
5025          * last I/O completes.
5026          */
5027         if (TAILQ_EMPTY(&lun->ooa_queue)) {
5028                 mtx_unlock(&lun->lun_lock);
5029                 mtx_lock(&softc->ctl_lock);
5030                 ctl_free_lun(lun);
5031                 mtx_unlock(&softc->ctl_lock);
5032         } else
5033                 mtx_unlock(&lun->lun_lock);
5034
5035         return (0);
5036 }
5037
5038 void
5039 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5040 {
5041         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5042         union ctl_ha_msg msg;
5043
5044         mtx_lock(&lun->lun_lock);
5045         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
5046         mtx_unlock(&lun->lun_lock);
5047         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5048                 /* Send msg to other side. */
5049                 bzero(&msg.ua, sizeof(msg.ua));
5050                 msg.hdr.msg_type = CTL_MSG_UA;
5051                 msg.hdr.nexus.initid = -1;
5052                 msg.hdr.nexus.targ_port = -1;
5053                 msg.hdr.nexus.targ_lun = lun->lun;
5054                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5055                 msg.ua.ua_all = 1;
5056                 msg.ua.ua_set = 1;
5057                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
5058                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5059                     M_WAITOK);
5060         }
5061 }
5062
5063 /*
5064  * Backend "memory move is complete" callback for requests that never
5065  * make it down to say RAIDCore's configuration code.
5066  */
5067 int
5068 ctl_config_move_done(union ctl_io *io)
5069 {
5070         int retval;
5071
5072         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5073         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5074             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5075
5076         if ((io->io_hdr.port_status != 0) &&
5077             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5078              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5079                 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
5080                     /*retry_count*/ io->io_hdr.port_status);
5081         } else if (io->scsiio.kern_data_resid != 0 &&
5082             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
5083             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5084              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5085                 ctl_set_invalid_field_ciu(&io->scsiio);
5086         }
5087
5088         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5089                 ctl_data_print(io);
5090         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5091             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5092              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5093             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5094                 /*
5095                  * XXX KDM just assuming a single pointer here, and not a
5096                  * S/G list.  If we start using S/G lists for config data,
5097                  * we'll need to know how to clean them up here as well.
5098                  */
5099                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5100                         free(io->scsiio.kern_data_ptr, M_CTL);
5101                 ctl_done(io);
5102                 retval = CTL_RETVAL_COMPLETE;
5103         } else {
5104                 /*
5105                  * XXX KDM now we need to continue data movement.  Some
5106                  * options:
5107                  * - call ctl_scsiio() again?  We don't do this for data
5108                  *   writes, because for those at least we know ahead of
5109                  *   time where the write will go and how long it is.  For
5110                  *   config writes, though, that information is largely
5111                  *   contained within the write itself, thus we need to
5112                  *   parse out the data again.
5113                  *
5114                  * - Call some other function once the data is in?
5115                  */
5116
5117                 /*
5118                  * XXX KDM call ctl_scsiio() again for now, and check flag
5119                  * bits to see whether we're allocated or not.
5120                  */
5121                 retval = ctl_scsiio(&io->scsiio);
5122         }
5123         return (retval);
5124 }
5125
5126 /*
5127  * This gets called by a backend driver when it is done with a
5128  * data_submit method.
5129  */
5130 void
5131 ctl_data_submit_done(union ctl_io *io)
5132 {
5133         /*
5134          * If the IO_CONT flag is set, we need to call the supplied
5135          * function to continue processing the I/O, instead of completing
5136          * the I/O just yet.
5137          *
5138          * If there is an error, though, we don't want to keep processing.
5139          * Instead, just send status back to the initiator.
5140          */
5141         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5142             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5143             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5144              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5145                 io->scsiio.io_cont(io);
5146                 return;
5147         }
5148         ctl_done(io);
5149 }
5150
5151 /*
5152  * This gets called by a backend driver when it is done with a
5153  * configuration write.
5154  */
5155 void
5156 ctl_config_write_done(union ctl_io *io)
5157 {
5158         uint8_t *buf;
5159
5160         /*
5161          * If the IO_CONT flag is set, we need to call the supplied
5162          * function to continue processing the I/O, instead of completing
5163          * the I/O just yet.
5164          *
5165          * If there is an error, though, we don't want to keep processing.
5166          * Instead, just send status back to the initiator.
5167          */
5168         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5169             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5170             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5171              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5172                 io->scsiio.io_cont(io);
5173                 return;
5174         }
5175         /*
5176          * Since a configuration write can be done for commands that actually
5177          * have data allocated, like write buffer, and commands that have
5178          * no data, like start/stop unit, we need to check here.
5179          */
5180         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5181                 buf = io->scsiio.kern_data_ptr;
5182         else
5183                 buf = NULL;
5184         ctl_done(io);
5185         if (buf)
5186                 free(buf, M_CTL);
5187 }
5188
5189 void
5190 ctl_config_read_done(union ctl_io *io)
5191 {
5192         uint8_t *buf;
5193
5194         /*
5195          * If there is some error -- we are done, skip data transfer.
5196          */
5197         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5198             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5199              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5200                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5201                         buf = io->scsiio.kern_data_ptr;
5202                 else
5203                         buf = NULL;
5204                 ctl_done(io);
5205                 if (buf)
5206                         free(buf, M_CTL);
5207                 return;
5208         }
5209
5210         /*
5211          * If the IO_CONT flag is set, we need to call the supplied
5212          * function to continue processing the I/O, instead of completing
5213          * the I/O just yet.
5214          */
5215         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5216                 io->scsiio.io_cont(io);
5217                 return;
5218         }
5219
5220         ctl_datamove(io);
5221 }
5222
5223 /*
5224  * SCSI release command.
5225  */
5226 int
5227 ctl_scsi_release(struct ctl_scsiio *ctsio)
5228 {
5229         struct ctl_lun *lun = CTL_LUN(ctsio);
5230         uint32_t residx;
5231
5232         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5233
5234         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5235
5236         /*
5237          * XXX KDM right now, we only support LUN reservation.  We don't
5238          * support 3rd party reservations, or extent reservations, which
5239          * might actually need the parameter list.  If we've gotten this
5240          * far, we've got a LUN reservation.  Anything else got kicked out
5241          * above.  So, according to SPC, ignore the length.
5242          */
5243
5244         mtx_lock(&lun->lun_lock);
5245
5246         /*
5247          * According to SPC, it is not an error for an intiator to attempt
5248          * to release a reservation on a LUN that isn't reserved, or that
5249          * is reserved by another initiator.  The reservation can only be
5250          * released, though, by the initiator who made it or by one of
5251          * several reset type events.
5252          */
5253         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5254                         lun->flags &= ~CTL_LUN_RESERVED;
5255
5256         mtx_unlock(&lun->lun_lock);
5257
5258         ctl_set_success(ctsio);
5259         ctl_done((union ctl_io *)ctsio);
5260         return (CTL_RETVAL_COMPLETE);
5261 }
5262
5263 int
5264 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5265 {
5266         struct ctl_lun *lun = CTL_LUN(ctsio);
5267         uint32_t residx;
5268
5269         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5270
5271         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5272
5273         /*
5274          * XXX KDM right now, we only support LUN reservation.  We don't
5275          * support 3rd party reservations, or extent reservations, which
5276          * might actually need the parameter list.  If we've gotten this
5277          * far, we've got a LUN reservation.  Anything else got kicked out
5278          * above.  So, according to SPC, ignore the length.
5279          */
5280
5281         mtx_lock(&lun->lun_lock);
5282         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5283                 ctl_set_reservation_conflict(ctsio);
5284                 goto bailout;
5285         }
5286
5287         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5288         if (lun->flags & CTL_LUN_PR_RESERVED) {
5289                 ctl_set_success(ctsio);
5290                 goto bailout;
5291         }
5292
5293         lun->flags |= CTL_LUN_RESERVED;
5294         lun->res_idx = residx;
5295         ctl_set_success(ctsio);
5296
5297 bailout:
5298         mtx_unlock(&lun->lun_lock);
5299         ctl_done((union ctl_io *)ctsio);
5300         return (CTL_RETVAL_COMPLETE);
5301 }
5302
5303 int
5304 ctl_start_stop(struct ctl_scsiio *ctsio)
5305 {
5306         struct ctl_lun *lun = CTL_LUN(ctsio);
5307         struct scsi_start_stop_unit *cdb;
5308         int retval;
5309
5310         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5311
5312         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5313
5314         if ((cdb->how & SSS_PC_MASK) == 0) {
5315                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5316                     (cdb->how & SSS_START) == 0) {
5317                         uint32_t residx;
5318
5319                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5320                         if (ctl_get_prkey(lun, residx) == 0 ||
5321                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5322
5323                                 ctl_set_reservation_conflict(ctsio);
5324                                 ctl_done((union ctl_io *)ctsio);
5325                                 return (CTL_RETVAL_COMPLETE);
5326                         }
5327                 }
5328
5329                 if ((cdb->how & SSS_LOEJ) &&
5330                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5331                         ctl_set_invalid_field(ctsio,
5332                                               /*sks_valid*/ 1,
5333                                               /*command*/ 1,
5334                                               /*field*/ 4,
5335                                               /*bit_valid*/ 1,
5336                                               /*bit*/ 1);
5337                         ctl_done((union ctl_io *)ctsio);
5338                         return (CTL_RETVAL_COMPLETE);
5339                 }
5340
5341                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5342                     lun->prevent_count > 0) {
5343                         /* "Medium removal prevented" */
5344                         ctl_set_sense(ctsio, /*current_error*/ 1,
5345                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5346                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5347                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5348                         ctl_done((union ctl_io *)ctsio);
5349                         return (CTL_RETVAL_COMPLETE);
5350                 }
5351         }
5352
5353         retval = lun->backend->config_write((union ctl_io *)ctsio);
5354         return (retval);
5355 }
5356
5357 int
5358 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5359 {
5360         struct ctl_lun *lun = CTL_LUN(ctsio);
5361         struct scsi_prevent *cdb;
5362         int retval;
5363         uint32_t initidx;
5364
5365         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5366
5367         cdb = (struct scsi_prevent *)ctsio->cdb;
5368
5369         if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) {
5370                 ctl_set_invalid_opcode(ctsio);
5371                 ctl_done((union ctl_io *)ctsio);
5372                 return (CTL_RETVAL_COMPLETE);
5373         }
5374
5375         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5376         mtx_lock(&lun->lun_lock);
5377         if ((cdb->how & PR_PREVENT) &&
5378             ctl_is_set(lun->prevent, initidx) == 0) {
5379                 ctl_set_mask(lun->prevent, initidx);
5380                 lun->prevent_count++;
5381         } else if ((cdb->how & PR_PREVENT) == 0 &&
5382             ctl_is_set(lun->prevent, initidx)) {
5383                 ctl_clear_mask(lun->prevent, initidx);
5384                 lun->prevent_count--;
5385         }
5386         mtx_unlock(&lun->lun_lock);
5387         retval = lun->backend->config_write((union ctl_io *)ctsio);
5388         return (retval);
5389 }
5390
5391 /*
5392  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5393  * we don't really do anything with the LBA and length fields if the user
5394  * passes them in.  Instead we'll just flush out the cache for the entire
5395  * LUN.
5396  */
5397 int
5398 ctl_sync_cache(struct ctl_scsiio *ctsio)
5399 {
5400         struct ctl_lun *lun = CTL_LUN(ctsio);
5401         struct ctl_lba_len_flags *lbalen;
5402         uint64_t starting_lba;
5403         uint32_t block_count;
5404         int retval;
5405         uint8_t byte2;
5406
5407         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5408
5409         retval = 0;
5410
5411         switch (ctsio->cdb[0]) {
5412         case SYNCHRONIZE_CACHE: {
5413                 struct scsi_sync_cache *cdb;
5414                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5415
5416                 starting_lba = scsi_4btoul(cdb->begin_lba);
5417                 block_count = scsi_2btoul(cdb->lb_count);
5418                 byte2 = cdb->byte2;
5419                 break;
5420         }
5421         case SYNCHRONIZE_CACHE_16: {
5422                 struct scsi_sync_cache_16 *cdb;
5423                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5424
5425                 starting_lba = scsi_8btou64(cdb->begin_lba);
5426                 block_count = scsi_4btoul(cdb->lb_count);
5427                 byte2 = cdb->byte2;
5428                 break;
5429         }
5430         default:
5431                 ctl_set_invalid_opcode(ctsio);
5432                 ctl_done((union ctl_io *)ctsio);
5433                 goto bailout;
5434                 break; /* NOTREACHED */
5435         }
5436
5437         /*
5438          * We check the LBA and length, but don't do anything with them.
5439          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5440          * get flushed.  This check will just help satisfy anyone who wants
5441          * to see an error for an out of range LBA.
5442          */
5443         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5444                 ctl_set_lba_out_of_range(ctsio,
5445                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5446                 ctl_done((union ctl_io *)ctsio);
5447                 goto bailout;
5448         }
5449
5450         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5451         lbalen->lba = starting_lba;
5452         lbalen->len = block_count;
5453         lbalen->flags = byte2;
5454         retval = lun->backend->config_write((union ctl_io *)ctsio);
5455
5456 bailout:
5457         return (retval);
5458 }
5459
5460 int
5461 ctl_format(struct ctl_scsiio *ctsio)
5462 {
5463         struct scsi_format *cdb;
5464         int length, defect_list_len;
5465
5466         CTL_DEBUG_PRINT(("ctl_format\n"));
5467
5468         cdb = (struct scsi_format *)ctsio->cdb;
5469
5470         length = 0;
5471         if (cdb->byte2 & SF_FMTDATA) {
5472                 if (cdb->byte2 & SF_LONGLIST)
5473                         length = sizeof(struct scsi_format_header_long);
5474                 else
5475                         length = sizeof(struct scsi_format_header_short);
5476         }
5477
5478         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5479          && (length > 0)) {
5480                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5481                 ctsio->kern_data_len = length;
5482                 ctsio->kern_total_len = length;
5483                 ctsio->kern_rel_offset = 0;
5484                 ctsio->kern_sg_entries = 0;
5485                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5486                 ctsio->be_move_done = ctl_config_move_done;
5487                 ctl_datamove((union ctl_io *)ctsio);
5488
5489                 return (CTL_RETVAL_COMPLETE);
5490         }
5491
5492         defect_list_len = 0;
5493
5494         if (cdb->byte2 & SF_FMTDATA) {
5495                 if (cdb->byte2 & SF_LONGLIST) {
5496                         struct scsi_format_header_long *header;
5497
5498                         header = (struct scsi_format_header_long *)
5499                                 ctsio->kern_data_ptr;
5500
5501                         defect_list_len = scsi_4btoul(header->defect_list_len);
5502                         if (defect_list_len != 0) {
5503                                 ctl_set_invalid_field(ctsio,
5504                                                       /*sks_valid*/ 1,
5505                                                       /*command*/ 0,
5506                                                       /*field*/ 2,
5507                                                       /*bit_valid*/ 0,
5508                                                       /*bit*/ 0);
5509                                 goto bailout;
5510                         }
5511                 } else {
5512                         struct scsi_format_header_short *header;
5513
5514                         header = (struct scsi_format_header_short *)
5515                                 ctsio->kern_data_ptr;
5516
5517                         defect_list_len = scsi_2btoul(header->defect_list_len);
5518                         if (defect_list_len != 0) {
5519                                 ctl_set_invalid_field(ctsio,
5520                                                       /*sks_valid*/ 1,
5521                                                       /*command*/ 0,
5522                                                       /*field*/ 2,
5523                                                       /*bit_valid*/ 0,
5524                                                       /*bit*/ 0);
5525                                 goto bailout;
5526                         }
5527                 }
5528         }
5529
5530         ctl_set_success(ctsio);
5531 bailout:
5532
5533         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5534                 free(ctsio->kern_data_ptr, M_CTL);
5535                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5536         }
5537
5538         ctl_done((union ctl_io *)ctsio);
5539         return (CTL_RETVAL_COMPLETE);
5540 }
5541
5542 int
5543 ctl_read_buffer(struct ctl_scsiio *ctsio)
5544 {
5545         struct ctl_lun *lun = CTL_LUN(ctsio);
5546         uint64_t buffer_offset;
5547         uint32_t len;
5548         uint8_t byte2;
5549         static uint8_t descr[4];
5550         static uint8_t echo_descr[4] = { 0 };
5551
5552         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5553
5554         switch (ctsio->cdb[0]) {
5555         case READ_BUFFER: {
5556                 struct scsi_read_buffer *cdb;
5557
5558                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5559                 buffer_offset = scsi_3btoul(cdb->offset);
5560                 len = scsi_3btoul(cdb->length);
5561                 byte2 = cdb->byte2;
5562                 break;
5563         }
5564         case READ_BUFFER_16: {
5565                 struct scsi_read_buffer_16 *cdb;
5566
5567                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5568                 buffer_offset = scsi_8btou64(cdb->offset);
5569                 len = scsi_4btoul(cdb->length);
5570                 byte2 = cdb->byte2;
5571                 break;
5572         }
5573         default: /* This shouldn't happen. */
5574                 ctl_set_invalid_opcode(ctsio);
5575                 ctl_done((union ctl_io *)ctsio);
5576                 return (CTL_RETVAL_COMPLETE);
5577         }
5578
5579         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5580             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5581                 ctl_set_invalid_field(ctsio,
5582                                       /*sks_valid*/ 1,
5583                                       /*command*/ 1,
5584                                       /*field*/ 6,
5585                                       /*bit_valid*/ 0,
5586                                       /*bit*/ 0);
5587                 ctl_done((union ctl_io *)ctsio);
5588                 return (CTL_RETVAL_COMPLETE);
5589         }
5590
5591         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5592                 descr[0] = 0;
5593                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5594                 ctsio->kern_data_ptr = descr;
5595                 len = min(len, sizeof(descr));
5596         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5597                 ctsio->kern_data_ptr = echo_descr;
5598                 len = min(len, sizeof(echo_descr));
5599         } else {
5600                 if (lun->write_buffer == NULL) {
5601                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5602                             M_CTL, M_WAITOK);
5603                 }
5604                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5605         }
5606         ctsio->kern_data_len = len;
5607         ctsio->kern_total_len = len;
5608         ctsio->kern_rel_offset = 0;
5609         ctsio->kern_sg_entries = 0;
5610         ctl_set_success(ctsio);
5611         ctsio->be_move_done = ctl_config_move_done;
5612         ctl_datamove((union ctl_io *)ctsio);
5613         return (CTL_RETVAL_COMPLETE);
5614 }
5615
5616 int
5617 ctl_write_buffer(struct ctl_scsiio *ctsio)
5618 {
5619         struct ctl_lun *lun = CTL_LUN(ctsio);
5620         struct scsi_write_buffer *cdb;
5621         int buffer_offset, len;
5622
5623         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5624
5625         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5626
5627         len = scsi_3btoul(cdb->length);
5628         buffer_offset = scsi_3btoul(cdb->offset);
5629
5630         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5631                 ctl_set_invalid_field(ctsio,
5632                                       /*sks_valid*/ 1,
5633                                       /*command*/ 1,
5634                                       /*field*/ 6,
5635                                       /*bit_valid*/ 0,
5636                                       /*bit*/ 0);
5637                 ctl_done((union ctl_io *)ctsio);
5638                 return (CTL_RETVAL_COMPLETE);
5639         }
5640
5641         /*
5642          * If we've got a kernel request that hasn't been malloced yet,
5643          * malloc it and tell the caller the data buffer is here.
5644          */
5645         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5646                 if (lun->write_buffer == NULL) {
5647                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5648                             M_CTL, M_WAITOK);
5649                 }
5650                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5651                 ctsio->kern_data_len = len;
5652                 ctsio->kern_total_len = len;
5653                 ctsio->kern_rel_offset = 0;
5654                 ctsio->kern_sg_entries = 0;
5655                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5656                 ctsio->be_move_done = ctl_config_move_done;
5657                 ctl_datamove((union ctl_io *)ctsio);
5658
5659                 return (CTL_RETVAL_COMPLETE);
5660         }
5661
5662         ctl_set_success(ctsio);
5663         ctl_done((union ctl_io *)ctsio);
5664         return (CTL_RETVAL_COMPLETE);
5665 }
5666
5667 int
5668 ctl_write_same(struct ctl_scsiio *ctsio)
5669 {
5670         struct ctl_lun *lun = CTL_LUN(ctsio);
5671         struct ctl_lba_len_flags *lbalen;
5672         uint64_t lba;
5673         uint32_t num_blocks;
5674         int len, retval;
5675         uint8_t byte2;
5676
5677         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5678
5679         switch (ctsio->cdb[0]) {
5680         case WRITE_SAME_10: {
5681                 struct scsi_write_same_10 *cdb;
5682
5683                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5684
5685                 lba = scsi_4btoul(cdb->addr);
5686                 num_blocks = scsi_2btoul(cdb->length);
5687                 byte2 = cdb->byte2;
5688                 break;
5689         }
5690         case WRITE_SAME_16: {
5691                 struct scsi_write_same_16 *cdb;
5692
5693                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5694
5695                 lba = scsi_8btou64(cdb->addr);
5696                 num_blocks = scsi_4btoul(cdb->length);
5697                 byte2 = cdb->byte2;
5698                 break;
5699         }
5700         default:
5701                 /*
5702                  * We got a command we don't support.  This shouldn't
5703                  * happen, commands should be filtered out above us.
5704                  */
5705                 ctl_set_invalid_opcode(ctsio);
5706                 ctl_done((union ctl_io *)ctsio);
5707
5708                 return (CTL_RETVAL_COMPLETE);
5709                 break; /* NOTREACHED */
5710         }
5711
5712         /* ANCHOR flag can be used only together with UNMAP */
5713         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5714                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5715                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5716                 ctl_done((union ctl_io *)ctsio);
5717                 return (CTL_RETVAL_COMPLETE);
5718         }
5719
5720         /*
5721          * The first check is to make sure we're in bounds, the second
5722          * check is to catch wrap-around problems.  If the lba + num blocks
5723          * is less than the lba, then we've wrapped around and the block
5724          * range is invalid anyway.
5725          */
5726         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5727          || ((lba + num_blocks) < lba)) {
5728                 ctl_set_lba_out_of_range(ctsio,
5729                     MAX(lba, lun->be_lun->maxlba + 1));
5730                 ctl_done((union ctl_io *)ctsio);
5731                 return (CTL_RETVAL_COMPLETE);
5732         }
5733
5734         /* Zero number of blocks means "to the last logical block" */
5735         if (num_blocks == 0) {
5736                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5737                         ctl_set_invalid_field(ctsio,
5738                                               /*sks_valid*/ 0,
5739                                               /*command*/ 1,
5740                                               /*field*/ 0,
5741                                               /*bit_valid*/ 0,
5742                                               /*bit*/ 0);
5743                         ctl_done((union ctl_io *)ctsio);
5744                         return (CTL_RETVAL_COMPLETE);
5745                 }
5746                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5747         }
5748
5749         len = lun->be_lun->blocksize;
5750
5751         /*
5752          * If we've got a kernel request that hasn't been malloced yet,
5753          * malloc it and tell the caller the data buffer is here.
5754          */
5755         if ((byte2 & SWS_NDOB) == 0 &&
5756             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5757                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5758                 ctsio->kern_data_len = len;
5759                 ctsio->kern_total_len = len;
5760                 ctsio->kern_rel_offset = 0;
5761                 ctsio->kern_sg_entries = 0;
5762                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5763                 ctsio->be_move_done = ctl_config_move_done;
5764                 ctl_datamove((union ctl_io *)ctsio);
5765
5766                 return (CTL_RETVAL_COMPLETE);
5767         }
5768
5769         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5770         lbalen->lba = lba;
5771         lbalen->len = num_blocks;
5772         lbalen->flags = byte2;
5773         retval = lun->backend->config_write((union ctl_io *)ctsio);
5774
5775         return (retval);
5776 }
5777
5778 int
5779 ctl_unmap(struct ctl_scsiio *ctsio)
5780 {
5781         struct ctl_lun *lun = CTL_LUN(ctsio);
5782         struct scsi_unmap *cdb;
5783         struct ctl_ptr_len_flags *ptrlen;
5784         struct scsi_unmap_header *hdr;
5785         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5786         uint64_t lba;
5787         uint32_t num_blocks;
5788         int len, retval;
5789         uint8_t byte2;
5790
5791         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5792
5793         cdb = (struct scsi_unmap *)ctsio->cdb;
5794         len = scsi_2btoul(cdb->length);
5795         byte2 = cdb->byte2;
5796
5797         /*
5798          * If we've got a kernel request that hasn't been malloced yet,
5799          * malloc it and tell the caller the data buffer is here.
5800          */
5801         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5802                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5803                 ctsio->kern_data_len = len;
5804                 ctsio->kern_total_len = len;
5805                 ctsio->kern_rel_offset = 0;
5806                 ctsio->kern_sg_entries = 0;
5807                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5808                 ctsio->be_move_done = ctl_config_move_done;
5809                 ctl_datamove((union ctl_io *)ctsio);
5810
5811                 return (CTL_RETVAL_COMPLETE);
5812         }
5813
5814         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5815         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5816         if (len < sizeof (*hdr) ||
5817             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5818             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5819             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5820                 ctl_set_invalid_field(ctsio,
5821                                       /*sks_valid*/ 0,
5822                                       /*command*/ 0,
5823                                       /*field*/ 0,
5824                                       /*bit_valid*/ 0,
5825                                       /*bit*/ 0);
5826                 goto done;
5827         }
5828         len = scsi_2btoul(hdr->desc_length);
5829         buf = (struct scsi_unmap_desc *)(hdr + 1);
5830         end = buf + len / sizeof(*buf);
5831
5832         endnz = buf;
5833         for (range = buf; range < end; range++) {
5834                 lba = scsi_8btou64(range->lba);
5835                 num_blocks = scsi_4btoul(range->length);
5836                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5837                  || ((lba + num_blocks) < lba)) {
5838                         ctl_set_lba_out_of_range(ctsio,
5839                             MAX(lba, lun->be_lun->maxlba + 1));
5840                         ctl_done((union ctl_io *)ctsio);
5841                         return (CTL_RETVAL_COMPLETE);
5842                 }
5843                 if (num_blocks != 0)
5844                         endnz = range + 1;
5845         }
5846
5847         /*
5848          * Block backend can not handle zero last range.
5849          * Filter it out and return if there is nothing left.
5850          */
5851         len = (uint8_t *)endnz - (uint8_t *)buf;
5852         if (len == 0) {
5853                 ctl_set_success(ctsio);
5854                 goto done;
5855         }
5856
5857         mtx_lock(&lun->lun_lock);
5858         ptrlen = (struct ctl_ptr_len_flags *)
5859             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5860         ptrlen->ptr = (void *)buf;
5861         ptrlen->len = len;
5862         ptrlen->flags = byte2;
5863         ctl_check_blocked(lun);
5864         mtx_unlock(&lun->lun_lock);
5865
5866         retval = lun->backend->config_write((union ctl_io *)ctsio);
5867         return (retval);
5868
5869 done:
5870         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5871                 free(ctsio->kern_data_ptr, M_CTL);
5872                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5873         }
5874         ctl_done((union ctl_io *)ctsio);
5875         return (CTL_RETVAL_COMPLETE);
5876 }
5877
5878 int
5879 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5880                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5881 {
5882         struct ctl_lun *lun = CTL_LUN(ctsio);
5883         uint8_t *current_cp;
5884         int set_ua;
5885         uint32_t initidx;
5886
5887         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5888         set_ua = 0;
5889
5890         current_cp = (page_index->page_data + (page_index->page_len *
5891             CTL_PAGE_CURRENT));
5892
5893         mtx_lock(&lun->lun_lock);
5894         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5895                 memcpy(current_cp, page_ptr, page_index->page_len);
5896                 set_ua = 1;
5897         }
5898         if (set_ua != 0)
5899                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5900         mtx_unlock(&lun->lun_lock);
5901         if (set_ua) {
5902                 ctl_isc_announce_mode(lun,
5903                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5904                     page_index->page_code, page_index->subpage);
5905         }
5906         return (CTL_RETVAL_COMPLETE);
5907 }
5908
5909 static void
5910 ctl_ie_timer(void *arg)
5911 {
5912         struct ctl_lun *lun = arg;
5913         uint64_t t;
5914
5915         if (lun->ie_asc == 0)
5916                 return;
5917
5918         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5919                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5920         else
5921                 lun->ie_reported = 0;
5922
5923         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5924                 lun->ie_reportcnt++;
5925                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5926                 if (t == 0 || t == UINT32_MAX)
5927                         t = 3000;  /* 5 min */
5928                 callout_schedule(&lun->ie_callout, t * hz / 10);
5929         }
5930 }
5931
5932 int
5933 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5934                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5935 {
5936         struct ctl_lun *lun = CTL_LUN(ctsio);
5937         struct scsi_info_exceptions_page *pg;
5938         uint64_t t;
5939
5940         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5941
5942         pg = (struct scsi_info_exceptions_page *)page_ptr;
5943         mtx_lock(&lun->lun_lock);
5944         if (pg->info_flags & SIEP_FLAGS_TEST) {
5945                 lun->ie_asc = 0x5d;
5946                 lun->ie_ascq = 0xff;
5947                 if (pg->mrie == SIEP_MRIE_UA) {
5948                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
5949                         lun->ie_reported = 1;
5950                 } else {
5951                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5952                         lun->ie_reported = -1;
5953                 }
5954                 lun->ie_reportcnt = 1;
5955                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
5956                         lun->ie_reportcnt++;
5957                         t = scsi_4btoul(pg->interval_timer);
5958                         if (t == 0 || t == UINT32_MAX)
5959                                 t = 3000;  /* 5 min */
5960                         callout_reset(&lun->ie_callout, t * hz / 10,
5961                             ctl_ie_timer, lun);
5962                 }
5963         } else {
5964                 lun->ie_asc = 0;
5965                 lun->ie_ascq = 0;
5966                 lun->ie_reported = 1;
5967                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5968                 lun->ie_reportcnt = UINT32_MAX;
5969                 callout_stop(&lun->ie_callout);
5970         }
5971         mtx_unlock(&lun->lun_lock);
5972         return (CTL_RETVAL_COMPLETE);
5973 }
5974
5975 static int
5976 ctl_do_mode_select(union ctl_io *io)
5977 {
5978         struct ctl_lun *lun = CTL_LUN(io);
5979         struct scsi_mode_page_header *page_header;
5980         struct ctl_page_index *page_index;
5981         struct ctl_scsiio *ctsio;
5982         int page_len, page_len_offset, page_len_size;
5983         union ctl_modepage_info *modepage_info;
5984         uint16_t *len_left, *len_used;
5985         int retval, i;
5986
5987         ctsio = &io->scsiio;
5988         page_index = NULL;
5989         page_len = 0;
5990
5991         modepage_info = (union ctl_modepage_info *)
5992                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5993         len_left = &modepage_info->header.len_left;
5994         len_used = &modepage_info->header.len_used;
5995
5996 do_next_page:
5997
5998         page_header = (struct scsi_mode_page_header *)
5999                 (ctsio->kern_data_ptr + *len_used);
6000
6001         if (*len_left == 0) {
6002                 free(ctsio->kern_data_ptr, M_CTL);
6003                 ctl_set_success(ctsio);
6004                 ctl_done((union ctl_io *)ctsio);
6005                 return (CTL_RETVAL_COMPLETE);
6006         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6007
6008                 free(ctsio->kern_data_ptr, M_CTL);
6009                 ctl_set_param_len_error(ctsio);
6010                 ctl_done((union ctl_io *)ctsio);
6011                 return (CTL_RETVAL_COMPLETE);
6012
6013         } else if ((page_header->page_code & SMPH_SPF)
6014                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6015
6016                 free(ctsio->kern_data_ptr, M_CTL);
6017                 ctl_set_param_len_error(ctsio);
6018                 ctl_done((union ctl_io *)ctsio);
6019                 return (CTL_RETVAL_COMPLETE);
6020         }
6021
6022
6023         /*
6024          * XXX KDM should we do something with the block descriptor?
6025          */
6026         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6027                 page_index = &lun->mode_pages.index[i];
6028                 if (lun->be_lun->lun_type == T_DIRECT &&
6029                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6030                         continue;
6031                 if (lun->be_lun->lun_type == T_PROCESSOR &&
6032                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6033                         continue;
6034                 if (lun->be_lun->lun_type == T_CDROM &&
6035                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6036                         continue;
6037
6038                 if ((page_index->page_code & SMPH_PC_MASK) !=
6039                     (page_header->page_code & SMPH_PC_MASK))
6040                         continue;
6041
6042                 /*
6043                  * If neither page has a subpage code, then we've got a
6044                  * match.
6045                  */
6046                 if (((page_index->page_code & SMPH_SPF) == 0)
6047                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6048                         page_len = page_header->page_length;
6049                         break;
6050                 }
6051
6052                 /*
6053                  * If both pages have subpages, then the subpage numbers
6054                  * have to match.
6055                  */
6056                 if ((page_index->page_code & SMPH_SPF)
6057                   && (page_header->page_code & SMPH_SPF)) {
6058                         struct scsi_mode_page_header_sp *sph;
6059
6060                         sph = (struct scsi_mode_page_header_sp *)page_header;
6061                         if (page_index->subpage == sph->subpage) {
6062                                 page_len = scsi_2btoul(sph->page_length);
6063                                 break;
6064                         }
6065                 }
6066         }
6067
6068         /*
6069          * If we couldn't find the page, or if we don't have a mode select
6070          * handler for it, send back an error to the user.
6071          */
6072         if ((i >= CTL_NUM_MODE_PAGES)
6073          || (page_index->select_handler == NULL)) {
6074                 ctl_set_invalid_field(ctsio,
6075                                       /*sks_valid*/ 1,
6076                                       /*command*/ 0,
6077                                       /*field*/ *len_used,
6078                                       /*bit_valid*/ 0,
6079                                       /*bit*/ 0);
6080                 free(ctsio->kern_data_ptr, M_CTL);
6081                 ctl_done((union ctl_io *)ctsio);
6082                 return (CTL_RETVAL_COMPLETE);
6083         }
6084
6085         if (page_index->page_code & SMPH_SPF) {
6086                 page_len_offset = 2;
6087                 page_len_size = 2;
6088         } else {
6089                 page_len_size = 1;
6090                 page_len_offset = 1;
6091         }
6092
6093         /*
6094          * If the length the initiator gives us isn't the one we specify in
6095          * the mode page header, or if they didn't specify enough data in
6096          * the CDB to avoid truncating this page, kick out the request.
6097          */
6098         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6099                 ctl_set_invalid_field(ctsio,
6100                                       /*sks_valid*/ 1,
6101                                       /*command*/ 0,
6102                                       /*field*/ *len_used + page_len_offset,
6103                                       /*bit_valid*/ 0,
6104                                       /*bit*/ 0);
6105                 free(ctsio->kern_data_ptr, M_CTL);
6106                 ctl_done((union ctl_io *)ctsio);
6107                 return (CTL_RETVAL_COMPLETE);
6108         }
6109         if (*len_left < page_index->page_len) {
6110                 free(ctsio->kern_data_ptr, M_CTL);
6111                 ctl_set_param_len_error(ctsio);
6112                 ctl_done((union ctl_io *)ctsio);
6113                 return (CTL_RETVAL_COMPLETE);
6114         }
6115
6116         /*
6117          * Run through the mode page, checking to make sure that the bits
6118          * the user changed are actually legal for him to change.
6119          */
6120         for (i = 0; i < page_index->page_len; i++) {
6121                 uint8_t *user_byte, *change_mask, *current_byte;
6122                 int bad_bit;
6123                 int j;
6124
6125                 user_byte = (uint8_t *)page_header + i;
6126                 change_mask = page_index->page_data +
6127                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6128                 current_byte = page_index->page_data +
6129                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6130
6131                 /*
6132                  * Check to see whether the user set any bits in this byte
6133                  * that he is not allowed to set.
6134                  */
6135                 if ((*user_byte & ~(*change_mask)) ==
6136                     (*current_byte & ~(*change_mask)))
6137                         continue;
6138
6139                 /*
6140                  * Go through bit by bit to determine which one is illegal.
6141                  */
6142                 bad_bit = 0;
6143                 for (j = 7; j >= 0; j--) {
6144                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6145                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6146                                 bad_bit = i;
6147                                 break;
6148                         }
6149                 }
6150                 ctl_set_invalid_field(ctsio,
6151                                       /*sks_valid*/ 1,
6152                                       /*command*/ 0,
6153                                       /*field*/ *len_used + i,
6154                                       /*bit_valid*/ 1,
6155                                       /*bit*/ bad_bit);
6156                 free(ctsio->kern_data_ptr, M_CTL);
6157                 ctl_done((union ctl_io *)ctsio);
6158                 return (CTL_RETVAL_COMPLETE);
6159         }
6160
6161         /*
6162          * Decrement these before we call the page handler, since we may
6163          * end up getting called back one way or another before the handler
6164          * returns to this context.
6165          */
6166         *len_left -= page_index->page_len;
6167         *len_used += page_index->page_len;
6168
6169         retval = page_index->select_handler(ctsio, page_index,
6170                                             (uint8_t *)page_header);
6171
6172         /*
6173          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6174          * wait until this queued command completes to finish processing
6175          * the mode page.  If it returns anything other than
6176          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6177          * already set the sense information, freed the data pointer, and
6178          * completed the io for us.
6179          */
6180         if (retval != CTL_RETVAL_COMPLETE)
6181                 goto bailout_no_done;
6182
6183         /*
6184          * If the initiator sent us more than one page, parse the next one.
6185          */
6186         if (*len_left > 0)
6187                 goto do_next_page;
6188
6189         ctl_set_success(ctsio);
6190         free(ctsio->kern_data_ptr, M_CTL);
6191         ctl_done((union ctl_io *)ctsio);
6192
6193 bailout_no_done:
6194
6195         return (CTL_RETVAL_COMPLETE);
6196
6197 }
6198
6199 int
6200 ctl_mode_select(struct ctl_scsiio *ctsio)
6201 {
6202         struct ctl_lun *lun = CTL_LUN(ctsio);
6203         union ctl_modepage_info *modepage_info;
6204         int bd_len, i, header_size, param_len, pf, rtd, sp;
6205         uint32_t initidx;
6206
6207         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6208         switch (ctsio->cdb[0]) {
6209         case MODE_SELECT_6: {
6210                 struct scsi_mode_select_6 *cdb;
6211
6212                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6213
6214                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6215                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6216                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6217                 param_len = cdb->length;
6218                 header_size = sizeof(struct scsi_mode_header_6);
6219                 break;
6220         }
6221         case MODE_SELECT_10: {
6222                 struct scsi_mode_select_10 *cdb;
6223
6224                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6225
6226                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6227                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6228                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6229                 param_len = scsi_2btoul(cdb->length);
6230                 header_size = sizeof(struct scsi_mode_header_10);
6231                 break;
6232         }
6233         default:
6234                 ctl_set_invalid_opcode(ctsio);
6235                 ctl_done((union ctl_io *)ctsio);
6236                 return (CTL_RETVAL_COMPLETE);
6237         }
6238
6239         if (rtd) {
6240                 if (param_len != 0) {
6241                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 0,
6242                             /*command*/ 1, /*field*/ 0,
6243                             /*bit_valid*/ 0, /*bit*/ 0);
6244                         ctl_done((union ctl_io *)ctsio);
6245                         return (CTL_RETVAL_COMPLETE);
6246                 }
6247
6248                 /* Revert to defaults. */
6249                 ctl_init_page_index(lun);
6250                 mtx_lock(&lun->lun_lock);
6251                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6252                 mtx_unlock(&lun->lun_lock);
6253                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6254                         ctl_isc_announce_mode(lun, -1,
6255                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
6256                             lun->mode_pages.index[i].subpage);
6257                 }
6258                 ctl_set_success(ctsio);
6259                 ctl_done((union ctl_io *)ctsio);
6260                 return (CTL_RETVAL_COMPLETE);
6261         }
6262
6263         /*
6264          * From SPC-3:
6265          * "A parameter list length of zero indicates that the Data-Out Buffer
6266          * shall be empty. This condition shall not be considered as an error."
6267          */
6268         if (param_len == 0) {
6269                 ctl_set_success(ctsio);
6270                 ctl_done((union ctl_io *)ctsio);
6271                 return (CTL_RETVAL_COMPLETE);
6272         }
6273
6274         /*
6275          * Since we'll hit this the first time through, prior to
6276          * allocation, we don't need to free a data buffer here.
6277          */
6278         if (param_len < header_size) {
6279                 ctl_set_param_len_error(ctsio);
6280                 ctl_done((union ctl_io *)ctsio);
6281                 return (CTL_RETVAL_COMPLETE);
6282         }
6283
6284         /*
6285          * Allocate the data buffer and grab the user's data.  In theory,
6286          * we shouldn't have to sanity check the parameter list length here
6287          * because the maximum size is 64K.  We should be able to malloc
6288          * that much without too many problems.
6289          */
6290         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6291                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6292                 ctsio->kern_data_len = param_len;
6293                 ctsio->kern_total_len = param_len;
6294                 ctsio->kern_rel_offset = 0;
6295                 ctsio->kern_sg_entries = 0;
6296                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6297                 ctsio->be_move_done = ctl_config_move_done;
6298                 ctl_datamove((union ctl_io *)ctsio);
6299
6300                 return (CTL_RETVAL_COMPLETE);
6301         }
6302
6303         switch (ctsio->cdb[0]) {
6304         case MODE_SELECT_6: {
6305                 struct scsi_mode_header_6 *mh6;
6306
6307                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6308                 bd_len = mh6->blk_desc_len;
6309                 break;
6310         }
6311         case MODE_SELECT_10: {
6312                 struct scsi_mode_header_10 *mh10;
6313
6314                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6315                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6316                 break;
6317         }
6318         default:
6319                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6320         }
6321
6322         if (param_len < (header_size + bd_len)) {
6323                 free(ctsio->kern_data_ptr, M_CTL);
6324                 ctl_set_param_len_error(ctsio);
6325                 ctl_done((union ctl_io *)ctsio);
6326                 return (CTL_RETVAL_COMPLETE);
6327         }
6328
6329         /*
6330          * Set the IO_CONT flag, so that if this I/O gets passed to
6331          * ctl_config_write_done(), it'll get passed back to
6332          * ctl_do_mode_select() for further processing, or completion if
6333          * we're all done.
6334          */
6335         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6336         ctsio->io_cont = ctl_do_mode_select;
6337
6338         modepage_info = (union ctl_modepage_info *)
6339                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6340         memset(modepage_info, 0, sizeof(*modepage_info));
6341         modepage_info->header.len_left = param_len - header_size - bd_len;
6342         modepage_info->header.len_used = header_size + bd_len;
6343
6344         return (ctl_do_mode_select((union ctl_io *)ctsio));
6345 }
6346
6347 int
6348 ctl_mode_sense(struct ctl_scsiio *ctsio)
6349 {
6350         struct ctl_lun *lun = CTL_LUN(ctsio);
6351         int pc, page_code, dbd, llba, subpage;
6352         int alloc_len, page_len, header_len, total_len;
6353         struct scsi_mode_block_descr *block_desc;
6354         struct ctl_page_index *page_index;
6355
6356         dbd = 0;
6357         llba = 0;
6358         block_desc = NULL;
6359
6360         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6361
6362         switch (ctsio->cdb[0]) {
6363         case MODE_SENSE_6: {
6364                 struct scsi_mode_sense_6 *cdb;
6365
6366                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6367
6368                 header_len = sizeof(struct scsi_mode_hdr_6);
6369                 if (cdb->byte2 & SMS_DBD)
6370                         dbd = 1;
6371                 else
6372                         header_len += sizeof(struct scsi_mode_block_descr);
6373
6374                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6375                 page_code = cdb->page & SMS_PAGE_CODE;
6376                 subpage = cdb->subpage;
6377                 alloc_len = cdb->length;
6378                 break;
6379         }
6380         case MODE_SENSE_10: {
6381                 struct scsi_mode_sense_10 *cdb;
6382
6383                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6384
6385                 header_len = sizeof(struct scsi_mode_hdr_10);
6386
6387                 if (cdb->byte2 & SMS_DBD)
6388                         dbd = 1;
6389                 else
6390                         header_len += sizeof(struct scsi_mode_block_descr);
6391                 if (cdb->byte2 & SMS10_LLBAA)
6392                         llba = 1;
6393                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6394                 page_code = cdb->page & SMS_PAGE_CODE;
6395                 subpage = cdb->subpage;
6396                 alloc_len = scsi_2btoul(cdb->length);
6397                 break;
6398         }
6399         default:
6400                 ctl_set_invalid_opcode(ctsio);
6401                 ctl_done((union ctl_io *)ctsio);
6402                 return (CTL_RETVAL_COMPLETE);
6403                 break; /* NOTREACHED */
6404         }
6405
6406         /*
6407          * We have to make a first pass through to calculate the size of
6408          * the pages that match the user's query.  Then we allocate enough
6409          * memory to hold it, and actually copy the data into the buffer.
6410          */
6411         switch (page_code) {
6412         case SMS_ALL_PAGES_PAGE: {
6413                 u_int i;
6414
6415                 page_len = 0;
6416
6417                 /*
6418                  * At the moment, values other than 0 and 0xff here are
6419                  * reserved according to SPC-3.
6420                  */
6421                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6422                  && (subpage != SMS_SUBPAGE_ALL)) {
6423                         ctl_set_invalid_field(ctsio,
6424                                               /*sks_valid*/ 1,
6425                                               /*command*/ 1,
6426                                               /*field*/ 3,
6427                                               /*bit_valid*/ 0,
6428                                               /*bit*/ 0);
6429                         ctl_done((union ctl_io *)ctsio);
6430                         return (CTL_RETVAL_COMPLETE);
6431                 }
6432
6433                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6434                         page_index = &lun->mode_pages.index[i];
6435
6436                         /* Make sure the page is supported for this dev type */
6437                         if (lun->be_lun->lun_type == T_DIRECT &&
6438                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6439                                 continue;
6440                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6441                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6442                                 continue;
6443                         if (lun->be_lun->lun_type == T_CDROM &&
6444                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6445                                 continue;
6446
6447                         /*
6448                          * We don't use this subpage if the user didn't
6449                          * request all subpages.
6450                          */
6451                         if ((page_index->subpage != 0)
6452                          && (subpage == SMS_SUBPAGE_PAGE_0))
6453                                 continue;
6454
6455 #if 0
6456                         printf("found page %#x len %d\n",
6457                                page_index->page_code & SMPH_PC_MASK,
6458                                page_index->page_len);
6459 #endif
6460                         page_len += page_index->page_len;
6461                 }
6462                 break;
6463         }
6464         default: {
6465                 u_int i;
6466
6467                 page_len = 0;
6468
6469                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6470                         page_index = &lun->mode_pages.index[i];
6471
6472                         /* Make sure the page is supported for this dev type */
6473                         if (lun->be_lun->lun_type == T_DIRECT &&
6474                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6475                                 continue;
6476                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6477                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6478                                 continue;
6479                         if (lun->be_lun->lun_type == T_CDROM &&
6480                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6481                                 continue;
6482
6483                         /* Look for the right page code */
6484                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6485                                 continue;
6486
6487                         /* Look for the right subpage or the subpage wildcard*/
6488                         if ((page_index->subpage != subpage)
6489                          && (subpage != SMS_SUBPAGE_ALL))
6490                                 continue;
6491
6492 #if 0
6493                         printf("found page %#x len %d\n",
6494                                page_index->page_code & SMPH_PC_MASK,
6495                                page_index->page_len);
6496 #endif
6497
6498                         page_len += page_index->page_len;
6499                 }
6500
6501                 if (page_len == 0) {
6502                         ctl_set_invalid_field(ctsio,
6503                                               /*sks_valid*/ 1,
6504                                               /*command*/ 1,
6505                                               /*field*/ 2,
6506                                               /*bit_valid*/ 1,
6507                                               /*bit*/ 5);
6508                         ctl_done((union ctl_io *)ctsio);
6509                         return (CTL_RETVAL_COMPLETE);
6510                 }
6511                 break;
6512         }
6513         }
6514
6515         total_len = header_len + page_len;
6516 #if 0
6517         printf("header_len = %d, page_len = %d, total_len = %d\n",
6518                header_len, page_len, total_len);
6519 #endif
6520
6521         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6522         ctsio->kern_sg_entries = 0;
6523         ctsio->kern_rel_offset = 0;
6524         ctsio->kern_data_len = min(total_len, alloc_len);
6525         ctsio->kern_total_len = ctsio->kern_data_len;
6526
6527         switch (ctsio->cdb[0]) {
6528         case MODE_SENSE_6: {
6529                 struct scsi_mode_hdr_6 *header;
6530
6531                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6532
6533                 header->datalen = MIN(total_len - 1, 254);
6534                 if (lun->be_lun->lun_type == T_DIRECT) {
6535                         header->dev_specific = 0x10; /* DPOFUA */
6536                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6537                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6538                                 header->dev_specific |= 0x80; /* WP */
6539                 }
6540                 if (dbd)
6541                         header->block_descr_len = 0;
6542                 else
6543                         header->block_descr_len =
6544                                 sizeof(struct scsi_mode_block_descr);
6545                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6546                 break;
6547         }
6548         case MODE_SENSE_10: {
6549                 struct scsi_mode_hdr_10 *header;
6550                 int datalen;
6551
6552                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6553
6554                 datalen = MIN(total_len - 2, 65533);
6555                 scsi_ulto2b(datalen, header->datalen);
6556                 if (lun->be_lun->lun_type == T_DIRECT) {
6557                         header->dev_specific = 0x10; /* DPOFUA */
6558                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6559                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6560                                 header->dev_specific |= 0x80; /* WP */
6561                 }
6562                 if (dbd)
6563                         scsi_ulto2b(0, header->block_descr_len);
6564                 else
6565                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6566                                     header->block_descr_len);
6567                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6568                 break;
6569         }
6570         default:
6571                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6572         }
6573
6574         /*
6575          * If we've got a disk, use its blocksize in the block
6576          * descriptor.  Otherwise, just set it to 0.
6577          */
6578         if (dbd == 0) {
6579                 if (lun->be_lun->lun_type == T_DIRECT)
6580                         scsi_ulto3b(lun->be_lun->blocksize,
6581                                     block_desc->block_len);
6582                 else
6583                         scsi_ulto3b(0, block_desc->block_len);
6584         }
6585
6586         switch (page_code) {
6587         case SMS_ALL_PAGES_PAGE: {
6588                 int i, data_used;
6589
6590                 data_used = header_len;
6591                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6592                         struct ctl_page_index *page_index;
6593
6594                         page_index = &lun->mode_pages.index[i];
6595                         if (lun->be_lun->lun_type == T_DIRECT &&
6596                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6597                                 continue;
6598                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6599                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6600                                 continue;
6601                         if (lun->be_lun->lun_type == T_CDROM &&
6602                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6603                                 continue;
6604
6605                         /*
6606                          * We don't use this subpage if the user didn't
6607                          * request all subpages.  We already checked (above)
6608                          * to make sure the user only specified a subpage
6609                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6610                          */
6611                         if ((page_index->subpage != 0)
6612                          && (subpage == SMS_SUBPAGE_PAGE_0))
6613                                 continue;
6614
6615                         /*
6616                          * Call the handler, if it exists, to update the
6617                          * page to the latest values.
6618                          */
6619                         if (page_index->sense_handler != NULL)
6620                                 page_index->sense_handler(ctsio, page_index,pc);
6621
6622                         memcpy(ctsio->kern_data_ptr + data_used,
6623                                page_index->page_data +
6624                                (page_index->page_len * pc),
6625                                page_index->page_len);
6626                         data_used += page_index->page_len;
6627                 }
6628                 break;
6629         }
6630         default: {
6631                 int i, data_used;
6632
6633                 data_used = header_len;
6634
6635                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6636                         struct ctl_page_index *page_index;
6637
6638                         page_index = &lun->mode_pages.index[i];
6639
6640                         /* Look for the right page code */
6641                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6642                                 continue;
6643
6644                         /* Look for the right subpage or the subpage wildcard*/
6645                         if ((page_index->subpage != subpage)
6646                          && (subpage != SMS_SUBPAGE_ALL))
6647                                 continue;
6648
6649                         /* Make sure the page is supported for this dev type */
6650                         if (lun->be_lun->lun_type == T_DIRECT &&
6651                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6652                                 continue;
6653                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6654                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6655                                 continue;
6656                         if (lun->be_lun->lun_type == T_CDROM &&
6657                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6658                                 continue;
6659
6660                         /*
6661                          * Call the handler, if it exists, to update the
6662                          * page to the latest values.
6663                          */
6664                         if (page_index->sense_handler != NULL)
6665                                 page_index->sense_handler(ctsio, page_index,pc);
6666
6667                         memcpy(ctsio->kern_data_ptr + data_used,
6668                                page_index->page_data +
6669                                (page_index->page_len * pc),
6670                                page_index->page_len);
6671                         data_used += page_index->page_len;
6672                 }
6673                 break;
6674         }
6675         }
6676
6677         ctl_set_success(ctsio);
6678         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6679         ctsio->be_move_done = ctl_config_move_done;
6680         ctl_datamove((union ctl_io *)ctsio);
6681         return (CTL_RETVAL_COMPLETE);
6682 }
6683
6684 int
6685 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6686                                struct ctl_page_index *page_index,
6687                                int pc)
6688 {
6689         struct ctl_lun *lun = CTL_LUN(ctsio);
6690         struct scsi_log_param_header *phdr;
6691         uint8_t *data;
6692         uint64_t val;
6693
6694         data = page_index->page_data;
6695
6696         if (lun->backend->lun_attr != NULL &&
6697             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6698              != UINT64_MAX) {
6699                 phdr = (struct scsi_log_param_header *)data;
6700                 scsi_ulto2b(0x0001, phdr->param_code);
6701                 phdr->param_control = SLP_LBIN | SLP_LP;
6702                 phdr->param_len = 8;
6703                 data = (uint8_t *)(phdr + 1);
6704                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6705                 data[4] = 0x02; /* per-pool */
6706                 data += phdr->param_len;
6707         }
6708
6709         if (lun->backend->lun_attr != NULL &&
6710             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6711              != UINT64_MAX) {
6712                 phdr = (struct scsi_log_param_header *)data;
6713                 scsi_ulto2b(0x0002, phdr->param_code);
6714                 phdr->param_control = SLP_LBIN | SLP_LP;
6715                 phdr->param_len = 8;
6716                 data = (uint8_t *)(phdr + 1);
6717                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6718                 data[4] = 0x01; /* per-LUN */
6719                 data += phdr->param_len;
6720         }
6721
6722         if (lun->backend->lun_attr != NULL &&
6723             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6724              != UINT64_MAX) {
6725                 phdr = (struct scsi_log_param_header *)data;
6726                 scsi_ulto2b(0x00f1, phdr->param_code);
6727                 phdr->param_control = SLP_LBIN | SLP_LP;
6728                 phdr->param_len = 8;
6729                 data = (uint8_t *)(phdr + 1);
6730                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6731                 data[4] = 0x02; /* per-pool */
6732                 data += phdr->param_len;
6733         }
6734
6735         if (lun->backend->lun_attr != NULL &&
6736             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6737              != UINT64_MAX) {
6738                 phdr = (struct scsi_log_param_header *)data;
6739                 scsi_ulto2b(0x00f2, phdr->param_code);
6740                 phdr->param_control = SLP_LBIN | SLP_LP;
6741                 phdr->param_len = 8;
6742                 data = (uint8_t *)(phdr + 1);
6743                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6744                 data[4] = 0x02; /* per-pool */
6745                 data += phdr->param_len;
6746         }
6747
6748         page_index->page_len = data - page_index->page_data;
6749         return (0);
6750 }
6751
6752 int
6753 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6754                                struct ctl_page_index *page_index,
6755                                int pc)
6756 {
6757         struct ctl_lun *lun = CTL_LUN(ctsio);
6758         struct stat_page *data;
6759         struct bintime *t;
6760
6761         data = (struct stat_page *)page_index->page_data;
6762
6763         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6764         data->sap.hdr.param_control = SLP_LBIN;
6765         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6766             sizeof(struct scsi_log_param_header);
6767         scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
6768             data->sap.read_num);
6769         scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
6770             data->sap.write_num);
6771         if (lun->be_lun->blocksize > 0) {
6772                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
6773                     lun->be_lun->blocksize, data->sap.recvieved_lba);
6774                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
6775                     lun->be_lun->blocksize, data->sap.transmitted_lba);
6776         }
6777         t = &lun->stats.time[CTL_STATS_READ];
6778         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6779             data->sap.read_int);
6780         t = &lun->stats.time[CTL_STATS_WRITE];
6781         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6782             data->sap.write_int);
6783         scsi_u64to8b(0, data->sap.weighted_num);
6784         scsi_u64to8b(0, data->sap.weighted_int);
6785         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6786         data->it.hdr.param_control = SLP_LBIN;
6787         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6788             sizeof(struct scsi_log_param_header);
6789 #ifdef CTL_TIME_IO
6790         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6791 #endif
6792         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6793         data->it.hdr.param_control = SLP_LBIN;
6794         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6795             sizeof(struct scsi_log_param_header);
6796         scsi_ulto4b(3, data->ti.exponent);
6797         scsi_ulto4b(1, data->ti.integer);
6798         return (0);
6799 }
6800
6801 int
6802 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6803                                struct ctl_page_index *page_index,
6804                                int pc)
6805 {
6806         struct ctl_lun *lun = CTL_LUN(ctsio);
6807         struct scsi_log_informational_exceptions *data;
6808
6809         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6810
6811         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6812         data->hdr.param_control = SLP_LBIN;
6813         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6814             sizeof(struct scsi_log_param_header);
6815         data->ie_asc = lun->ie_asc;
6816         data->ie_ascq = lun->ie_ascq;
6817         data->temperature = 0xff;
6818         return (0);
6819 }
6820
6821 int
6822 ctl_log_sense(struct ctl_scsiio *ctsio)
6823 {
6824         struct ctl_lun *lun = CTL_LUN(ctsio);
6825         int i, pc, page_code, subpage;
6826         int alloc_len, total_len;
6827         struct ctl_page_index *page_index;
6828         struct scsi_log_sense *cdb;
6829         struct scsi_log_header *header;
6830
6831         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6832
6833         cdb = (struct scsi_log_sense *)ctsio->cdb;
6834         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6835         page_code = cdb->page & SLS_PAGE_CODE;
6836         subpage = cdb->subpage;
6837         alloc_len = scsi_2btoul(cdb->length);
6838
6839         page_index = NULL;
6840         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6841                 page_index = &lun->log_pages.index[i];
6842
6843                 /* Look for the right page code */
6844                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6845                         continue;
6846
6847                 /* Look for the right subpage or the subpage wildcard*/
6848                 if (page_index->subpage != subpage)
6849                         continue;
6850
6851                 break;
6852         }
6853         if (i >= CTL_NUM_LOG_PAGES) {
6854                 ctl_set_invalid_field(ctsio,
6855                                       /*sks_valid*/ 1,
6856                                       /*command*/ 1,
6857                                       /*field*/ 2,
6858                                       /*bit_valid*/ 0,
6859                                       /*bit*/ 0);
6860                 ctl_done((union ctl_io *)ctsio);
6861                 return (CTL_RETVAL_COMPLETE);
6862         }
6863
6864         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6865
6866         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6867         ctsio->kern_sg_entries = 0;
6868         ctsio->kern_rel_offset = 0;
6869         ctsio->kern_data_len = min(total_len, alloc_len);
6870         ctsio->kern_total_len = ctsio->kern_data_len;
6871
6872         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6873         header->page = page_index->page_code;
6874         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6875                 header->page |= SL_DS;
6876         if (page_index->subpage) {
6877                 header->page |= SL_SPF;
6878                 header->subpage = page_index->subpage;
6879         }
6880         scsi_ulto2b(page_index->page_len, header->datalen);
6881
6882         /*
6883          * Call the handler, if it exists, to update the
6884          * page to the latest values.
6885          */
6886         if (page_index->sense_handler != NULL)
6887                 page_index->sense_handler(ctsio, page_index, pc);
6888
6889         memcpy(header + 1, page_index->page_data, page_index->page_len);
6890
6891         ctl_set_success(ctsio);
6892         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6893         ctsio->be_move_done = ctl_config_move_done;
6894         ctl_datamove((union ctl_io *)ctsio);
6895         return (CTL_RETVAL_COMPLETE);
6896 }
6897
6898 int
6899 ctl_read_capacity(struct ctl_scsiio *ctsio)
6900 {
6901         struct ctl_lun *lun = CTL_LUN(ctsio);
6902         struct scsi_read_capacity *cdb;
6903         struct scsi_read_capacity_data *data;
6904         uint32_t lba;
6905
6906         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6907
6908         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6909
6910         lba = scsi_4btoul(cdb->addr);
6911         if (((cdb->pmi & SRC_PMI) == 0)
6912          && (lba != 0)) {
6913                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6914                                       /*sks_valid*/ 1,
6915                                       /*command*/ 1,
6916                                       /*field*/ 2,
6917                                       /*bit_valid*/ 0,
6918                                       /*bit*/ 0);
6919                 ctl_done((union ctl_io *)ctsio);
6920                 return (CTL_RETVAL_COMPLETE);
6921         }
6922
6923         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6924         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6925         ctsio->kern_data_len = sizeof(*data);
6926         ctsio->kern_total_len = sizeof(*data);
6927         ctsio->kern_rel_offset = 0;
6928         ctsio->kern_sg_entries = 0;
6929
6930         /*
6931          * If the maximum LBA is greater than 0xfffffffe, the user must
6932          * issue a SERVICE ACTION IN (16) command, with the read capacity
6933          * serivce action set.
6934          */
6935         if (lun->be_lun->maxlba > 0xfffffffe)
6936                 scsi_ulto4b(0xffffffff, data->addr);
6937         else
6938                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6939
6940         /*
6941          * XXX KDM this may not be 512 bytes...
6942          */
6943         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6944
6945         ctl_set_success(ctsio);
6946         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6947         ctsio->be_move_done = ctl_config_move_done;
6948         ctl_datamove((union ctl_io *)ctsio);
6949         return (CTL_RETVAL_COMPLETE);
6950 }
6951
6952 int
6953 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6954 {
6955         struct ctl_lun *lun = CTL_LUN(ctsio);
6956         struct scsi_read_capacity_16 *cdb;
6957         struct scsi_read_capacity_data_long *data;
6958         uint64_t lba;
6959         uint32_t alloc_len;
6960
6961         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6962
6963         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6964
6965         alloc_len = scsi_4btoul(cdb->alloc_len);
6966         lba = scsi_8btou64(cdb->addr);
6967
6968         if ((cdb->reladr & SRC16_PMI)
6969          && (lba != 0)) {
6970                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6971                                       /*sks_valid*/ 1,
6972                                       /*command*/ 1,
6973                                       /*field*/ 2,
6974                                       /*bit_valid*/ 0,
6975                                       /*bit*/ 0);
6976                 ctl_done((union ctl_io *)ctsio);
6977                 return (CTL_RETVAL_COMPLETE);
6978         }
6979
6980         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6981         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6982         ctsio->kern_rel_offset = 0;
6983         ctsio->kern_sg_entries = 0;
6984         ctsio->kern_data_len = min(sizeof(*data), alloc_len);
6985         ctsio->kern_total_len = ctsio->kern_data_len;
6986
6987         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6988         /* XXX KDM this may not be 512 bytes... */
6989         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6990         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6991         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6992         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6993                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6994
6995         ctl_set_success(ctsio);
6996         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6997         ctsio->be_move_done = ctl_config_move_done;
6998         ctl_datamove((union ctl_io *)ctsio);
6999         return (CTL_RETVAL_COMPLETE);
7000 }
7001
7002 int
7003 ctl_get_lba_status(struct ctl_scsiio *ctsio)
7004 {
7005         struct ctl_lun *lun = CTL_LUN(ctsio);
7006         struct scsi_get_lba_status *cdb;
7007         struct scsi_get_lba_status_data *data;
7008         struct ctl_lba_len_flags *lbalen;
7009         uint64_t lba;
7010         uint32_t alloc_len, total_len;
7011         int retval;
7012
7013         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7014
7015         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7016         lba = scsi_8btou64(cdb->addr);
7017         alloc_len = scsi_4btoul(cdb->alloc_len);
7018
7019         if (lba > lun->be_lun->maxlba) {
7020                 ctl_set_lba_out_of_range(ctsio, lba);
7021                 ctl_done((union ctl_io *)ctsio);
7022                 return (CTL_RETVAL_COMPLETE);
7023         }
7024
7025         total_len = sizeof(*data) + sizeof(data->descr[0]);
7026         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7027         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7028         ctsio->kern_rel_offset = 0;
7029         ctsio->kern_sg_entries = 0;
7030         ctsio->kern_data_len = min(total_len, alloc_len);
7031         ctsio->kern_total_len = ctsio->kern_data_len;
7032
7033         /* Fill dummy data in case backend can't tell anything. */
7034         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7035         scsi_u64to8b(lba, data->descr[0].addr);
7036         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7037             data->descr[0].length);
7038         data->descr[0].status = 0; /* Mapped or unknown. */
7039
7040         ctl_set_success(ctsio);
7041         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7042         ctsio->be_move_done = ctl_config_move_done;
7043
7044         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7045         lbalen->lba = lba;
7046         lbalen->len = total_len;
7047         lbalen->flags = 0;
7048         retval = lun->backend->config_read((union ctl_io *)ctsio);
7049         return (CTL_RETVAL_COMPLETE);
7050 }
7051
7052 int
7053 ctl_read_defect(struct ctl_scsiio *ctsio)
7054 {
7055         struct scsi_read_defect_data_10 *ccb10;
7056         struct scsi_read_defect_data_12 *ccb12;
7057         struct scsi_read_defect_data_hdr_10 *data10;
7058         struct scsi_read_defect_data_hdr_12 *data12;
7059         uint32_t alloc_len, data_len;
7060         uint8_t format;
7061
7062         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7063
7064         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7065                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7066                 format = ccb10->format;
7067                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7068                 data_len = sizeof(*data10);
7069         } else {
7070                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7071                 format = ccb12->format;
7072                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7073                 data_len = sizeof(*data12);
7074         }
7075         if (alloc_len == 0) {
7076                 ctl_set_success(ctsio);
7077                 ctl_done((union ctl_io *)ctsio);
7078                 return (CTL_RETVAL_COMPLETE);
7079         }
7080
7081         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7082         ctsio->kern_rel_offset = 0;
7083         ctsio->kern_sg_entries = 0;
7084         ctsio->kern_data_len = min(data_len, alloc_len);
7085         ctsio->kern_total_len = ctsio->kern_data_len;
7086
7087         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7088                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7089                     ctsio->kern_data_ptr;
7090                 data10->format = format;
7091                 scsi_ulto2b(0, data10->length);
7092         } else {
7093                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7094                     ctsio->kern_data_ptr;
7095                 data12->format = format;
7096                 scsi_ulto2b(0, data12->generation);
7097                 scsi_ulto4b(0, data12->length);
7098         }
7099
7100         ctl_set_success(ctsio);
7101         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7102         ctsio->be_move_done = ctl_config_move_done;
7103         ctl_datamove((union ctl_io *)ctsio);
7104         return (CTL_RETVAL_COMPLETE);
7105 }
7106
7107 int
7108 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7109 {
7110         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7111         struct ctl_lun *lun = CTL_LUN(ctsio);
7112         struct scsi_maintenance_in *cdb;
7113         int retval;
7114         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7115         int num_ha_groups, num_target_ports, shared_group;
7116         struct ctl_port *port;
7117         struct scsi_target_group_data *rtg_ptr;
7118         struct scsi_target_group_data_extended *rtg_ext_ptr;
7119         struct scsi_target_port_group_descriptor *tpg_desc;
7120
7121         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7122
7123         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7124         retval = CTL_RETVAL_COMPLETE;
7125
7126         switch (cdb->byte2 & STG_PDF_MASK) {
7127         case STG_PDF_LENGTH:
7128                 ext = 0;
7129                 break;
7130         case STG_PDF_EXTENDED:
7131                 ext = 1;
7132                 break;
7133         default:
7134                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7135                                       /*sks_valid*/ 1,
7136                                       /*command*/ 1,
7137                                       /*field*/ 2,
7138                                       /*bit_valid*/ 1,
7139                                       /*bit*/ 5);
7140                 ctl_done((union ctl_io *)ctsio);
7141                 return(retval);
7142         }
7143
7144         num_target_ports = 0;
7145         shared_group = (softc->is_single != 0);
7146         mtx_lock(&softc->ctl_lock);
7147         STAILQ_FOREACH(port, &softc->port_list, links) {
7148                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7149                         continue;
7150                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7151                         continue;
7152                 num_target_ports++;
7153                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7154                         shared_group = 1;
7155         }
7156         mtx_unlock(&softc->ctl_lock);
7157         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7158
7159         if (ext)
7160                 total_len = sizeof(struct scsi_target_group_data_extended);
7161         else
7162                 total_len = sizeof(struct scsi_target_group_data);
7163         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7164                 (shared_group + num_ha_groups) +
7165             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7166
7167         alloc_len = scsi_4btoul(cdb->length);
7168
7169         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7170         ctsio->kern_sg_entries = 0;
7171         ctsio->kern_rel_offset = 0;
7172         ctsio->kern_data_len = min(total_len, alloc_len);
7173         ctsio->kern_total_len = ctsio->kern_data_len;
7174
7175         if (ext) {
7176                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7177                     ctsio->kern_data_ptr;
7178                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7179                 rtg_ext_ptr->format_type = 0x10;
7180                 rtg_ext_ptr->implicit_transition_time = 0;
7181                 tpg_desc = &rtg_ext_ptr->groups[0];
7182         } else {
7183                 rtg_ptr = (struct scsi_target_group_data *)
7184                     ctsio->kern_data_ptr;
7185                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7186                 tpg_desc = &rtg_ptr->groups[0];
7187         }
7188
7189         mtx_lock(&softc->ctl_lock);
7190         pg = softc->port_min / softc->port_cnt;
7191         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7192                 /* Some shelf is known to be primary. */
7193                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7194                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7195                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7196                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7197                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7198                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7199                 else
7200                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7201                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7202                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7203                 } else {
7204                         ts = os;
7205                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7206                 }
7207         } else {
7208                 /* No known primary shelf. */
7209                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7210                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7211                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7212                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7213                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7214                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7215                 } else {
7216                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7217                 }
7218         }
7219         if (shared_group) {
7220                 tpg_desc->pref_state = ts;
7221                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7222                     TPG_U_SUP | TPG_T_SUP;
7223                 scsi_ulto2b(1, tpg_desc->target_port_group);
7224                 tpg_desc->status = TPG_IMPLICIT;
7225                 pc = 0;
7226                 STAILQ_FOREACH(port, &softc->port_list, links) {
7227                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7228                                 continue;
7229                         if (!softc->is_single &&
7230                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7231                                 continue;
7232                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7233                                 continue;
7234                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7235                             relative_target_port_identifier);
7236                         pc++;
7237                 }
7238                 tpg_desc->target_port_count = pc;
7239                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7240                     &tpg_desc->descriptors[pc];
7241         }
7242         for (g = 0; g < num_ha_groups; g++) {
7243                 tpg_desc->pref_state = (g == pg) ? ts : os;
7244                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7245                     TPG_U_SUP | TPG_T_SUP;
7246                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7247                 tpg_desc->status = TPG_IMPLICIT;
7248                 pc = 0;
7249                 STAILQ_FOREACH(port, &softc->port_list, links) {
7250                         if (port->targ_port < g * softc->port_cnt ||
7251                             port->targ_port >= (g + 1) * softc->port_cnt)
7252                                 continue;
7253                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7254                                 continue;
7255                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7256                                 continue;
7257                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7258                                 continue;
7259                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7260                             relative_target_port_identifier);
7261                         pc++;
7262                 }
7263                 tpg_desc->target_port_count = pc;
7264                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7265                     &tpg_desc->descriptors[pc];
7266         }
7267         mtx_unlock(&softc->ctl_lock);
7268
7269         ctl_set_success(ctsio);
7270         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7271         ctsio->be_move_done = ctl_config_move_done;
7272         ctl_datamove((union ctl_io *)ctsio);
7273         return(retval);
7274 }
7275
7276 int
7277 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7278 {
7279         struct ctl_lun *lun = CTL_LUN(ctsio);
7280         struct scsi_report_supported_opcodes *cdb;
7281         const struct ctl_cmd_entry *entry, *sentry;
7282         struct scsi_report_supported_opcodes_all *all;
7283         struct scsi_report_supported_opcodes_descr *descr;
7284         struct scsi_report_supported_opcodes_one *one;
7285         int retval;
7286         int alloc_len, total_len;
7287         int opcode, service_action, i, j, num;
7288
7289         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7290
7291         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7292         retval = CTL_RETVAL_COMPLETE;
7293
7294         opcode = cdb->requested_opcode;
7295         service_action = scsi_2btoul(cdb->requested_service_action);
7296         switch (cdb->options & RSO_OPTIONS_MASK) {
7297         case RSO_OPTIONS_ALL:
7298                 num = 0;
7299                 for (i = 0; i < 256; i++) {
7300                         entry = &ctl_cmd_table[i];
7301                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7302                                 for (j = 0; j < 32; j++) {
7303                                         sentry = &((const struct ctl_cmd_entry *)
7304                                             entry->execute)[j];
7305                                         if (ctl_cmd_applicable(
7306                                             lun->be_lun->lun_type, sentry))
7307                                                 num++;
7308                                 }
7309                         } else {
7310                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7311                                     entry))
7312                                         num++;
7313                         }
7314                 }
7315                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7316                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7317                 break;
7318         case RSO_OPTIONS_OC:
7319                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7320                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7321                                               /*sks_valid*/ 1,
7322                                               /*command*/ 1,
7323                                               /*field*/ 2,
7324                                               /*bit_valid*/ 1,
7325                                               /*bit*/ 2);
7326                         ctl_done((union ctl_io *)ctsio);
7327                         return (CTL_RETVAL_COMPLETE);
7328                 }
7329                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7330                 break;
7331         case RSO_OPTIONS_OC_SA:
7332                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7333                     service_action >= 32) {
7334                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7335                                               /*sks_valid*/ 1,
7336                                               /*command*/ 1,
7337                                               /*field*/ 2,
7338                                               /*bit_valid*/ 1,
7339                                               /*bit*/ 2);
7340                         ctl_done((union ctl_io *)ctsio);
7341                         return (CTL_RETVAL_COMPLETE);
7342                 }
7343                 /* FALLTHROUGH */
7344         case RSO_OPTIONS_OC_ASA:
7345                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7346                 break;
7347         default:
7348                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7349                                       /*sks_valid*/ 1,
7350                                       /*command*/ 1,
7351                                       /*field*/ 2,
7352                                       /*bit_valid*/ 1,
7353                                       /*bit*/ 2);
7354                 ctl_done((union ctl_io *)ctsio);
7355                 return (CTL_RETVAL_COMPLETE);
7356         }
7357
7358         alloc_len = scsi_4btoul(cdb->length);
7359
7360         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7361         ctsio->kern_sg_entries = 0;
7362         ctsio->kern_rel_offset = 0;
7363         ctsio->kern_data_len = min(total_len, alloc_len);
7364         ctsio->kern_total_len = ctsio->kern_data_len;
7365
7366         switch (cdb->options & RSO_OPTIONS_MASK) {
7367         case RSO_OPTIONS_ALL:
7368                 all = (struct scsi_report_supported_opcodes_all *)
7369                     ctsio->kern_data_ptr;
7370                 num = 0;
7371                 for (i = 0; i < 256; i++) {
7372                         entry = &ctl_cmd_table[i];
7373                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7374                                 for (j = 0; j < 32; j++) {
7375                                         sentry = &((const struct ctl_cmd_entry *)
7376                                             entry->execute)[j];
7377                                         if (!ctl_cmd_applicable(
7378                                             lun->be_lun->lun_type, sentry))
7379                                                 continue;
7380                                         descr = &all->descr[num++];
7381                                         descr->opcode = i;
7382                                         scsi_ulto2b(j, descr->service_action);
7383                                         descr->flags = RSO_SERVACTV;
7384                                         scsi_ulto2b(sentry->length,
7385                                             descr->cdb_length);
7386                                 }
7387                         } else {
7388                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7389                                     entry))
7390                                         continue;
7391                                 descr = &all->descr[num++];
7392                                 descr->opcode = i;
7393                                 scsi_ulto2b(0, descr->service_action);
7394                                 descr->flags = 0;
7395                                 scsi_ulto2b(entry->length, descr->cdb_length);
7396                         }
7397                 }
7398                 scsi_ulto4b(
7399                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7400                     all->length);
7401                 break;
7402         case RSO_OPTIONS_OC:
7403                 one = (struct scsi_report_supported_opcodes_one *)
7404                     ctsio->kern_data_ptr;
7405                 entry = &ctl_cmd_table[opcode];
7406                 goto fill_one;
7407         case RSO_OPTIONS_OC_SA:
7408                 one = (struct scsi_report_supported_opcodes_one *)
7409                     ctsio->kern_data_ptr;
7410                 entry = &ctl_cmd_table[opcode];
7411                 entry = &((const struct ctl_cmd_entry *)
7412                     entry->execute)[service_action];
7413 fill_one:
7414                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7415                         one->support = 3;
7416                         scsi_ulto2b(entry->length, one->cdb_length);
7417                         one->cdb_usage[0] = opcode;
7418                         memcpy(&one->cdb_usage[1], entry->usage,
7419                             entry->length - 1);
7420                 } else
7421                         one->support = 1;
7422                 break;
7423         case RSO_OPTIONS_OC_ASA:
7424                 one = (struct scsi_report_supported_opcodes_one *)
7425                     ctsio->kern_data_ptr;
7426                 entry = &ctl_cmd_table[opcode];
7427                 if (entry->flags & CTL_CMD_FLAG_SA5) {
7428                         entry = &((const struct ctl_cmd_entry *)
7429                             entry->execute)[service_action];
7430                 } else if (service_action != 0) {
7431                         one->support = 1;
7432                         break;
7433                 }
7434                 goto fill_one;
7435         }
7436
7437         ctl_set_success(ctsio);
7438         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7439         ctsio->be_move_done = ctl_config_move_done;
7440         ctl_datamove((union ctl_io *)ctsio);
7441         return(retval);
7442 }
7443
7444 int
7445 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7446 {
7447         struct scsi_report_supported_tmf *cdb;
7448         struct scsi_report_supported_tmf_ext_data *data;
7449         int retval;
7450         int alloc_len, total_len;
7451
7452         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7453
7454         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7455
7456         retval = CTL_RETVAL_COMPLETE;
7457
7458         if (cdb->options & RST_REPD)
7459                 total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7460         else
7461                 total_len = sizeof(struct scsi_report_supported_tmf_data);
7462         alloc_len = scsi_4btoul(cdb->length);
7463
7464         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7465         ctsio->kern_sg_entries = 0;
7466         ctsio->kern_rel_offset = 0;
7467         ctsio->kern_data_len = min(total_len, alloc_len);
7468         ctsio->kern_total_len = ctsio->kern_data_len;
7469
7470         data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7471         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7472             RST_TRS;
7473         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7474         data->length = total_len - 4;
7475
7476         ctl_set_success(ctsio);
7477         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7478         ctsio->be_move_done = ctl_config_move_done;
7479         ctl_datamove((union ctl_io *)ctsio);
7480         return (retval);
7481 }
7482
7483 int
7484 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7485 {
7486         struct scsi_report_timestamp *cdb;
7487         struct scsi_report_timestamp_data *data;
7488         struct timeval tv;
7489         int64_t timestamp;
7490         int retval;
7491         int alloc_len, total_len;
7492
7493         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7494
7495         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7496
7497         retval = CTL_RETVAL_COMPLETE;
7498
7499         total_len = sizeof(struct scsi_report_timestamp_data);
7500         alloc_len = scsi_4btoul(cdb->length);
7501
7502         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7503         ctsio->kern_sg_entries = 0;
7504         ctsio->kern_rel_offset = 0;
7505         ctsio->kern_data_len = min(total_len, alloc_len);
7506         ctsio->kern_total_len = ctsio->kern_data_len;
7507
7508         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7509         scsi_ulto2b(sizeof(*data) - 2, data->length);
7510         data->origin = RTS_ORIG_OUTSIDE;
7511         getmicrotime(&tv);
7512         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7513         scsi_ulto4b(timestamp >> 16, data->timestamp);
7514         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7515
7516         ctl_set_success(ctsio);
7517         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7518         ctsio->be_move_done = ctl_config_move_done;
7519         ctl_datamove((union ctl_io *)ctsio);
7520         return (retval);
7521 }
7522
7523 int
7524 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7525 {
7526         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7527         struct ctl_lun *lun = CTL_LUN(ctsio);
7528         struct scsi_per_res_in *cdb;
7529         int alloc_len, total_len = 0;
7530         /* struct scsi_per_res_in_rsrv in_data; */
7531         uint64_t key;
7532
7533         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7534
7535         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7536
7537         alloc_len = scsi_2btoul(cdb->length);
7538
7539 retry:
7540         mtx_lock(&lun->lun_lock);
7541         switch (cdb->action) {
7542         case SPRI_RK: /* read keys */
7543                 total_len = sizeof(struct scsi_per_res_in_keys) +
7544                         lun->pr_key_count *
7545                         sizeof(struct scsi_per_res_key);
7546                 break;
7547         case SPRI_RR: /* read reservation */
7548                 if (lun->flags & CTL_LUN_PR_RESERVED)
7549                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7550                 else
7551                         total_len = sizeof(struct scsi_per_res_in_header);
7552                 break;
7553         case SPRI_RC: /* report capabilities */
7554                 total_len = sizeof(struct scsi_per_res_cap);
7555                 break;
7556         case SPRI_RS: /* read full status */
7557                 total_len = sizeof(struct scsi_per_res_in_header) +
7558                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7559                     lun->pr_key_count;
7560                 break;
7561         default:
7562                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7563         }
7564         mtx_unlock(&lun->lun_lock);
7565
7566         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7567         ctsio->kern_rel_offset = 0;
7568         ctsio->kern_sg_entries = 0;
7569         ctsio->kern_data_len = min(total_len, alloc_len);
7570         ctsio->kern_total_len = ctsio->kern_data_len;
7571
7572         mtx_lock(&lun->lun_lock);
7573         switch (cdb->action) {
7574         case SPRI_RK: { // read keys
7575         struct scsi_per_res_in_keys *res_keys;
7576                 int i, key_count;
7577
7578                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7579
7580                 /*
7581                  * We had to drop the lock to allocate our buffer, which
7582                  * leaves time for someone to come in with another
7583                  * persistent reservation.  (That is unlikely, though,
7584                  * since this should be the only persistent reservation
7585                  * command active right now.)
7586                  */
7587                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7588                     (lun->pr_key_count *
7589                      sizeof(struct scsi_per_res_key)))){
7590                         mtx_unlock(&lun->lun_lock);
7591                         free(ctsio->kern_data_ptr, M_CTL);
7592                         printf("%s: reservation length changed, retrying\n",
7593                                __func__);
7594                         goto retry;
7595                 }
7596
7597                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7598
7599                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7600                              lun->pr_key_count, res_keys->header.length);
7601
7602                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7603                         if ((key = ctl_get_prkey(lun, i)) == 0)
7604                                 continue;
7605
7606                         /*
7607                          * We used lun->pr_key_count to calculate the
7608                          * size to allocate.  If it turns out the number of
7609                          * initiators with the registered flag set is
7610                          * larger than that (i.e. they haven't been kept in
7611                          * sync), we've got a problem.
7612                          */
7613                         if (key_count >= lun->pr_key_count) {
7614                                 key_count++;
7615                                 continue;
7616                         }
7617                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7618                         key_count++;
7619                 }
7620                 break;
7621         }
7622         case SPRI_RR: { // read reservation
7623                 struct scsi_per_res_in_rsrv *res;
7624                 int tmp_len, header_only;
7625
7626                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7627
7628                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7629
7630                 if (lun->flags & CTL_LUN_PR_RESERVED)
7631                 {
7632                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7633                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7634                                     res->header.length);
7635                         header_only = 0;
7636                 } else {
7637                         tmp_len = sizeof(struct scsi_per_res_in_header);
7638                         scsi_ulto4b(0, res->header.length);
7639                         header_only = 1;
7640                 }
7641
7642                 /*
7643                  * We had to drop the lock to allocate our buffer, which
7644                  * leaves time for someone to come in with another
7645                  * persistent reservation.  (That is unlikely, though,
7646                  * since this should be the only persistent reservation
7647                  * command active right now.)
7648                  */
7649                 if (tmp_len != total_len) {
7650                         mtx_unlock(&lun->lun_lock);
7651                         free(ctsio->kern_data_ptr, M_CTL);
7652                         printf("%s: reservation status changed, retrying\n",
7653                                __func__);
7654                         goto retry;
7655                 }
7656
7657                 /*
7658                  * No reservation held, so we're done.
7659                  */
7660                 if (header_only != 0)
7661                         break;
7662
7663                 /*
7664                  * If the registration is an All Registrants type, the key
7665                  * is 0, since it doesn't really matter.
7666                  */
7667                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7668                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7669                             res->data.reservation);
7670                 }
7671                 res->data.scopetype = lun->pr_res_type;
7672                 break;
7673         }
7674         case SPRI_RC:     //report capabilities
7675         {
7676                 struct scsi_per_res_cap *res_cap;
7677                 uint16_t type_mask;
7678
7679                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7680                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7681                 res_cap->flags1 = SPRI_CRH;
7682                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7683                 type_mask = SPRI_TM_WR_EX_AR |
7684                             SPRI_TM_EX_AC_RO |
7685                             SPRI_TM_WR_EX_RO |
7686                             SPRI_TM_EX_AC |
7687                             SPRI_TM_WR_EX |
7688                             SPRI_TM_EX_AC_AR;
7689                 scsi_ulto2b(type_mask, res_cap->type_mask);
7690                 break;
7691         }
7692         case SPRI_RS: { // read full status
7693                 struct scsi_per_res_in_full *res_status;
7694                 struct scsi_per_res_in_full_desc *res_desc;
7695                 struct ctl_port *port;
7696                 int i, len;
7697
7698                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7699
7700                 /*
7701                  * We had to drop the lock to allocate our buffer, which
7702                  * leaves time for someone to come in with another
7703                  * persistent reservation.  (That is unlikely, though,
7704                  * since this should be the only persistent reservation
7705                  * command active right now.)
7706                  */
7707                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7708                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7709                      lun->pr_key_count)){
7710                         mtx_unlock(&lun->lun_lock);
7711                         free(ctsio->kern_data_ptr, M_CTL);
7712                         printf("%s: reservation length changed, retrying\n",
7713                                __func__);
7714                         goto retry;
7715                 }
7716
7717                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7718
7719                 res_desc = &res_status->desc[0];
7720                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7721                         if ((key = ctl_get_prkey(lun, i)) == 0)
7722                                 continue;
7723
7724                         scsi_u64to8b(key, res_desc->res_key.key);
7725                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7726                             (lun->pr_res_idx == i ||
7727                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7728                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7729                                 res_desc->scopetype = lun->pr_res_type;
7730                         }
7731                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7732                             res_desc->rel_trgt_port_id);
7733                         len = 0;
7734                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7735                         if (port != NULL)
7736                                 len = ctl_create_iid(port,
7737                                     i % CTL_MAX_INIT_PER_PORT,
7738                                     res_desc->transport_id);
7739                         scsi_ulto4b(len, res_desc->additional_length);
7740                         res_desc = (struct scsi_per_res_in_full_desc *)
7741                             &res_desc->transport_id[len];
7742                 }
7743                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7744                     res_status->header.length);
7745                 break;
7746         }
7747         default:
7748                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7749         }
7750         mtx_unlock(&lun->lun_lock);
7751
7752         ctl_set_success(ctsio);
7753         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7754         ctsio->be_move_done = ctl_config_move_done;
7755         ctl_datamove((union ctl_io *)ctsio);
7756         return (CTL_RETVAL_COMPLETE);
7757 }
7758
7759 /*
7760  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7761  * it should return.
7762  */
7763 static int
7764 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7765                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7766                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7767                 struct scsi_per_res_out_parms* param)
7768 {
7769         union ctl_ha_msg persis_io;
7770         int i;
7771
7772         mtx_lock(&lun->lun_lock);
7773         if (sa_res_key == 0) {
7774                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7775                         /* validate scope and type */
7776                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7777                              SPR_LU_SCOPE) {
7778                                 mtx_unlock(&lun->lun_lock);
7779                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7780                                                       /*sks_valid*/ 1,
7781                                                       /*command*/ 1,
7782                                                       /*field*/ 2,
7783                                                       /*bit_valid*/ 1,
7784                                                       /*bit*/ 4);
7785                                 ctl_done((union ctl_io *)ctsio);
7786                                 return (1);
7787                         }
7788
7789                         if (type>8 || type==2 || type==4 || type==0) {
7790                                 mtx_unlock(&lun->lun_lock);
7791                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7792                                                       /*sks_valid*/ 1,
7793                                                       /*command*/ 1,
7794                                                       /*field*/ 2,
7795                                                       /*bit_valid*/ 1,
7796                                                       /*bit*/ 0);
7797                                 ctl_done((union ctl_io *)ctsio);
7798                                 return (1);
7799                         }
7800
7801                         /*
7802                          * Unregister everybody else and build UA for
7803                          * them
7804                          */
7805                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7806                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7807                                         continue;
7808
7809                                 ctl_clr_prkey(lun, i);
7810                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7811                         }
7812                         lun->pr_key_count = 1;
7813                         lun->pr_res_type = type;
7814                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7815                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7816                                 lun->pr_res_idx = residx;
7817                         lun->pr_generation++;
7818                         mtx_unlock(&lun->lun_lock);
7819
7820                         /* send msg to other side */
7821                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7822                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7823                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7824                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7825                         persis_io.pr.pr_info.res_type = type;
7826                         memcpy(persis_io.pr.pr_info.sa_res_key,
7827                                param->serv_act_res_key,
7828                                sizeof(param->serv_act_res_key));
7829                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7830                             sizeof(persis_io.pr), M_WAITOK);
7831                 } else {
7832                         /* not all registrants */
7833                         mtx_unlock(&lun->lun_lock);
7834                         free(ctsio->kern_data_ptr, M_CTL);
7835                         ctl_set_invalid_field(ctsio,
7836                                               /*sks_valid*/ 1,
7837                                               /*command*/ 0,
7838                                               /*field*/ 8,
7839                                               /*bit_valid*/ 0,
7840                                               /*bit*/ 0);
7841                         ctl_done((union ctl_io *)ctsio);
7842                         return (1);
7843                 }
7844         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7845                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7846                 int found = 0;
7847
7848                 if (res_key == sa_res_key) {
7849                         /* special case */
7850                         /*
7851                          * The spec implies this is not good but doesn't
7852                          * say what to do. There are two choices either
7853                          * generate a res conflict or check condition
7854                          * with illegal field in parameter data. Since
7855                          * that is what is done when the sa_res_key is
7856                          * zero I'll take that approach since this has
7857                          * to do with the sa_res_key.
7858                          */
7859                         mtx_unlock(&lun->lun_lock);
7860                         free(ctsio->kern_data_ptr, M_CTL);
7861                         ctl_set_invalid_field(ctsio,
7862                                               /*sks_valid*/ 1,
7863                                               /*command*/ 0,
7864                                               /*field*/ 8,
7865                                               /*bit_valid*/ 0,
7866                                               /*bit*/ 0);
7867                         ctl_done((union ctl_io *)ctsio);
7868                         return (1);
7869                 }
7870
7871                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7872                         if (ctl_get_prkey(lun, i) != sa_res_key)
7873                                 continue;
7874
7875                         found = 1;
7876                         ctl_clr_prkey(lun, i);
7877                         lun->pr_key_count--;
7878                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7879                 }
7880                 if (!found) {
7881                         mtx_unlock(&lun->lun_lock);
7882                         free(ctsio->kern_data_ptr, M_CTL);
7883                         ctl_set_reservation_conflict(ctsio);
7884                         ctl_done((union ctl_io *)ctsio);
7885                         return (CTL_RETVAL_COMPLETE);
7886                 }
7887                 lun->pr_generation++;
7888                 mtx_unlock(&lun->lun_lock);
7889
7890                 /* send msg to other side */
7891                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7892                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7893                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7894                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7895                 persis_io.pr.pr_info.res_type = type;
7896                 memcpy(persis_io.pr.pr_info.sa_res_key,
7897                        param->serv_act_res_key,
7898                        sizeof(param->serv_act_res_key));
7899                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7900                     sizeof(persis_io.pr), M_WAITOK);
7901         } else {
7902                 /* Reserved but not all registrants */
7903                 /* sa_res_key is res holder */
7904                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7905                         /* validate scope and type */
7906                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7907                              SPR_LU_SCOPE) {
7908                                 mtx_unlock(&lun->lun_lock);
7909                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7910                                                       /*sks_valid*/ 1,
7911                                                       /*command*/ 1,
7912                                                       /*field*/ 2,
7913                                                       /*bit_valid*/ 1,
7914                                                       /*bit*/ 4);
7915                                 ctl_done((union ctl_io *)ctsio);
7916                                 return (1);
7917                         }
7918
7919                         if (type>8 || type==2 || type==4 || type==0) {
7920                                 mtx_unlock(&lun->lun_lock);
7921                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7922                                                       /*sks_valid*/ 1,
7923                                                       /*command*/ 1,
7924                                                       /*field*/ 2,
7925                                                       /*bit_valid*/ 1,
7926                                                       /*bit*/ 0);
7927                                 ctl_done((union ctl_io *)ctsio);
7928                                 return (1);
7929                         }
7930
7931                         /*
7932                          * Do the following:
7933                          * if sa_res_key != res_key remove all
7934                          * registrants w/sa_res_key and generate UA
7935                          * for these registrants(Registrations
7936                          * Preempted) if it wasn't an exclusive
7937                          * reservation generate UA(Reservations
7938                          * Preempted) for all other registered nexuses
7939                          * if the type has changed. Establish the new
7940                          * reservation and holder. If res_key and
7941                          * sa_res_key are the same do the above
7942                          * except don't unregister the res holder.
7943                          */
7944
7945                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7946                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7947                                         continue;
7948
7949                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
7950                                         ctl_clr_prkey(lun, i);
7951                                         lun->pr_key_count--;
7952                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7953                                 } else if (type != lun->pr_res_type &&
7954                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
7955                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
7956                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
7957                                 }
7958                         }
7959                         lun->pr_res_type = type;
7960                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7961                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7962                                 lun->pr_res_idx = residx;
7963                         else
7964                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7965                         lun->pr_generation++;
7966                         mtx_unlock(&lun->lun_lock);
7967
7968                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7969                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7970                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7971                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7972                         persis_io.pr.pr_info.res_type = type;
7973                         memcpy(persis_io.pr.pr_info.sa_res_key,
7974                                param->serv_act_res_key,
7975                                sizeof(param->serv_act_res_key));
7976                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7977                             sizeof(persis_io.pr), M_WAITOK);
7978                 } else {
7979                         /*
7980                          * sa_res_key is not the res holder just
7981                          * remove registrants
7982                          */
7983                         int found=0;
7984
7985                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7986                                 if (sa_res_key != ctl_get_prkey(lun, i))
7987                                         continue;
7988
7989                                 found = 1;
7990                                 ctl_clr_prkey(lun, i);
7991                                 lun->pr_key_count--;
7992                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7993                         }
7994
7995                         if (!found) {
7996                                 mtx_unlock(&lun->lun_lock);
7997                                 free(ctsio->kern_data_ptr, M_CTL);
7998                                 ctl_set_reservation_conflict(ctsio);
7999                                 ctl_done((union ctl_io *)ctsio);
8000                                 return (1);
8001                         }
8002                         lun->pr_generation++;
8003                         mtx_unlock(&lun->lun_lock);
8004
8005                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8006                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8007                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8008                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8009                         persis_io.pr.pr_info.res_type = type;
8010                         memcpy(persis_io.pr.pr_info.sa_res_key,
8011                                param->serv_act_res_key,
8012                                sizeof(param->serv_act_res_key));
8013                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8014                             sizeof(persis_io.pr), M_WAITOK);
8015                 }
8016         }
8017         return (0);
8018 }
8019
8020 static void
8021 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8022 {
8023         uint64_t sa_res_key;
8024         int i;
8025
8026         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8027
8028         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8029          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8030          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8031                 if (sa_res_key == 0) {
8032                         /*
8033                          * Unregister everybody else and build UA for
8034                          * them
8035                          */
8036                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8037                                 if (i == msg->pr.pr_info.residx ||
8038                                     ctl_get_prkey(lun, i) == 0)
8039                                         continue;
8040
8041                                 ctl_clr_prkey(lun, i);
8042                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8043                         }
8044
8045                         lun->pr_key_count = 1;
8046                         lun->pr_res_type = msg->pr.pr_info.res_type;
8047                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8048                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8049                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8050                 } else {
8051                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8052                                 if (sa_res_key == ctl_get_prkey(lun, i))
8053                                         continue;
8054
8055                                 ctl_clr_prkey(lun, i);
8056                                 lun->pr_key_count--;
8057                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8058                         }
8059                 }
8060         } else {
8061                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8062                         if (i == msg->pr.pr_info.residx ||
8063                             ctl_get_prkey(lun, i) == 0)
8064                                 continue;
8065
8066                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8067                                 ctl_clr_prkey(lun, i);
8068                                 lun->pr_key_count--;
8069                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8070                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8071                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8072                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8073                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8074                         }
8075                 }
8076                 lun->pr_res_type = msg->pr.pr_info.res_type;
8077                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8078                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8079                         lun->pr_res_idx = msg->pr.pr_info.residx;
8080                 else
8081                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8082         }
8083         lun->pr_generation++;
8084
8085 }
8086
8087
8088 int
8089 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8090 {
8091         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8092         struct ctl_lun *lun = CTL_LUN(ctsio);
8093         int retval;
8094         u_int32_t param_len;
8095         struct scsi_per_res_out *cdb;
8096         struct scsi_per_res_out_parms* param;
8097         uint32_t residx;
8098         uint64_t res_key, sa_res_key, key;
8099         uint8_t type;
8100         union ctl_ha_msg persis_io;
8101         int    i;
8102
8103         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8104
8105         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8106         retval = CTL_RETVAL_COMPLETE;
8107
8108         /*
8109          * We only support whole-LUN scope.  The scope & type are ignored for
8110          * register, register and ignore existing key and clear.
8111          * We sometimes ignore scope and type on preempts too!!
8112          * Verify reservation type here as well.
8113          */
8114         type = cdb->scope_type & SPR_TYPE_MASK;
8115         if ((cdb->action == SPRO_RESERVE)
8116          || (cdb->action == SPRO_RELEASE)) {
8117                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8118                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8119                                               /*sks_valid*/ 1,
8120                                               /*command*/ 1,
8121                                               /*field*/ 2,
8122                                               /*bit_valid*/ 1,
8123                                               /*bit*/ 4);
8124                         ctl_done((union ctl_io *)ctsio);
8125                         return (CTL_RETVAL_COMPLETE);
8126                 }
8127
8128                 if (type>8 || type==2 || type==4 || type==0) {
8129                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8130                                               /*sks_valid*/ 1,
8131                                               /*command*/ 1,
8132                                               /*field*/ 2,
8133                                               /*bit_valid*/ 1,
8134                                               /*bit*/ 0);
8135                         ctl_done((union ctl_io *)ctsio);
8136                         return (CTL_RETVAL_COMPLETE);
8137                 }
8138         }
8139
8140         param_len = scsi_4btoul(cdb->length);
8141
8142         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8143                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8144                 ctsio->kern_data_len = param_len;
8145                 ctsio->kern_total_len = param_len;
8146                 ctsio->kern_rel_offset = 0;
8147                 ctsio->kern_sg_entries = 0;
8148                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8149                 ctsio->be_move_done = ctl_config_move_done;
8150                 ctl_datamove((union ctl_io *)ctsio);
8151
8152                 return (CTL_RETVAL_COMPLETE);
8153         }
8154
8155         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8156
8157         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8158         res_key = scsi_8btou64(param->res_key.key);
8159         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8160
8161         /*
8162          * Validate the reservation key here except for SPRO_REG_IGNO
8163          * This must be done for all other service actions
8164          */
8165         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8166                 mtx_lock(&lun->lun_lock);
8167                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8168                         if (res_key != key) {
8169                                 /*
8170                                  * The current key passed in doesn't match
8171                                  * the one the initiator previously
8172                                  * registered.
8173                                  */
8174                                 mtx_unlock(&lun->lun_lock);
8175                                 free(ctsio->kern_data_ptr, M_CTL);
8176                                 ctl_set_reservation_conflict(ctsio);
8177                                 ctl_done((union ctl_io *)ctsio);
8178                                 return (CTL_RETVAL_COMPLETE);
8179                         }
8180                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8181                         /*
8182                          * We are not registered
8183                          */
8184                         mtx_unlock(&lun->lun_lock);
8185                         free(ctsio->kern_data_ptr, M_CTL);
8186                         ctl_set_reservation_conflict(ctsio);
8187                         ctl_done((union ctl_io *)ctsio);
8188                         return (CTL_RETVAL_COMPLETE);
8189                 } else if (res_key != 0) {
8190                         /*
8191                          * We are not registered and trying to register but
8192                          * the register key isn't zero.
8193                          */
8194                         mtx_unlock(&lun->lun_lock);
8195                         free(ctsio->kern_data_ptr, M_CTL);
8196                         ctl_set_reservation_conflict(ctsio);
8197                         ctl_done((union ctl_io *)ctsio);
8198                         return (CTL_RETVAL_COMPLETE);
8199                 }
8200                 mtx_unlock(&lun->lun_lock);
8201         }
8202
8203         switch (cdb->action & SPRO_ACTION_MASK) {
8204         case SPRO_REGISTER:
8205         case SPRO_REG_IGNO: {
8206
8207 #if 0
8208                 printf("Registration received\n");
8209 #endif
8210
8211                 /*
8212                  * We don't support any of these options, as we report in
8213                  * the read capabilities request (see
8214                  * ctl_persistent_reserve_in(), above).
8215                  */
8216                 if ((param->flags & SPR_SPEC_I_PT)
8217                  || (param->flags & SPR_ALL_TG_PT)
8218                  || (param->flags & SPR_APTPL)) {
8219                         int bit_ptr;
8220
8221                         if (param->flags & SPR_APTPL)
8222                                 bit_ptr = 0;
8223                         else if (param->flags & SPR_ALL_TG_PT)
8224                                 bit_ptr = 2;
8225                         else /* SPR_SPEC_I_PT */
8226                                 bit_ptr = 3;
8227
8228                         free(ctsio->kern_data_ptr, M_CTL);
8229                         ctl_set_invalid_field(ctsio,
8230                                               /*sks_valid*/ 1,
8231                                               /*command*/ 0,
8232                                               /*field*/ 20,
8233                                               /*bit_valid*/ 1,
8234                                               /*bit*/ bit_ptr);
8235                         ctl_done((union ctl_io *)ctsio);
8236                         return (CTL_RETVAL_COMPLETE);
8237                 }
8238
8239                 mtx_lock(&lun->lun_lock);
8240
8241                 /*
8242                  * The initiator wants to clear the
8243                  * key/unregister.
8244                  */
8245                 if (sa_res_key == 0) {
8246                         if ((res_key == 0
8247                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8248                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8249                           && ctl_get_prkey(lun, residx) == 0)) {
8250                                 mtx_unlock(&lun->lun_lock);
8251                                 goto done;
8252                         }
8253
8254                         ctl_clr_prkey(lun, residx);
8255                         lun->pr_key_count--;
8256
8257                         if (residx == lun->pr_res_idx) {
8258                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8259                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8260
8261                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8262                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8263                                     lun->pr_key_count) {
8264                                         /*
8265                                          * If the reservation is a registrants
8266                                          * only type we need to generate a UA
8267                                          * for other registered inits.  The
8268                                          * sense code should be RESERVATIONS
8269                                          * RELEASED
8270                                          */
8271
8272                                         for (i = softc->init_min; i < softc->init_max; i++){
8273                                                 if (ctl_get_prkey(lun, i) == 0)
8274                                                         continue;
8275                                                 ctl_est_ua(lun, i,
8276                                                     CTL_UA_RES_RELEASE);
8277                                         }
8278                                 }
8279                                 lun->pr_res_type = 0;
8280                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8281                                 if (lun->pr_key_count==0) {
8282                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8283                                         lun->pr_res_type = 0;
8284                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8285                                 }
8286                         }
8287                         lun->pr_generation++;
8288                         mtx_unlock(&lun->lun_lock);
8289
8290                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8291                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8292                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8293                         persis_io.pr.pr_info.residx = residx;
8294                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8295                             sizeof(persis_io.pr), M_WAITOK);
8296                 } else /* sa_res_key != 0 */ {
8297
8298                         /*
8299                          * If we aren't registered currently then increment
8300                          * the key count and set the registered flag.
8301                          */
8302                         ctl_alloc_prkey(lun, residx);
8303                         if (ctl_get_prkey(lun, residx) == 0)
8304                                 lun->pr_key_count++;
8305                         ctl_set_prkey(lun, residx, sa_res_key);
8306                         lun->pr_generation++;
8307                         mtx_unlock(&lun->lun_lock);
8308
8309                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8310                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8311                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8312                         persis_io.pr.pr_info.residx = residx;
8313                         memcpy(persis_io.pr.pr_info.sa_res_key,
8314                                param->serv_act_res_key,
8315                                sizeof(param->serv_act_res_key));
8316                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8317                             sizeof(persis_io.pr), M_WAITOK);
8318                 }
8319
8320                 break;
8321         }
8322         case SPRO_RESERVE:
8323 #if 0
8324                 printf("Reserve executed type %d\n", type);
8325 #endif
8326                 mtx_lock(&lun->lun_lock);
8327                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8328                         /*
8329                          * if this isn't the reservation holder and it's
8330                          * not a "all registrants" type or if the type is
8331                          * different then we have a conflict
8332                          */
8333                         if ((lun->pr_res_idx != residx
8334                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8335                          || lun->pr_res_type != type) {
8336                                 mtx_unlock(&lun->lun_lock);
8337                                 free(ctsio->kern_data_ptr, M_CTL);
8338                                 ctl_set_reservation_conflict(ctsio);
8339                                 ctl_done((union ctl_io *)ctsio);
8340                                 return (CTL_RETVAL_COMPLETE);
8341                         }
8342                         mtx_unlock(&lun->lun_lock);
8343                 } else /* create a reservation */ {
8344                         /*
8345                          * If it's not an "all registrants" type record
8346                          * reservation holder
8347                          */
8348                         if (type != SPR_TYPE_WR_EX_AR
8349                          && type != SPR_TYPE_EX_AC_AR)
8350                                 lun->pr_res_idx = residx; /* Res holder */
8351                         else
8352                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8353
8354                         lun->flags |= CTL_LUN_PR_RESERVED;
8355                         lun->pr_res_type = type;
8356
8357                         mtx_unlock(&lun->lun_lock);
8358
8359                         /* send msg to other side */
8360                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8361                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8362                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8363                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8364                         persis_io.pr.pr_info.res_type = type;
8365                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8366                             sizeof(persis_io.pr), M_WAITOK);
8367                 }
8368                 break;
8369
8370         case SPRO_RELEASE:
8371                 mtx_lock(&lun->lun_lock);
8372                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8373                         /* No reservation exists return good status */
8374                         mtx_unlock(&lun->lun_lock);
8375                         goto done;
8376                 }
8377                 /*
8378                  * Is this nexus a reservation holder?
8379                  */
8380                 if (lun->pr_res_idx != residx
8381                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8382                         /*
8383                          * not a res holder return good status but
8384                          * do nothing
8385                          */
8386                         mtx_unlock(&lun->lun_lock);
8387                         goto done;
8388                 }
8389
8390                 if (lun->pr_res_type != type) {
8391                         mtx_unlock(&lun->lun_lock);
8392                         free(ctsio->kern_data_ptr, M_CTL);
8393                         ctl_set_illegal_pr_release(ctsio);
8394                         ctl_done((union ctl_io *)ctsio);
8395                         return (CTL_RETVAL_COMPLETE);
8396                 }
8397
8398                 /* okay to release */
8399                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8400                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8401                 lun->pr_res_type = 0;
8402
8403                 /*
8404                  * If this isn't an exclusive access reservation and NUAR
8405                  * is not set, generate UA for all other registrants.
8406                  */
8407                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8408                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8409                         for (i = softc->init_min; i < softc->init_max; i++) {
8410                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8411                                         continue;
8412                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8413                         }
8414                 }
8415                 mtx_unlock(&lun->lun_lock);
8416
8417                 /* Send msg to other side */
8418                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8419                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8420                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8421                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8422                      sizeof(persis_io.pr), M_WAITOK);
8423                 break;
8424
8425         case SPRO_CLEAR:
8426                 /* send msg to other side */
8427
8428                 mtx_lock(&lun->lun_lock);
8429                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8430                 lun->pr_res_type = 0;
8431                 lun->pr_key_count = 0;
8432                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8433
8434                 ctl_clr_prkey(lun, residx);
8435                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8436                         if (ctl_get_prkey(lun, i) != 0) {
8437                                 ctl_clr_prkey(lun, i);
8438                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8439                         }
8440                 lun->pr_generation++;
8441                 mtx_unlock(&lun->lun_lock);
8442
8443                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8444                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8445                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8446                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8447                      sizeof(persis_io.pr), M_WAITOK);
8448                 break;
8449
8450         case SPRO_PREEMPT:
8451         case SPRO_PRE_ABO: {
8452                 int nretval;
8453
8454                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8455                                           residx, ctsio, cdb, param);
8456                 if (nretval != 0)
8457                         return (CTL_RETVAL_COMPLETE);
8458                 break;
8459         }
8460         default:
8461                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8462         }
8463
8464 done:
8465         free(ctsio->kern_data_ptr, M_CTL);
8466         ctl_set_success(ctsio);
8467         ctl_done((union ctl_io *)ctsio);
8468
8469         return (retval);
8470 }
8471
8472 /*
8473  * This routine is for handling a message from the other SC pertaining to
8474  * persistent reserve out. All the error checking will have been done
8475  * so only perorming the action need be done here to keep the two
8476  * in sync.
8477  */
8478 static void
8479 ctl_hndl_per_res_out_on_other_sc(union ctl_io *io)
8480 {
8481         struct ctl_softc *softc = CTL_SOFTC(io);
8482         union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg;
8483         struct ctl_lun *lun;
8484         int i;
8485         uint32_t residx, targ_lun;
8486
8487         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8488         mtx_lock(&softc->ctl_lock);
8489         if (targ_lun >= CTL_MAX_LUNS ||
8490             (lun = softc->ctl_luns[targ_lun]) == NULL) {
8491                 mtx_unlock(&softc->ctl_lock);
8492                 return;
8493         }
8494         mtx_lock(&lun->lun_lock);
8495         mtx_unlock(&softc->ctl_lock);
8496         if (lun->flags & CTL_LUN_DISABLED) {
8497                 mtx_unlock(&lun->lun_lock);
8498                 return;
8499         }
8500         residx = ctl_get_initindex(&msg->hdr.nexus);
8501         switch(msg->pr.pr_info.action) {
8502         case CTL_PR_REG_KEY:
8503                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8504                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8505                         lun->pr_key_count++;
8506                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8507                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8508                 lun->pr_generation++;
8509                 break;
8510
8511         case CTL_PR_UNREG_KEY:
8512                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8513                 lun->pr_key_count--;
8514
8515                 /* XXX Need to see if the reservation has been released */
8516                 /* if so do we need to generate UA? */
8517                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8518                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8519                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8520
8521                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8522                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8523                             lun->pr_key_count) {
8524                                 /*
8525                                  * If the reservation is a registrants
8526                                  * only type we need to generate a UA
8527                                  * for other registered inits.  The
8528                                  * sense code should be RESERVATIONS
8529                                  * RELEASED
8530                                  */
8531
8532                                 for (i = softc->init_min; i < softc->init_max; i++) {
8533                                         if (ctl_get_prkey(lun, i) == 0)
8534                                                 continue;
8535
8536                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8537                                 }
8538                         }
8539                         lun->pr_res_type = 0;
8540                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8541                         if (lun->pr_key_count==0) {
8542                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8543                                 lun->pr_res_type = 0;
8544                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8545                         }
8546                 }
8547                 lun->pr_generation++;
8548                 break;
8549
8550         case CTL_PR_RESERVE:
8551                 lun->flags |= CTL_LUN_PR_RESERVED;
8552                 lun->pr_res_type = msg->pr.pr_info.res_type;
8553                 lun->pr_res_idx = msg->pr.pr_info.residx;
8554
8555                 break;
8556
8557         case CTL_PR_RELEASE:
8558                 /*
8559                  * If this isn't an exclusive access reservation and NUAR
8560                  * is not set, generate UA for all other registrants.
8561                  */
8562                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8563                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8564                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8565                         for (i = softc->init_min; i < softc->init_max; i++)
8566                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8567                                         continue;
8568                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8569                 }
8570
8571                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8572                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8573                 lun->pr_res_type = 0;
8574                 break;
8575
8576         case CTL_PR_PREEMPT:
8577                 ctl_pro_preempt_other(lun, msg);
8578                 break;
8579         case CTL_PR_CLEAR:
8580                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8581                 lun->pr_res_type = 0;
8582                 lun->pr_key_count = 0;
8583                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8584
8585                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8586                         if (ctl_get_prkey(lun, i) == 0)
8587                                 continue;
8588                         ctl_clr_prkey(lun, i);
8589                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8590                 }
8591                 lun->pr_generation++;
8592                 break;
8593         }
8594
8595         mtx_unlock(&lun->lun_lock);
8596 }
8597
8598 int
8599 ctl_read_write(struct ctl_scsiio *ctsio)
8600 {
8601         struct ctl_lun *lun = CTL_LUN(ctsio);
8602         struct ctl_lba_len_flags *lbalen;
8603         uint64_t lba;
8604         uint32_t num_blocks;
8605         int flags, retval;
8606         int isread;
8607
8608         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8609
8610         flags = 0;
8611         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8612               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8613         switch (ctsio->cdb[0]) {
8614         case READ_6:
8615         case WRITE_6: {
8616                 struct scsi_rw_6 *cdb;
8617
8618                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8619
8620                 lba = scsi_3btoul(cdb->addr);
8621                 /* only 5 bits are valid in the most significant address byte */
8622                 lba &= 0x1fffff;
8623                 num_blocks = cdb->length;
8624                 /*
8625                  * This is correct according to SBC-2.
8626                  */
8627                 if (num_blocks == 0)
8628                         num_blocks = 256;
8629                 break;
8630         }
8631         case READ_10:
8632         case WRITE_10: {
8633                 struct scsi_rw_10 *cdb;
8634
8635                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8636                 if (cdb->byte2 & SRW10_FUA)
8637                         flags |= CTL_LLF_FUA;
8638                 if (cdb->byte2 & SRW10_DPO)
8639                         flags |= CTL_LLF_DPO;
8640                 lba = scsi_4btoul(cdb->addr);
8641                 num_blocks = scsi_2btoul(cdb->length);
8642                 break;
8643         }
8644         case WRITE_VERIFY_10: {
8645                 struct scsi_write_verify_10 *cdb;
8646
8647                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8648                 flags |= CTL_LLF_FUA;
8649                 if (cdb->byte2 & SWV_DPO)
8650                         flags |= CTL_LLF_DPO;
8651                 lba = scsi_4btoul(cdb->addr);
8652                 num_blocks = scsi_2btoul(cdb->length);
8653                 break;
8654         }
8655         case READ_12:
8656         case WRITE_12: {
8657                 struct scsi_rw_12 *cdb;
8658
8659                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8660                 if (cdb->byte2 & SRW12_FUA)
8661                         flags |= CTL_LLF_FUA;
8662                 if (cdb->byte2 & SRW12_DPO)
8663                         flags |= CTL_LLF_DPO;
8664                 lba = scsi_4btoul(cdb->addr);
8665                 num_blocks = scsi_4btoul(cdb->length);
8666                 break;
8667         }
8668         case WRITE_VERIFY_12: {
8669                 struct scsi_write_verify_12 *cdb;
8670
8671                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8672                 flags |= CTL_LLF_FUA;
8673                 if (cdb->byte2 & SWV_DPO)
8674                         flags |= CTL_LLF_DPO;
8675                 lba = scsi_4btoul(cdb->addr);
8676                 num_blocks = scsi_4btoul(cdb->length);
8677                 break;
8678         }
8679         case READ_16:
8680         case WRITE_16: {
8681                 struct scsi_rw_16 *cdb;
8682
8683                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8684                 if (cdb->byte2 & SRW12_FUA)
8685                         flags |= CTL_LLF_FUA;
8686                 if (cdb->byte2 & SRW12_DPO)
8687                         flags |= CTL_LLF_DPO;
8688                 lba = scsi_8btou64(cdb->addr);
8689                 num_blocks = scsi_4btoul(cdb->length);
8690                 break;
8691         }
8692         case WRITE_ATOMIC_16: {
8693                 struct scsi_write_atomic_16 *cdb;
8694
8695                 if (lun->be_lun->atomicblock == 0) {
8696                         ctl_set_invalid_opcode(ctsio);
8697                         ctl_done((union ctl_io *)ctsio);
8698                         return (CTL_RETVAL_COMPLETE);
8699                 }
8700
8701                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8702                 if (cdb->byte2 & SRW12_FUA)
8703                         flags |= CTL_LLF_FUA;
8704                 if (cdb->byte2 & SRW12_DPO)
8705                         flags |= CTL_LLF_DPO;
8706                 lba = scsi_8btou64(cdb->addr);
8707                 num_blocks = scsi_2btoul(cdb->length);
8708                 if (num_blocks > lun->be_lun->atomicblock) {
8709                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8710                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8711                             /*bit*/ 0);
8712                         ctl_done((union ctl_io *)ctsio);
8713                         return (CTL_RETVAL_COMPLETE);
8714                 }
8715                 break;
8716         }
8717         case WRITE_VERIFY_16: {
8718                 struct scsi_write_verify_16 *cdb;
8719
8720                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8721                 flags |= CTL_LLF_FUA;
8722                 if (cdb->byte2 & SWV_DPO)
8723                         flags |= CTL_LLF_DPO;
8724                 lba = scsi_8btou64(cdb->addr);
8725                 num_blocks = scsi_4btoul(cdb->length);
8726                 break;
8727         }
8728         default:
8729                 /*
8730                  * We got a command we don't support.  This shouldn't
8731                  * happen, commands should be filtered out above us.
8732                  */
8733                 ctl_set_invalid_opcode(ctsio);
8734                 ctl_done((union ctl_io *)ctsio);
8735
8736                 return (CTL_RETVAL_COMPLETE);
8737                 break; /* NOTREACHED */
8738         }
8739
8740         /*
8741          * The first check is to make sure we're in bounds, the second
8742          * check is to catch wrap-around problems.  If the lba + num blocks
8743          * is less than the lba, then we've wrapped around and the block
8744          * range is invalid anyway.
8745          */
8746         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8747          || ((lba + num_blocks) < lba)) {
8748                 ctl_set_lba_out_of_range(ctsio,
8749                     MAX(lba, lun->be_lun->maxlba + 1));
8750                 ctl_done((union ctl_io *)ctsio);
8751                 return (CTL_RETVAL_COMPLETE);
8752         }
8753
8754         /*
8755          * According to SBC-3, a transfer length of 0 is not an error.
8756          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8757          * translates to 256 blocks for those commands.
8758          */
8759         if (num_blocks == 0) {
8760                 ctl_set_success(ctsio);
8761                 ctl_done((union ctl_io *)ctsio);
8762                 return (CTL_RETVAL_COMPLETE);
8763         }
8764
8765         /* Set FUA and/or DPO if caches are disabled. */
8766         if (isread) {
8767                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8768                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8769         } else {
8770                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8771                         flags |= CTL_LLF_FUA;
8772         }
8773
8774         lbalen = (struct ctl_lba_len_flags *)
8775             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8776         lbalen->lba = lba;
8777         lbalen->len = num_blocks;
8778         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8779
8780         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8781         ctsio->kern_rel_offset = 0;
8782
8783         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8784
8785         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8786         return (retval);
8787 }
8788
8789 static int
8790 ctl_cnw_cont(union ctl_io *io)
8791 {
8792         struct ctl_lun *lun = CTL_LUN(io);
8793         struct ctl_scsiio *ctsio;
8794         struct ctl_lba_len_flags *lbalen;
8795         int retval;
8796
8797         ctsio = &io->scsiio;
8798         ctsio->io_hdr.status = CTL_STATUS_NONE;
8799         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8800         lbalen = (struct ctl_lba_len_flags *)
8801             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8802         lbalen->flags &= ~CTL_LLF_COMPARE;
8803         lbalen->flags |= CTL_LLF_WRITE;
8804
8805         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8806         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8807         return (retval);
8808 }
8809
8810 int
8811 ctl_cnw(struct ctl_scsiio *ctsio)
8812 {
8813         struct ctl_lun *lun = CTL_LUN(ctsio);
8814         struct ctl_lba_len_flags *lbalen;
8815         uint64_t lba;
8816         uint32_t num_blocks;
8817         int flags, retval;
8818
8819         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8820
8821         flags = 0;
8822         switch (ctsio->cdb[0]) {
8823         case COMPARE_AND_WRITE: {
8824                 struct scsi_compare_and_write *cdb;
8825
8826                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8827                 if (cdb->byte2 & SRW10_FUA)
8828                         flags |= CTL_LLF_FUA;
8829                 if (cdb->byte2 & SRW10_DPO)
8830                         flags |= CTL_LLF_DPO;
8831                 lba = scsi_8btou64(cdb->addr);
8832                 num_blocks = cdb->length;
8833                 break;
8834         }
8835         default:
8836                 /*
8837                  * We got a command we don't support.  This shouldn't
8838                  * happen, commands should be filtered out above us.
8839                  */
8840                 ctl_set_invalid_opcode(ctsio);
8841                 ctl_done((union ctl_io *)ctsio);
8842
8843                 return (CTL_RETVAL_COMPLETE);
8844                 break; /* NOTREACHED */
8845         }
8846
8847         /*
8848          * The first check is to make sure we're in bounds, the second
8849          * check is to catch wrap-around problems.  If the lba + num blocks
8850          * is less than the lba, then we've wrapped around and the block
8851          * range is invalid anyway.
8852          */
8853         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8854          || ((lba + num_blocks) < lba)) {
8855                 ctl_set_lba_out_of_range(ctsio,
8856                     MAX(lba, lun->be_lun->maxlba + 1));
8857                 ctl_done((union ctl_io *)ctsio);
8858                 return (CTL_RETVAL_COMPLETE);
8859         }
8860
8861         /*
8862          * According to SBC-3, a transfer length of 0 is not an error.
8863          */
8864         if (num_blocks == 0) {
8865                 ctl_set_success(ctsio);
8866                 ctl_done((union ctl_io *)ctsio);
8867                 return (CTL_RETVAL_COMPLETE);
8868         }
8869
8870         /* Set FUA if write cache is disabled. */
8871         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8872                 flags |= CTL_LLF_FUA;
8873
8874         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8875         ctsio->kern_rel_offset = 0;
8876
8877         /*
8878          * Set the IO_CONT flag, so that if this I/O gets passed to
8879          * ctl_data_submit_done(), it'll get passed back to
8880          * ctl_ctl_cnw_cont() for further processing.
8881          */
8882         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8883         ctsio->io_cont = ctl_cnw_cont;
8884
8885         lbalen = (struct ctl_lba_len_flags *)
8886             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8887         lbalen->lba = lba;
8888         lbalen->len = num_blocks;
8889         lbalen->flags = CTL_LLF_COMPARE | flags;
8890
8891         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8892         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8893         return (retval);
8894 }
8895
8896 int
8897 ctl_verify(struct ctl_scsiio *ctsio)
8898 {
8899         struct ctl_lun *lun = CTL_LUN(ctsio);
8900         struct ctl_lba_len_flags *lbalen;
8901         uint64_t lba;
8902         uint32_t num_blocks;
8903         int bytchk, flags;
8904         int retval;
8905
8906         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8907
8908         bytchk = 0;
8909         flags = CTL_LLF_FUA;
8910         switch (ctsio->cdb[0]) {
8911         case VERIFY_10: {
8912                 struct scsi_verify_10 *cdb;
8913
8914                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8915                 if (cdb->byte2 & SVFY_BYTCHK)
8916                         bytchk = 1;
8917                 if (cdb->byte2 & SVFY_DPO)
8918                         flags |= CTL_LLF_DPO;
8919                 lba = scsi_4btoul(cdb->addr);
8920                 num_blocks = scsi_2btoul(cdb->length);
8921                 break;
8922         }
8923         case VERIFY_12: {
8924                 struct scsi_verify_12 *cdb;
8925
8926                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8927                 if (cdb->byte2 & SVFY_BYTCHK)
8928                         bytchk = 1;
8929                 if (cdb->byte2 & SVFY_DPO)
8930                         flags |= CTL_LLF_DPO;
8931                 lba = scsi_4btoul(cdb->addr);
8932                 num_blocks = scsi_4btoul(cdb->length);
8933                 break;
8934         }
8935         case VERIFY_16: {
8936                 struct scsi_rw_16 *cdb;
8937
8938                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8939                 if (cdb->byte2 & SVFY_BYTCHK)
8940                         bytchk = 1;
8941                 if (cdb->byte2 & SVFY_DPO)
8942                         flags |= CTL_LLF_DPO;
8943                 lba = scsi_8btou64(cdb->addr);
8944                 num_blocks = scsi_4btoul(cdb->length);
8945                 break;
8946         }
8947         default:
8948                 /*
8949                  * We got a command we don't support.  This shouldn't
8950                  * happen, commands should be filtered out above us.
8951                  */
8952                 ctl_set_invalid_opcode(ctsio);
8953                 ctl_done((union ctl_io *)ctsio);
8954                 return (CTL_RETVAL_COMPLETE);
8955         }
8956
8957         /*
8958          * The first check is to make sure we're in bounds, the second
8959          * check is to catch wrap-around problems.  If the lba + num blocks
8960          * is less than the lba, then we've wrapped around and the block
8961          * range is invalid anyway.
8962          */
8963         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8964          || ((lba + num_blocks) < lba)) {
8965                 ctl_set_lba_out_of_range(ctsio,
8966                     MAX(lba, lun->be_lun->maxlba + 1));
8967                 ctl_done((union ctl_io *)ctsio);
8968                 return (CTL_RETVAL_COMPLETE);
8969         }
8970
8971         /*
8972          * According to SBC-3, a transfer length of 0 is not an error.
8973          */
8974         if (num_blocks == 0) {
8975                 ctl_set_success(ctsio);
8976                 ctl_done((union ctl_io *)ctsio);
8977                 return (CTL_RETVAL_COMPLETE);
8978         }
8979
8980         lbalen = (struct ctl_lba_len_flags *)
8981             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8982         lbalen->lba = lba;
8983         lbalen->len = num_blocks;
8984         if (bytchk) {
8985                 lbalen->flags = CTL_LLF_COMPARE | flags;
8986                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8987         } else {
8988                 lbalen->flags = CTL_LLF_VERIFY | flags;
8989                 ctsio->kern_total_len = 0;
8990         }
8991         ctsio->kern_rel_offset = 0;
8992
8993         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
8994         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8995         return (retval);
8996 }
8997
8998 int
8999 ctl_report_luns(struct ctl_scsiio *ctsio)
9000 {
9001         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9002         struct ctl_port *port = CTL_PORT(ctsio);
9003         struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio);
9004         struct scsi_report_luns *cdb;
9005         struct scsi_report_luns_data *lun_data;
9006         int num_filled, num_luns, num_port_luns, retval;
9007         uint32_t alloc_len, lun_datalen;
9008         uint32_t initidx, targ_lun_id, lun_id;
9009
9010         retval = CTL_RETVAL_COMPLETE;
9011         cdb = (struct scsi_report_luns *)ctsio->cdb;
9012
9013         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9014
9015         num_luns = 0;
9016         num_port_luns = port->lun_map ? port->lun_map_size : CTL_MAX_LUNS;
9017         mtx_lock(&softc->ctl_lock);
9018         for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) {
9019                 if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX)
9020                         num_luns++;
9021         }
9022         mtx_unlock(&softc->ctl_lock);
9023
9024         switch (cdb->select_report) {
9025         case RPL_REPORT_DEFAULT:
9026         case RPL_REPORT_ALL:
9027         case RPL_REPORT_NONSUBSID:
9028                 break;
9029         case RPL_REPORT_WELLKNOWN:
9030         case RPL_REPORT_ADMIN:
9031         case RPL_REPORT_CONGLOM:
9032                 num_luns = 0;
9033                 break;
9034         default:
9035                 ctl_set_invalid_field(ctsio,
9036                                       /*sks_valid*/ 1,
9037                                       /*command*/ 1,
9038                                       /*field*/ 2,
9039                                       /*bit_valid*/ 0,
9040                                       /*bit*/ 0);
9041                 ctl_done((union ctl_io *)ctsio);
9042                 return (retval);
9043                 break; /* NOTREACHED */
9044         }
9045
9046         alloc_len = scsi_4btoul(cdb->length);
9047         /*
9048          * The initiator has to allocate at least 16 bytes for this request,
9049          * so he can at least get the header and the first LUN.  Otherwise
9050          * we reject the request (per SPC-3 rev 14, section 6.21).
9051          */
9052         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9053             sizeof(struct scsi_report_luns_lundata))) {
9054                 ctl_set_invalid_field(ctsio,
9055                                       /*sks_valid*/ 1,
9056                                       /*command*/ 1,
9057                                       /*field*/ 6,
9058                                       /*bit_valid*/ 0,
9059                                       /*bit*/ 0);
9060                 ctl_done((union ctl_io *)ctsio);
9061                 return (retval);
9062         }
9063
9064         lun_datalen = sizeof(*lun_data) +
9065                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9066
9067         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9068         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9069         ctsio->kern_sg_entries = 0;
9070
9071         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9072
9073         mtx_lock(&softc->ctl_lock);
9074         for (targ_lun_id = 0, num_filled = 0;
9075             targ_lun_id < num_port_luns && num_filled < num_luns;
9076             targ_lun_id++) {
9077                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9078                 if (lun_id == UINT32_MAX)
9079                         continue;
9080                 lun = softc->ctl_luns[lun_id];
9081                 if (lun == NULL)
9082                         continue;
9083
9084                 be64enc(lun_data->luns[num_filled++].lundata,
9085                     ctl_encode_lun(targ_lun_id));
9086
9087                 /*
9088                  * According to SPC-3, rev 14 section 6.21:
9089                  *
9090                  * "The execution of a REPORT LUNS command to any valid and
9091                  * installed logical unit shall clear the REPORTED LUNS DATA
9092                  * HAS CHANGED unit attention condition for all logical
9093                  * units of that target with respect to the requesting
9094                  * initiator. A valid and installed logical unit is one
9095                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9096                  * INQUIRY data (see 6.4.2)."
9097                  *
9098                  * If request_lun is NULL, the LUN this report luns command
9099                  * was issued to is either disabled or doesn't exist. In that
9100                  * case, we shouldn't clear any pending lun change unit
9101                  * attention.
9102                  */
9103                 if (request_lun != NULL) {
9104                         mtx_lock(&lun->lun_lock);
9105                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9106                         mtx_unlock(&lun->lun_lock);
9107                 }
9108         }
9109         mtx_unlock(&softc->ctl_lock);
9110
9111         /*
9112          * It's quite possible that we've returned fewer LUNs than we allocated
9113          * space for.  Trim it.
9114          */
9115         lun_datalen = sizeof(*lun_data) +
9116                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9117         ctsio->kern_rel_offset = 0;
9118         ctsio->kern_sg_entries = 0;
9119         ctsio->kern_data_len = min(lun_datalen, alloc_len);
9120         ctsio->kern_total_len = ctsio->kern_data_len;
9121
9122         /*
9123          * We set this to the actual data length, regardless of how much
9124          * space we actually have to return results.  If the user looks at
9125          * this value, he'll know whether or not he allocated enough space
9126          * and reissue the command if necessary.  We don't support well
9127          * known logical units, so if the user asks for that, return none.
9128          */
9129         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9130
9131         /*
9132          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9133          * this request.
9134          */
9135         ctl_set_success(ctsio);
9136         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9137         ctsio->be_move_done = ctl_config_move_done;
9138         ctl_datamove((union ctl_io *)ctsio);
9139         return (retval);
9140 }
9141
9142 int
9143 ctl_request_sense(struct ctl_scsiio *ctsio)
9144 {
9145         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9146         struct ctl_lun *lun = CTL_LUN(ctsio);
9147         struct scsi_request_sense *cdb;
9148         struct scsi_sense_data *sense_ptr;
9149         uint32_t initidx;
9150         int have_error;
9151         u_int sense_len = SSD_FULL_SIZE;
9152         scsi_sense_data_type sense_format;
9153         ctl_ua_type ua_type;
9154         uint8_t asc = 0, ascq = 0;
9155
9156         cdb = (struct scsi_request_sense *)ctsio->cdb;
9157
9158         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9159
9160         /*
9161          * Determine which sense format the user wants.
9162          */
9163         if (cdb->byte2 & SRS_DESC)
9164                 sense_format = SSD_TYPE_DESC;
9165         else
9166                 sense_format = SSD_TYPE_FIXED;
9167
9168         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9169         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9170         ctsio->kern_sg_entries = 0;
9171         ctsio->kern_rel_offset = 0;
9172
9173         /*
9174          * struct scsi_sense_data, which is currently set to 256 bytes, is
9175          * larger than the largest allowed value for the length field in the
9176          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9177          */
9178         ctsio->kern_data_len = cdb->length;
9179         ctsio->kern_total_len = cdb->length;
9180
9181         /*
9182          * If we don't have a LUN, we don't have any pending sense.
9183          */
9184         if (lun == NULL ||
9185             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9186              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9187                 /* "Logical unit not supported" */
9188                 ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9189                     /*current_error*/ 1,
9190                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9191                     /*asc*/ 0x25,
9192                     /*ascq*/ 0x00,
9193                     SSD_ELEM_NONE);
9194                 goto send;
9195         }
9196
9197         have_error = 0;
9198         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9199         /*
9200          * Check for pending sense, and then for pending unit attentions.
9201          * Pending sense gets returned first, then pending unit attentions.
9202          */
9203         mtx_lock(&lun->lun_lock);
9204 #ifdef CTL_WITH_CA
9205         if (ctl_is_set(lun->have_ca, initidx)) {
9206                 scsi_sense_data_type stored_format;
9207
9208                 /*
9209                  * Check to see which sense format was used for the stored
9210                  * sense data.
9211                  */
9212                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9213
9214                 /*
9215                  * If the user requested a different sense format than the
9216                  * one we stored, then we need to convert it to the other
9217                  * format.  If we're going from descriptor to fixed format
9218                  * sense data, we may lose things in translation, depending
9219                  * on what options were used.
9220                  *
9221                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9222                  * for some reason we'll just copy it out as-is.
9223                  */
9224                 if ((stored_format == SSD_TYPE_FIXED)
9225                  && (sense_format == SSD_TYPE_DESC))
9226                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9227                             &lun->pending_sense[initidx],
9228                             (struct scsi_sense_data_desc *)sense_ptr);
9229                 else if ((stored_format == SSD_TYPE_DESC)
9230                       && (sense_format == SSD_TYPE_FIXED))
9231                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9232                             &lun->pending_sense[initidx],
9233                             (struct scsi_sense_data_fixed *)sense_ptr);
9234                 else
9235                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9236                                MIN(sizeof(*sense_ptr),
9237                                sizeof(lun->pending_sense[initidx])));
9238
9239                 ctl_clear_mask(lun->have_ca, initidx);
9240                 have_error = 1;
9241         } else
9242 #endif
9243         if (have_error == 0) {
9244                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9245                     sense_format);
9246                 if (ua_type != CTL_UA_NONE)
9247                         have_error = 1;
9248         }
9249         if (have_error == 0) {
9250                 /*
9251                  * Report informational exception if have one and allowed.
9252                  */
9253                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9254                         asc = lun->ie_asc;
9255                         ascq = lun->ie_ascq;
9256                 }
9257                 ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9258                     /*current_error*/ 1,
9259                     /*sense_key*/ SSD_KEY_NO_SENSE,
9260                     /*asc*/ asc,
9261                     /*ascq*/ ascq,
9262                     SSD_ELEM_NONE);
9263         }
9264         mtx_unlock(&lun->lun_lock);
9265
9266 send:
9267         /*
9268          * We report the SCSI status as OK, since the status of the command
9269          * itself is OK.  We're reporting sense as parameter data.
9270          */
9271         ctl_set_success(ctsio);
9272         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9273         ctsio->be_move_done = ctl_config_move_done;
9274         ctl_datamove((union ctl_io *)ctsio);
9275         return (CTL_RETVAL_COMPLETE);
9276 }
9277
9278 int
9279 ctl_tur(struct ctl_scsiio *ctsio)
9280 {
9281
9282         CTL_DEBUG_PRINT(("ctl_tur\n"));
9283
9284         ctl_set_success(ctsio);
9285         ctl_done((union ctl_io *)ctsio);
9286
9287         return (CTL_RETVAL_COMPLETE);
9288 }
9289
9290 /*
9291  * SCSI VPD page 0x00, the Supported VPD Pages page.
9292  */
9293 static int
9294 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9295 {
9296         struct ctl_lun *lun = CTL_LUN(ctsio);
9297         struct scsi_vpd_supported_pages *pages;
9298         int sup_page_size;
9299         int p;
9300
9301         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9302             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9303         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9304         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9305         ctsio->kern_rel_offset = 0;
9306         ctsio->kern_sg_entries = 0;
9307         ctsio->kern_data_len = min(sup_page_size, alloc_len);
9308         ctsio->kern_total_len = ctsio->kern_data_len;
9309
9310         /*
9311          * The control device is always connected.  The disk device, on the
9312          * other hand, may not be online all the time.  Need to change this
9313          * to figure out whether the disk device is actually online or not.
9314          */
9315         if (lun != NULL)
9316                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9317                                 lun->be_lun->lun_type;
9318         else
9319                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9320
9321         p = 0;
9322         /* Supported VPD pages */
9323         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9324         /* Serial Number */
9325         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9326         /* Device Identification */
9327         pages->page_list[p++] = SVPD_DEVICE_ID;
9328         /* Extended INQUIRY Data */
9329         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9330         /* Mode Page Policy */
9331         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9332         /* SCSI Ports */
9333         pages->page_list[p++] = SVPD_SCSI_PORTS;
9334         /* Third-party Copy */
9335         pages->page_list[p++] = SVPD_SCSI_TPC;
9336         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9337                 /* Block limits */
9338                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9339                 /* Block Device Characteristics */
9340                 pages->page_list[p++] = SVPD_BDC;
9341                 /* Logical Block Provisioning */
9342                 pages->page_list[p++] = SVPD_LBP;
9343         }
9344         pages->length = p;
9345
9346         ctl_set_success(ctsio);
9347         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9348         ctsio->be_move_done = ctl_config_move_done;
9349         ctl_datamove((union ctl_io *)ctsio);
9350         return (CTL_RETVAL_COMPLETE);
9351 }
9352
9353 /*
9354  * SCSI VPD page 0x80, the Unit Serial Number page.
9355  */
9356 static int
9357 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9358 {
9359         struct ctl_lun *lun = CTL_LUN(ctsio);
9360         struct scsi_vpd_unit_serial_number *sn_ptr;
9361         int data_len;
9362
9363         data_len = 4 + CTL_SN_LEN;
9364         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9365         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9366         ctsio->kern_rel_offset = 0;
9367         ctsio->kern_sg_entries = 0;
9368         ctsio->kern_data_len = min(data_len, alloc_len);
9369         ctsio->kern_total_len = ctsio->kern_data_len;
9370
9371         /*
9372          * The control device is always connected.  The disk device, on the
9373          * other hand, may not be online all the time.  Need to change this
9374          * to figure out whether the disk device is actually online or not.
9375          */
9376         if (lun != NULL)
9377                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9378                                   lun->be_lun->lun_type;
9379         else
9380                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9381
9382         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9383         sn_ptr->length = CTL_SN_LEN;
9384         /*
9385          * If we don't have a LUN, we just leave the serial number as
9386          * all spaces.
9387          */
9388         if (lun != NULL) {
9389                 strncpy((char *)sn_ptr->serial_num,
9390                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9391         } else
9392                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9393
9394         ctl_set_success(ctsio);
9395         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9396         ctsio->be_move_done = ctl_config_move_done;
9397         ctl_datamove((union ctl_io *)ctsio);
9398         return (CTL_RETVAL_COMPLETE);
9399 }
9400
9401
9402 /*
9403  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9404  */
9405 static int
9406 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9407 {
9408         struct ctl_lun *lun = CTL_LUN(ctsio);
9409         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9410         int data_len;
9411
9412         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9413         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9414         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9415         ctsio->kern_sg_entries = 0;
9416         ctsio->kern_rel_offset = 0;
9417         ctsio->kern_data_len = min(data_len, alloc_len);
9418         ctsio->kern_total_len = ctsio->kern_data_len;
9419
9420         /*
9421          * The control device is always connected.  The disk device, on the
9422          * other hand, may not be online all the time.
9423          */
9424         if (lun != NULL)
9425                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9426                                      lun->be_lun->lun_type;
9427         else
9428                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9429         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9430         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9431         /*
9432          * We support head of queue, ordered and simple tags.
9433          */
9434         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9435         /*
9436          * Volatile cache supported.
9437          */
9438         eid_ptr->flags3 = SVPD_EID_V_SUP;
9439
9440         /*
9441          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9442          * attention for a particular IT nexus on all LUNs once we report
9443          * it to that nexus once.  This bit is required as of SPC-4.
9444          */
9445         eid_ptr->flags4 = SVPD_EID_LUICLR;
9446
9447         /*
9448          * We support revert to defaults (RTD) bit in MODE SELECT.
9449          */
9450         eid_ptr->flags5 = SVPD_EID_RTD_SUP;
9451
9452         /*
9453          * XXX KDM in order to correctly answer this, we would need
9454          * information from the SIM to determine how much sense data it
9455          * can send.  So this would really be a path inquiry field, most
9456          * likely.  This can be set to a maximum of 252 according to SPC-4,
9457          * but the hardware may or may not be able to support that much.
9458          * 0 just means that the maximum sense data length is not reported.
9459          */
9460         eid_ptr->max_sense_length = 0;
9461
9462         ctl_set_success(ctsio);
9463         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9464         ctsio->be_move_done = ctl_config_move_done;
9465         ctl_datamove((union ctl_io *)ctsio);
9466         return (CTL_RETVAL_COMPLETE);
9467 }
9468
9469 static int
9470 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9471 {
9472         struct ctl_lun *lun = CTL_LUN(ctsio);
9473         struct scsi_vpd_mode_page_policy *mpp_ptr;
9474         int data_len;
9475
9476         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9477             sizeof(struct scsi_vpd_mode_page_policy_descr);
9478
9479         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9480         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9481         ctsio->kern_rel_offset = 0;
9482         ctsio->kern_sg_entries = 0;
9483         ctsio->kern_data_len = min(data_len, alloc_len);
9484         ctsio->kern_total_len = ctsio->kern_data_len;
9485
9486         /*
9487          * The control device is always connected.  The disk device, on the
9488          * other hand, may not be online all the time.
9489          */
9490         if (lun != NULL)
9491                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9492                                      lun->be_lun->lun_type;
9493         else
9494                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9495         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9496         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9497         mpp_ptr->descr[0].page_code = 0x3f;
9498         mpp_ptr->descr[0].subpage_code = 0xff;
9499         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9500
9501         ctl_set_success(ctsio);
9502         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9503         ctsio->be_move_done = ctl_config_move_done;
9504         ctl_datamove((union ctl_io *)ctsio);
9505         return (CTL_RETVAL_COMPLETE);
9506 }
9507
9508 /*
9509  * SCSI VPD page 0x83, the Device Identification page.
9510  */
9511 static int
9512 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9513 {
9514         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9515         struct ctl_port *port = CTL_PORT(ctsio);
9516         struct ctl_lun *lun = CTL_LUN(ctsio);
9517         struct scsi_vpd_device_id *devid_ptr;
9518         struct scsi_vpd_id_descriptor *desc;
9519         int data_len, g;
9520         uint8_t proto;
9521
9522         data_len = sizeof(struct scsi_vpd_device_id) +
9523             sizeof(struct scsi_vpd_id_descriptor) +
9524                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9525             sizeof(struct scsi_vpd_id_descriptor) +
9526                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9527         if (lun && lun->lun_devid)
9528                 data_len += lun->lun_devid->len;
9529         if (port && port->port_devid)
9530                 data_len += port->port_devid->len;
9531         if (port && port->target_devid)
9532                 data_len += port->target_devid->len;
9533
9534         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9535         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9536         ctsio->kern_sg_entries = 0;
9537         ctsio->kern_rel_offset = 0;
9538         ctsio->kern_sg_entries = 0;
9539         ctsio->kern_data_len = min(data_len, alloc_len);
9540         ctsio->kern_total_len = ctsio->kern_data_len;
9541
9542         /*
9543          * The control device is always connected.  The disk device, on the
9544          * other hand, may not be online all the time.
9545          */
9546         if (lun != NULL)
9547                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9548                                      lun->be_lun->lun_type;
9549         else
9550                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9551         devid_ptr->page_code = SVPD_DEVICE_ID;
9552         scsi_ulto2b(data_len - 4, devid_ptr->length);
9553
9554         if (port && port->port_type == CTL_PORT_FC)
9555                 proto = SCSI_PROTO_FC << 4;
9556         else if (port && port->port_type == CTL_PORT_ISCSI)
9557                 proto = SCSI_PROTO_ISCSI << 4;
9558         else
9559                 proto = SCSI_PROTO_SPI << 4;
9560         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9561
9562         /*
9563          * We're using a LUN association here.  i.e., this device ID is a
9564          * per-LUN identifier.
9565          */
9566         if (lun && lun->lun_devid) {
9567                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9568                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9569                     lun->lun_devid->len);
9570         }
9571
9572         /*
9573          * This is for the WWPN which is a port association.
9574          */
9575         if (port && port->port_devid) {
9576                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9577                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9578                     port->port_devid->len);
9579         }
9580
9581         /*
9582          * This is for the Relative Target Port(type 4h) identifier
9583          */
9584         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9585         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9586             SVPD_ID_TYPE_RELTARG;
9587         desc->length = 4;
9588         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9589         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9590             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9591
9592         /*
9593          * This is for the Target Port Group(type 5h) identifier
9594          */
9595         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9596         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9597             SVPD_ID_TYPE_TPORTGRP;
9598         desc->length = 4;
9599         if (softc->is_single ||
9600             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9601                 g = 1;
9602         else
9603                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9604         scsi_ulto2b(g, &desc->identifier[2]);
9605         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9606             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9607
9608         /*
9609          * This is for the Target identifier
9610          */
9611         if (port && port->target_devid) {
9612                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9613         }
9614
9615         ctl_set_success(ctsio);
9616         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9617         ctsio->be_move_done = ctl_config_move_done;
9618         ctl_datamove((union ctl_io *)ctsio);
9619         return (CTL_RETVAL_COMPLETE);
9620 }
9621
9622 static int
9623 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9624 {
9625         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9626         struct ctl_lun *lun = CTL_LUN(ctsio);
9627         struct scsi_vpd_scsi_ports *sp;
9628         struct scsi_vpd_port_designation *pd;
9629         struct scsi_vpd_port_designation_cont *pdc;
9630         struct ctl_port *port;
9631         int data_len, num_target_ports, iid_len, id_len;
9632
9633         num_target_ports = 0;
9634         iid_len = 0;
9635         id_len = 0;
9636         mtx_lock(&softc->ctl_lock);
9637         STAILQ_FOREACH(port, &softc->port_list, links) {
9638                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9639                         continue;
9640                 if (lun != NULL &&
9641                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9642                         continue;
9643                 num_target_ports++;
9644                 if (port->init_devid)
9645                         iid_len += port->init_devid->len;
9646                 if (port->port_devid)
9647                         id_len += port->port_devid->len;
9648         }
9649         mtx_unlock(&softc->ctl_lock);
9650
9651         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9652             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9653              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9654         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9655         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9656         ctsio->kern_sg_entries = 0;
9657         ctsio->kern_rel_offset = 0;
9658         ctsio->kern_sg_entries = 0;
9659         ctsio->kern_data_len = min(data_len, alloc_len);
9660         ctsio->kern_total_len = ctsio->kern_data_len;
9661
9662         /*
9663          * The control device is always connected.  The disk device, on the
9664          * other hand, may not be online all the time.  Need to change this
9665          * to figure out whether the disk device is actually online or not.
9666          */
9667         if (lun != NULL)
9668                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9669                                   lun->be_lun->lun_type;
9670         else
9671                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9672
9673         sp->page_code = SVPD_SCSI_PORTS;
9674         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9675             sp->page_length);
9676         pd = &sp->design[0];
9677
9678         mtx_lock(&softc->ctl_lock);
9679         STAILQ_FOREACH(port, &softc->port_list, links) {
9680                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9681                         continue;
9682                 if (lun != NULL &&
9683                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9684                         continue;
9685                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9686                 if (port->init_devid) {
9687                         iid_len = port->init_devid->len;
9688                         memcpy(pd->initiator_transportid,
9689                             port->init_devid->data, port->init_devid->len);
9690                 } else
9691                         iid_len = 0;
9692                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9693                 pdc = (struct scsi_vpd_port_designation_cont *)
9694                     (&pd->initiator_transportid[iid_len]);
9695                 if (port->port_devid) {
9696                         id_len = port->port_devid->len;
9697                         memcpy(pdc->target_port_descriptors,
9698                             port->port_devid->data, port->port_devid->len);
9699                 } else
9700                         id_len = 0;
9701                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9702                 pd = (struct scsi_vpd_port_designation *)
9703                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9704         }
9705         mtx_unlock(&softc->ctl_lock);
9706
9707         ctl_set_success(ctsio);
9708         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9709         ctsio->be_move_done = ctl_config_move_done;
9710         ctl_datamove((union ctl_io *)ctsio);
9711         return (CTL_RETVAL_COMPLETE);
9712 }
9713
9714 static int
9715 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9716 {
9717         struct ctl_lun *lun = CTL_LUN(ctsio);
9718         struct scsi_vpd_block_limits *bl_ptr;
9719         uint64_t ival;
9720
9721         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9722         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9723         ctsio->kern_sg_entries = 0;
9724         ctsio->kern_rel_offset = 0;
9725         ctsio->kern_sg_entries = 0;
9726         ctsio->kern_data_len = min(sizeof(*bl_ptr), alloc_len);
9727         ctsio->kern_total_len = ctsio->kern_data_len;
9728
9729         /*
9730          * The control device is always connected.  The disk device, on the
9731          * other hand, may not be online all the time.  Need to change this
9732          * to figure out whether the disk device is actually online or not.
9733          */
9734         if (lun != NULL)
9735                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9736                                   lun->be_lun->lun_type;
9737         else
9738                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9739
9740         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9741         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9742         bl_ptr->max_cmp_write_len = 0xff;
9743         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9744         if (lun != NULL) {
9745                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9746                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9747                         ival = 0xffffffff;
9748                         ctl_get_opt_number(&lun->be_lun->options,
9749                             "unmap_max_lba", &ival);
9750                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9751                         ival = 0xffffffff;
9752                         ctl_get_opt_number(&lun->be_lun->options,
9753                             "unmap_max_descr", &ival);
9754                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9755                         if (lun->be_lun->ublockexp != 0) {
9756                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9757                                     bl_ptr->opt_unmap_grain);
9758                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9759                                     bl_ptr->unmap_grain_align);
9760                         }
9761                 }
9762                 scsi_ulto4b(lun->be_lun->atomicblock,
9763                     bl_ptr->max_atomic_transfer_length);
9764                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9765                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9766                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9767                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9768                 ival = UINT64_MAX;
9769                 ctl_get_opt_number(&lun->be_lun->options, "write_same_max_lba", &ival);
9770                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9771         }
9772
9773         ctl_set_success(ctsio);
9774         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9775         ctsio->be_move_done = ctl_config_move_done;
9776         ctl_datamove((union ctl_io *)ctsio);
9777         return (CTL_RETVAL_COMPLETE);
9778 }
9779
9780 static int
9781 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9782 {
9783         struct ctl_lun *lun = CTL_LUN(ctsio);
9784         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9785         const char *value;
9786         u_int i;
9787
9788         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9789         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9790         ctsio->kern_sg_entries = 0;
9791         ctsio->kern_rel_offset = 0;
9792         ctsio->kern_data_len = min(sizeof(*bdc_ptr), alloc_len);
9793         ctsio->kern_total_len = ctsio->kern_data_len;
9794
9795         /*
9796          * The control device is always connected.  The disk device, on the
9797          * other hand, may not be online all the time.  Need to change this
9798          * to figure out whether the disk device is actually online or not.
9799          */
9800         if (lun != NULL)
9801                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9802                                   lun->be_lun->lun_type;
9803         else
9804                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9805         bdc_ptr->page_code = SVPD_BDC;
9806         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9807         if (lun != NULL &&
9808             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9809                 i = strtol(value, NULL, 0);
9810         else
9811                 i = CTL_DEFAULT_ROTATION_RATE;
9812         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9813         if (lun != NULL &&
9814             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9815                 i = strtol(value, NULL, 0);
9816         else
9817                 i = 0;
9818         bdc_ptr->wab_wac_ff = (i & 0x0f);
9819         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9820
9821         ctl_set_success(ctsio);
9822         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9823         ctsio->be_move_done = ctl_config_move_done;
9824         ctl_datamove((union ctl_io *)ctsio);
9825         return (CTL_RETVAL_COMPLETE);
9826 }
9827
9828 static int
9829 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9830 {
9831         struct ctl_lun *lun = CTL_LUN(ctsio);
9832         struct scsi_vpd_logical_block_prov *lbp_ptr;
9833         const char *value;
9834
9835         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9836         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9837         ctsio->kern_sg_entries = 0;
9838         ctsio->kern_rel_offset = 0;
9839         ctsio->kern_data_len = min(sizeof(*lbp_ptr), alloc_len);
9840         ctsio->kern_total_len = ctsio->kern_data_len;
9841
9842         /*
9843          * The control device is always connected.  The disk device, on the
9844          * other hand, may not be online all the time.  Need to change this
9845          * to figure out whether the disk device is actually online or not.
9846          */
9847         if (lun != NULL)
9848                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9849                                   lun->be_lun->lun_type;
9850         else
9851                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9852
9853         lbp_ptr->page_code = SVPD_LBP;
9854         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9855         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
9856         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9857                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
9858                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
9859                 value = ctl_get_opt(&lun->be_lun->options, "provisioning_type");
9860                 if (value != NULL) {
9861                         if (strcmp(value, "resource") == 0)
9862                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
9863                         else if (strcmp(value, "thin") == 0)
9864                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
9865                 } else
9866                         lbp_ptr->prov_type = SVPD_LBP_THIN;
9867         }
9868
9869         ctl_set_success(ctsio);
9870         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9871         ctsio->be_move_done = ctl_config_move_done;
9872         ctl_datamove((union ctl_io *)ctsio);
9873         return (CTL_RETVAL_COMPLETE);
9874 }
9875
9876 /*
9877  * INQUIRY with the EVPD bit set.
9878  */
9879 static int
9880 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9881 {
9882         struct ctl_lun *lun = CTL_LUN(ctsio);
9883         struct scsi_inquiry *cdb;
9884         int alloc_len, retval;
9885
9886         cdb = (struct scsi_inquiry *)ctsio->cdb;
9887         alloc_len = scsi_2btoul(cdb->length);
9888
9889         switch (cdb->page_code) {
9890         case SVPD_SUPPORTED_PAGES:
9891                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9892                 break;
9893         case SVPD_UNIT_SERIAL_NUMBER:
9894                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9895                 break;
9896         case SVPD_DEVICE_ID:
9897                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9898                 break;
9899         case SVPD_EXTENDED_INQUIRY_DATA:
9900                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
9901                 break;
9902         case SVPD_MODE_PAGE_POLICY:
9903                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
9904                 break;
9905         case SVPD_SCSI_PORTS:
9906                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
9907                 break;
9908         case SVPD_SCSI_TPC:
9909                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
9910                 break;
9911         case SVPD_BLOCK_LIMITS:
9912                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9913                         goto err;
9914                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9915                 break;
9916         case SVPD_BDC:
9917                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9918                         goto err;
9919                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
9920                 break;
9921         case SVPD_LBP:
9922                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9923                         goto err;
9924                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9925                 break;
9926         default:
9927 err:
9928                 ctl_set_invalid_field(ctsio,
9929                                       /*sks_valid*/ 1,
9930                                       /*command*/ 1,
9931                                       /*field*/ 2,
9932                                       /*bit_valid*/ 0,
9933                                       /*bit*/ 0);
9934                 ctl_done((union ctl_io *)ctsio);
9935                 retval = CTL_RETVAL_COMPLETE;
9936                 break;
9937         }
9938
9939         return (retval);
9940 }
9941
9942 /*
9943  * Standard INQUIRY data.
9944  */
9945 static int
9946 ctl_inquiry_std(struct ctl_scsiio *ctsio)
9947 {
9948         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9949         struct ctl_port *port = CTL_PORT(ctsio);
9950         struct ctl_lun *lun = CTL_LUN(ctsio);
9951         struct scsi_inquiry_data *inq_ptr;
9952         struct scsi_inquiry *cdb;
9953         char *val;
9954         uint32_t alloc_len, data_len;
9955         ctl_port_type port_type;
9956
9957         port_type = port->port_type;
9958         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
9959                 port_type = CTL_PORT_SCSI;
9960
9961         cdb = (struct scsi_inquiry *)ctsio->cdb;
9962         alloc_len = scsi_2btoul(cdb->length);
9963
9964         /*
9965          * We malloc the full inquiry data size here and fill it
9966          * in.  If the user only asks for less, we'll give him
9967          * that much.
9968          */
9969         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
9970         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9971         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9972         ctsio->kern_sg_entries = 0;
9973         ctsio->kern_rel_offset = 0;
9974         ctsio->kern_data_len = min(data_len, alloc_len);
9975         ctsio->kern_total_len = ctsio->kern_data_len;
9976
9977         if (lun != NULL) {
9978                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
9979                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
9980                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9981                             lun->be_lun->lun_type;
9982                 } else {
9983                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
9984                             lun->be_lun->lun_type;
9985                 }
9986                 if (lun->flags & CTL_LUN_REMOVABLE)
9987                         inq_ptr->dev_qual2 |= SID_RMB;
9988         } else
9989                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9990
9991         /* RMB in byte 2 is 0 */
9992         inq_ptr->version = SCSI_REV_SPC5;
9993
9994         /*
9995          * According to SAM-3, even if a device only supports a single
9996          * level of LUN addressing, it should still set the HISUP bit:
9997          *
9998          * 4.9.1 Logical unit numbers overview
9999          *
10000          * All logical unit number formats described in this standard are
10001          * hierarchical in structure even when only a single level in that
10002          * hierarchy is used. The HISUP bit shall be set to one in the
10003          * standard INQUIRY data (see SPC-2) when any logical unit number
10004          * format described in this standard is used.  Non-hierarchical
10005          * formats are outside the scope of this standard.
10006          *
10007          * Therefore we set the HiSup bit here.
10008          *
10009          * The response format is 2, per SPC-3.
10010          */
10011         inq_ptr->response_format = SID_HiSup | 2;
10012
10013         inq_ptr->additional_length = data_len -
10014             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10015         CTL_DEBUG_PRINT(("additional_length = %d\n",
10016                          inq_ptr->additional_length));
10017
10018         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10019         if (port_type == CTL_PORT_SCSI)
10020                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10021         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10022         inq_ptr->flags = SID_CmdQue;
10023         if (port_type == CTL_PORT_SCSI)
10024                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10025
10026         /*
10027          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10028          * We have 8 bytes for the vendor name, and 16 bytes for the device
10029          * name and 4 bytes for the revision.
10030          */
10031         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10032             "vendor")) == NULL) {
10033                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10034         } else {
10035                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10036                 strncpy(inq_ptr->vendor, val,
10037                     min(sizeof(inq_ptr->vendor), strlen(val)));
10038         }
10039         if (lun == NULL) {
10040                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10041                     sizeof(inq_ptr->product));
10042         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10043                 switch (lun->be_lun->lun_type) {
10044                 case T_DIRECT:
10045                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10046                             sizeof(inq_ptr->product));
10047                         break;
10048                 case T_PROCESSOR:
10049                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10050                             sizeof(inq_ptr->product));
10051                         break;
10052                 case T_CDROM:
10053                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10054                             sizeof(inq_ptr->product));
10055                         break;
10056                 default:
10057                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10058                             sizeof(inq_ptr->product));
10059                         break;
10060                 }
10061         } else {
10062                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10063                 strncpy(inq_ptr->product, val,
10064                     min(sizeof(inq_ptr->product), strlen(val)));
10065         }
10066
10067         /*
10068          * XXX make this a macro somewhere so it automatically gets
10069          * incremented when we make changes.
10070          */
10071         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10072             "revision")) == NULL) {
10073                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10074         } else {
10075                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10076                 strncpy(inq_ptr->revision, val,
10077                     min(sizeof(inq_ptr->revision), strlen(val)));
10078         }
10079
10080         /*
10081          * For parallel SCSI, we support double transition and single
10082          * transition clocking.  We also support QAS (Quick Arbitration
10083          * and Selection) and Information Unit transfers on both the
10084          * control and array devices.
10085          */
10086         if (port_type == CTL_PORT_SCSI)
10087                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10088                                     SID_SPI_IUS;
10089
10090         /* SAM-6 (no version claimed) */
10091         scsi_ulto2b(0x00C0, inq_ptr->version1);
10092         /* SPC-5 (no version claimed) */
10093         scsi_ulto2b(0x05C0, inq_ptr->version2);
10094         if (port_type == CTL_PORT_FC) {
10095                 /* FCP-2 ANSI INCITS.350:2003 */
10096                 scsi_ulto2b(0x0917, inq_ptr->version3);
10097         } else if (port_type == CTL_PORT_SCSI) {
10098                 /* SPI-4 ANSI INCITS.362:200x */
10099                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10100         } else if (port_type == CTL_PORT_ISCSI) {
10101                 /* iSCSI (no version claimed) */
10102                 scsi_ulto2b(0x0960, inq_ptr->version3);
10103         } else if (port_type == CTL_PORT_SAS) {
10104                 /* SAS (no version claimed) */
10105                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10106         } else if (port_type == CTL_PORT_UMASS) {
10107                 /* USB Mass Storage Class Bulk-Only Transport, Revision 1.0 */
10108                 scsi_ulto2b(0x1730, inq_ptr->version3);
10109         }
10110
10111         if (lun == NULL) {
10112                 /* SBC-4 (no version claimed) */
10113                 scsi_ulto2b(0x0600, inq_ptr->version4);
10114         } else {
10115                 switch (lun->be_lun->lun_type) {
10116                 case T_DIRECT:
10117                         /* SBC-4 (no version claimed) */
10118                         scsi_ulto2b(0x0600, inq_ptr->version4);
10119                         break;
10120                 case T_PROCESSOR:
10121                         break;
10122                 case T_CDROM:
10123                         /* MMC-6 (no version claimed) */
10124                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10125                         break;
10126                 default:
10127                         break;
10128                 }
10129         }
10130
10131         ctl_set_success(ctsio);
10132         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10133         ctsio->be_move_done = ctl_config_move_done;
10134         ctl_datamove((union ctl_io *)ctsio);
10135         return (CTL_RETVAL_COMPLETE);
10136 }
10137
10138 int
10139 ctl_inquiry(struct ctl_scsiio *ctsio)
10140 {
10141         struct scsi_inquiry *cdb;
10142         int retval;
10143
10144         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10145
10146         cdb = (struct scsi_inquiry *)ctsio->cdb;
10147         if (cdb->byte2 & SI_EVPD)
10148                 retval = ctl_inquiry_evpd(ctsio);
10149         else if (cdb->page_code == 0)
10150                 retval = ctl_inquiry_std(ctsio);
10151         else {
10152                 ctl_set_invalid_field(ctsio,
10153                                       /*sks_valid*/ 1,
10154                                       /*command*/ 1,
10155                                       /*field*/ 2,
10156                                       /*bit_valid*/ 0,
10157                                       /*bit*/ 0);
10158                 ctl_done((union ctl_io *)ctsio);
10159                 return (CTL_RETVAL_COMPLETE);
10160         }
10161
10162         return (retval);
10163 }
10164
10165 int
10166 ctl_get_config(struct ctl_scsiio *ctsio)
10167 {
10168         struct ctl_lun *lun = CTL_LUN(ctsio);
10169         struct scsi_get_config_header *hdr;
10170         struct scsi_get_config_feature *feature;
10171         struct scsi_get_config *cdb;
10172         uint32_t alloc_len, data_len;
10173         int rt, starting;
10174
10175         cdb = (struct scsi_get_config *)ctsio->cdb;
10176         rt = (cdb->rt & SGC_RT_MASK);
10177         starting = scsi_2btoul(cdb->starting_feature);
10178         alloc_len = scsi_2btoul(cdb->length);
10179
10180         data_len = sizeof(struct scsi_get_config_header) +
10181             sizeof(struct scsi_get_config_feature) + 8 +
10182             sizeof(struct scsi_get_config_feature) + 8 +
10183             sizeof(struct scsi_get_config_feature) + 4 +
10184             sizeof(struct scsi_get_config_feature) + 4 +
10185             sizeof(struct scsi_get_config_feature) + 8 +
10186             sizeof(struct scsi_get_config_feature) +
10187             sizeof(struct scsi_get_config_feature) + 4 +
10188             sizeof(struct scsi_get_config_feature) + 4 +
10189             sizeof(struct scsi_get_config_feature) + 4 +
10190             sizeof(struct scsi_get_config_feature) + 4 +
10191             sizeof(struct scsi_get_config_feature) + 4 +
10192             sizeof(struct scsi_get_config_feature) + 4;
10193         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10194         ctsio->kern_sg_entries = 0;
10195         ctsio->kern_rel_offset = 0;
10196
10197         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10198         if (lun->flags & CTL_LUN_NO_MEDIA)
10199                 scsi_ulto2b(0x0000, hdr->current_profile);
10200         else
10201                 scsi_ulto2b(0x0010, hdr->current_profile);
10202         feature = (struct scsi_get_config_feature *)(hdr + 1);
10203
10204         if (starting > 0x003b)
10205                 goto done;
10206         if (starting > 0x003a)
10207                 goto f3b;
10208         if (starting > 0x002b)
10209                 goto f3a;
10210         if (starting > 0x002a)
10211                 goto f2b;
10212         if (starting > 0x001f)
10213                 goto f2a;
10214         if (starting > 0x001e)
10215                 goto f1f;
10216         if (starting > 0x001d)
10217                 goto f1e;
10218         if (starting > 0x0010)
10219                 goto f1d;
10220         if (starting > 0x0003)
10221                 goto f10;
10222         if (starting > 0x0002)
10223                 goto f3;
10224         if (starting > 0x0001)
10225                 goto f2;
10226         if (starting > 0x0000)
10227                 goto f1;
10228
10229         /* Profile List */
10230         scsi_ulto2b(0x0000, feature->feature_code);
10231         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10232         feature->add_length = 8;
10233         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10234         feature->feature_data[2] = 0x00;
10235         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10236         feature->feature_data[6] = 0x01;
10237         feature = (struct scsi_get_config_feature *)
10238             &feature->feature_data[feature->add_length];
10239
10240 f1:     /* Core */
10241         scsi_ulto2b(0x0001, feature->feature_code);
10242         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10243         feature->add_length = 8;
10244         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10245         feature->feature_data[4] = 0x03;
10246         feature = (struct scsi_get_config_feature *)
10247             &feature->feature_data[feature->add_length];
10248
10249 f2:     /* Morphing */
10250         scsi_ulto2b(0x0002, feature->feature_code);
10251         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10252         feature->add_length = 4;
10253         feature->feature_data[0] = 0x02;
10254         feature = (struct scsi_get_config_feature *)
10255             &feature->feature_data[feature->add_length];
10256
10257 f3:     /* Removable Medium */
10258         scsi_ulto2b(0x0003, feature->feature_code);
10259         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10260         feature->add_length = 4;
10261         feature->feature_data[0] = 0x39;
10262         feature = (struct scsi_get_config_feature *)
10263             &feature->feature_data[feature->add_length];
10264
10265         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10266                 goto done;
10267
10268 f10:    /* Random Read */
10269         scsi_ulto2b(0x0010, feature->feature_code);
10270         feature->flags = 0x00;
10271         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10272                 feature->flags |= SGC_F_CURRENT;
10273         feature->add_length = 8;
10274         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10275         scsi_ulto2b(1, &feature->feature_data[4]);
10276         feature->feature_data[6] = 0x00;
10277         feature = (struct scsi_get_config_feature *)
10278             &feature->feature_data[feature->add_length];
10279
10280 f1d:    /* Multi-Read */
10281         scsi_ulto2b(0x001D, feature->feature_code);
10282         feature->flags = 0x00;
10283         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10284                 feature->flags |= SGC_F_CURRENT;
10285         feature->add_length = 0;
10286         feature = (struct scsi_get_config_feature *)
10287             &feature->feature_data[feature->add_length];
10288
10289 f1e:    /* CD Read */
10290         scsi_ulto2b(0x001E, feature->feature_code);
10291         feature->flags = 0x00;
10292         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10293                 feature->flags |= SGC_F_CURRENT;
10294         feature->add_length = 4;
10295         feature->feature_data[0] = 0x00;
10296         feature = (struct scsi_get_config_feature *)
10297             &feature->feature_data[feature->add_length];
10298
10299 f1f:    /* DVD Read */
10300         scsi_ulto2b(0x001F, feature->feature_code);
10301         feature->flags = 0x08;
10302         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10303                 feature->flags |= SGC_F_CURRENT;
10304         feature->add_length = 4;
10305         feature->feature_data[0] = 0x01;
10306         feature->feature_data[2] = 0x03;
10307         feature = (struct scsi_get_config_feature *)
10308             &feature->feature_data[feature->add_length];
10309
10310 f2a:    /* DVD+RW */
10311         scsi_ulto2b(0x002A, feature->feature_code);
10312         feature->flags = 0x04;
10313         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10314                 feature->flags |= SGC_F_CURRENT;
10315         feature->add_length = 4;
10316         feature->feature_data[0] = 0x00;
10317         feature->feature_data[1] = 0x00;
10318         feature = (struct scsi_get_config_feature *)
10319             &feature->feature_data[feature->add_length];
10320
10321 f2b:    /* DVD+R */
10322         scsi_ulto2b(0x002B, feature->feature_code);
10323         feature->flags = 0x00;
10324         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10325                 feature->flags |= SGC_F_CURRENT;
10326         feature->add_length = 4;
10327         feature->feature_data[0] = 0x00;
10328         feature = (struct scsi_get_config_feature *)
10329             &feature->feature_data[feature->add_length];
10330
10331 f3a:    /* DVD+RW Dual Layer */
10332         scsi_ulto2b(0x003A, feature->feature_code);
10333         feature->flags = 0x00;
10334         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10335                 feature->flags |= SGC_F_CURRENT;
10336         feature->add_length = 4;
10337         feature->feature_data[0] = 0x00;
10338         feature->feature_data[1] = 0x00;
10339         feature = (struct scsi_get_config_feature *)
10340             &feature->feature_data[feature->add_length];
10341
10342 f3b:    /* DVD+R Dual Layer */
10343         scsi_ulto2b(0x003B, feature->feature_code);
10344         feature->flags = 0x00;
10345         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10346                 feature->flags |= SGC_F_CURRENT;
10347         feature->add_length = 4;
10348         feature->feature_data[0] = 0x00;
10349         feature = (struct scsi_get_config_feature *)
10350             &feature->feature_data[feature->add_length];
10351
10352 done:
10353         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10354         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10355                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10356                 if (scsi_2btoul(feature->feature_code) == starting)
10357                         feature = (struct scsi_get_config_feature *)
10358                             &feature->feature_data[feature->add_length];
10359                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10360         }
10361         scsi_ulto4b(data_len - 4, hdr->data_length);
10362         ctsio->kern_data_len = min(data_len, alloc_len);
10363         ctsio->kern_total_len = ctsio->kern_data_len;
10364
10365         ctl_set_success(ctsio);
10366         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10367         ctsio->be_move_done = ctl_config_move_done;
10368         ctl_datamove((union ctl_io *)ctsio);
10369         return (CTL_RETVAL_COMPLETE);
10370 }
10371
10372 int
10373 ctl_get_event_status(struct ctl_scsiio *ctsio)
10374 {
10375         struct scsi_get_event_status_header *hdr;
10376         struct scsi_get_event_status *cdb;
10377         uint32_t alloc_len, data_len;
10378         int notif_class;
10379
10380         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10381         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10382                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10383                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10384                 ctl_done((union ctl_io *)ctsio);
10385                 return (CTL_RETVAL_COMPLETE);
10386         }
10387         notif_class = cdb->notif_class;
10388         alloc_len = scsi_2btoul(cdb->length);
10389
10390         data_len = sizeof(struct scsi_get_event_status_header);
10391         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10392         ctsio->kern_sg_entries = 0;
10393         ctsio->kern_rel_offset = 0;
10394         ctsio->kern_data_len = min(data_len, alloc_len);
10395         ctsio->kern_total_len = ctsio->kern_data_len;
10396
10397         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10398         scsi_ulto2b(0, hdr->descr_length);
10399         hdr->nea_class = SGESN_NEA;
10400         hdr->supported_class = 0;
10401
10402         ctl_set_success(ctsio);
10403         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10404         ctsio->be_move_done = ctl_config_move_done;
10405         ctl_datamove((union ctl_io *)ctsio);
10406         return (CTL_RETVAL_COMPLETE);
10407 }
10408
10409 int
10410 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10411 {
10412         struct scsi_mechanism_status_header *hdr;
10413         struct scsi_mechanism_status *cdb;
10414         uint32_t alloc_len, data_len;
10415
10416         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10417         alloc_len = scsi_2btoul(cdb->length);
10418
10419         data_len = sizeof(struct scsi_mechanism_status_header);
10420         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10421         ctsio->kern_sg_entries = 0;
10422         ctsio->kern_rel_offset = 0;
10423         ctsio->kern_data_len = min(data_len, alloc_len);
10424         ctsio->kern_total_len = ctsio->kern_data_len;
10425
10426         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10427         hdr->state1 = 0x00;
10428         hdr->state2 = 0xe0;
10429         scsi_ulto3b(0, hdr->lba);
10430         hdr->slots_num = 0;
10431         scsi_ulto2b(0, hdr->slots_length);
10432
10433         ctl_set_success(ctsio);
10434         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10435         ctsio->be_move_done = ctl_config_move_done;
10436         ctl_datamove((union ctl_io *)ctsio);
10437         return (CTL_RETVAL_COMPLETE);
10438 }
10439
10440 static void
10441 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10442 {
10443
10444         lba += 150;
10445         buf[0] = 0;
10446         buf[1] = bin2bcd((lba / 75) / 60);
10447         buf[2] = bin2bcd((lba / 75) % 60);
10448         buf[3] = bin2bcd(lba % 75);
10449 }
10450
10451 int
10452 ctl_read_toc(struct ctl_scsiio *ctsio)
10453 {
10454         struct ctl_lun *lun = CTL_LUN(ctsio);
10455         struct scsi_read_toc_hdr *hdr;
10456         struct scsi_read_toc_type01_descr *descr;
10457         struct scsi_read_toc *cdb;
10458         uint32_t alloc_len, data_len;
10459         int format, msf;
10460
10461         cdb = (struct scsi_read_toc *)ctsio->cdb;
10462         msf = (cdb->byte2 & CD_MSF) != 0;
10463         format = cdb->format;
10464         alloc_len = scsi_2btoul(cdb->data_len);
10465
10466         data_len = sizeof(struct scsi_read_toc_hdr);
10467         if (format == 0)
10468                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10469         else
10470                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10471         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10472         ctsio->kern_sg_entries = 0;
10473         ctsio->kern_rel_offset = 0;
10474         ctsio->kern_data_len = min(data_len, alloc_len);
10475         ctsio->kern_total_len = ctsio->kern_data_len;
10476
10477         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10478         if (format == 0) {
10479                 scsi_ulto2b(0x12, hdr->data_length);
10480                 hdr->first = 1;
10481                 hdr->last = 1;
10482                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10483                 descr->addr_ctl = 0x14;
10484                 descr->track_number = 1;
10485                 if (msf)
10486                         ctl_ultomsf(0, descr->track_start);
10487                 else
10488                         scsi_ulto4b(0, descr->track_start);
10489                 descr++;
10490                 descr->addr_ctl = 0x14;
10491                 descr->track_number = 0xaa;
10492                 if (msf)
10493                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10494                 else
10495                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10496         } else {
10497                 scsi_ulto2b(0x0a, hdr->data_length);
10498                 hdr->first = 1;
10499                 hdr->last = 1;
10500                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10501                 descr->addr_ctl = 0x14;
10502                 descr->track_number = 1;
10503                 if (msf)
10504                         ctl_ultomsf(0, descr->track_start);
10505                 else
10506                         scsi_ulto4b(0, descr->track_start);
10507         }
10508
10509         ctl_set_success(ctsio);
10510         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10511         ctsio->be_move_done = ctl_config_move_done;
10512         ctl_datamove((union ctl_io *)ctsio);
10513         return (CTL_RETVAL_COMPLETE);
10514 }
10515
10516 /*
10517  * For known CDB types, parse the LBA and length.
10518  */
10519 static int
10520 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10521 {
10522         if (io->io_hdr.io_type != CTL_IO_SCSI)
10523                 return (1);
10524
10525         switch (io->scsiio.cdb[0]) {
10526         case COMPARE_AND_WRITE: {
10527                 struct scsi_compare_and_write *cdb;
10528
10529                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10530
10531                 *lba = scsi_8btou64(cdb->addr);
10532                 *len = cdb->length;
10533                 break;
10534         }
10535         case READ_6:
10536         case WRITE_6: {
10537                 struct scsi_rw_6 *cdb;
10538
10539                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10540
10541                 *lba = scsi_3btoul(cdb->addr);
10542                 /* only 5 bits are valid in the most significant address byte */
10543                 *lba &= 0x1fffff;
10544                 *len = cdb->length;
10545                 break;
10546         }
10547         case READ_10:
10548         case WRITE_10: {
10549                 struct scsi_rw_10 *cdb;
10550
10551                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10552
10553                 *lba = scsi_4btoul(cdb->addr);
10554                 *len = scsi_2btoul(cdb->length);
10555                 break;
10556         }
10557         case WRITE_VERIFY_10: {
10558                 struct scsi_write_verify_10 *cdb;
10559
10560                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10561
10562                 *lba = scsi_4btoul(cdb->addr);
10563                 *len = scsi_2btoul(cdb->length);
10564                 break;
10565         }
10566         case READ_12:
10567         case WRITE_12: {
10568                 struct scsi_rw_12 *cdb;
10569
10570                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10571
10572                 *lba = scsi_4btoul(cdb->addr);
10573                 *len = scsi_4btoul(cdb->length);
10574                 break;
10575         }
10576         case WRITE_VERIFY_12: {
10577                 struct scsi_write_verify_12 *cdb;
10578
10579                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10580
10581                 *lba = scsi_4btoul(cdb->addr);
10582                 *len = scsi_4btoul(cdb->length);
10583                 break;
10584         }
10585         case READ_16:
10586         case WRITE_16: {
10587                 struct scsi_rw_16 *cdb;
10588
10589                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10590
10591                 *lba = scsi_8btou64(cdb->addr);
10592                 *len = scsi_4btoul(cdb->length);
10593                 break;
10594         }
10595         case WRITE_ATOMIC_16: {
10596                 struct scsi_write_atomic_16 *cdb;
10597
10598                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10599
10600                 *lba = scsi_8btou64(cdb->addr);
10601                 *len = scsi_2btoul(cdb->length);
10602                 break;
10603         }
10604         case WRITE_VERIFY_16: {
10605                 struct scsi_write_verify_16 *cdb;
10606
10607                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10608
10609                 *lba = scsi_8btou64(cdb->addr);
10610                 *len = scsi_4btoul(cdb->length);
10611                 break;
10612         }
10613         case WRITE_SAME_10: {
10614                 struct scsi_write_same_10 *cdb;
10615
10616                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10617
10618                 *lba = scsi_4btoul(cdb->addr);
10619                 *len = scsi_2btoul(cdb->length);
10620                 break;
10621         }
10622         case WRITE_SAME_16: {
10623                 struct scsi_write_same_16 *cdb;
10624
10625                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10626
10627                 *lba = scsi_8btou64(cdb->addr);
10628                 *len = scsi_4btoul(cdb->length);
10629                 break;
10630         }
10631         case VERIFY_10: {
10632                 struct scsi_verify_10 *cdb;
10633
10634                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10635
10636                 *lba = scsi_4btoul(cdb->addr);
10637                 *len = scsi_2btoul(cdb->length);
10638                 break;
10639         }
10640         case VERIFY_12: {
10641                 struct scsi_verify_12 *cdb;
10642
10643                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10644
10645                 *lba = scsi_4btoul(cdb->addr);
10646                 *len = scsi_4btoul(cdb->length);
10647                 break;
10648         }
10649         case VERIFY_16: {
10650                 struct scsi_verify_16 *cdb;
10651
10652                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10653
10654                 *lba = scsi_8btou64(cdb->addr);
10655                 *len = scsi_4btoul(cdb->length);
10656                 break;
10657         }
10658         case UNMAP: {
10659                 *lba = 0;
10660                 *len = UINT64_MAX;
10661                 break;
10662         }
10663         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10664                 struct scsi_get_lba_status *cdb;
10665
10666                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10667                 *lba = scsi_8btou64(cdb->addr);
10668                 *len = UINT32_MAX;
10669                 break;
10670         }
10671         default:
10672                 return (1);
10673                 break; /* NOTREACHED */
10674         }
10675
10676         return (0);
10677 }
10678
10679 static ctl_action
10680 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10681     bool seq)
10682 {
10683         uint64_t endlba1, endlba2;
10684
10685         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10686         endlba2 = lba2 + len2 - 1;
10687
10688         if ((endlba1 < lba2) || (endlba2 < lba1))
10689                 return (CTL_ACTION_PASS);
10690         else
10691                 return (CTL_ACTION_BLOCK);
10692 }
10693
10694 static int
10695 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10696 {
10697         struct ctl_ptr_len_flags *ptrlen;
10698         struct scsi_unmap_desc *buf, *end, *range;
10699         uint64_t lba;
10700         uint32_t len;
10701
10702         /* If not UNMAP -- go other way. */
10703         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10704             io->scsiio.cdb[0] != UNMAP)
10705                 return (CTL_ACTION_ERROR);
10706
10707         /* If UNMAP without data -- block and wait for data. */
10708         ptrlen = (struct ctl_ptr_len_flags *)
10709             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10710         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10711             ptrlen->ptr == NULL)
10712                 return (CTL_ACTION_BLOCK);
10713
10714         /* UNMAP with data -- check for collision. */
10715         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10716         end = buf + ptrlen->len / sizeof(*buf);
10717         for (range = buf; range < end; range++) {
10718                 lba = scsi_8btou64(range->lba);
10719                 len = scsi_4btoul(range->length);
10720                 if ((lba < lba2 + len2) && (lba + len > lba2))
10721                         return (CTL_ACTION_BLOCK);
10722         }
10723         return (CTL_ACTION_PASS);
10724 }
10725
10726 static ctl_action
10727 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10728 {
10729         uint64_t lba1, lba2;
10730         uint64_t len1, len2;
10731         int retval;
10732
10733         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10734                 return (CTL_ACTION_ERROR);
10735
10736         retval = ctl_extent_check_unmap(io1, lba2, len2);
10737         if (retval != CTL_ACTION_ERROR)
10738                 return (retval);
10739
10740         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10741                 return (CTL_ACTION_ERROR);
10742
10743         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10744                 seq = FALSE;
10745         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10746 }
10747
10748 static ctl_action
10749 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10750 {
10751         uint64_t lba1, lba2;
10752         uint64_t len1, len2;
10753
10754         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10755                 return (CTL_ACTION_PASS);
10756         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10757                 return (CTL_ACTION_ERROR);
10758         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10759                 return (CTL_ACTION_ERROR);
10760
10761         if (lba1 + len1 == lba2)
10762                 return (CTL_ACTION_BLOCK);
10763         return (CTL_ACTION_PASS);
10764 }
10765
10766 static ctl_action
10767 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10768     union ctl_io *ooa_io)
10769 {
10770         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10771         const ctl_serialize_action *serialize_row;
10772
10773         /*
10774          * The initiator attempted multiple untagged commands at the same
10775          * time.  Can't do that.
10776          */
10777         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10778          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10779          && ((pending_io->io_hdr.nexus.targ_port ==
10780               ooa_io->io_hdr.nexus.targ_port)
10781           && (pending_io->io_hdr.nexus.initid ==
10782               ooa_io->io_hdr.nexus.initid))
10783          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10784               CTL_FLAG_STATUS_SENT)) == 0))
10785                 return (CTL_ACTION_OVERLAP);
10786
10787         /*
10788          * The initiator attempted to send multiple tagged commands with
10789          * the same ID.  (It's fine if different initiators have the same
10790          * tag ID.)
10791          *
10792          * Even if all of those conditions are true, we don't kill the I/O
10793          * if the command ahead of us has been aborted.  We won't end up
10794          * sending it to the FETD, and it's perfectly legal to resend a
10795          * command with the same tag number as long as the previous
10796          * instance of this tag number has been aborted somehow.
10797          */
10798         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10799          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10800          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10801          && ((pending_io->io_hdr.nexus.targ_port ==
10802               ooa_io->io_hdr.nexus.targ_port)
10803           && (pending_io->io_hdr.nexus.initid ==
10804               ooa_io->io_hdr.nexus.initid))
10805          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10806               CTL_FLAG_STATUS_SENT)) == 0))
10807                 return (CTL_ACTION_OVERLAP_TAG);
10808
10809         /*
10810          * If we get a head of queue tag, SAM-3 says that we should
10811          * immediately execute it.
10812          *
10813          * What happens if this command would normally block for some other
10814          * reason?  e.g. a request sense with a head of queue tag
10815          * immediately after a write.  Normally that would block, but this
10816          * will result in its getting executed immediately...
10817          *
10818          * We currently return "pass" instead of "skip", so we'll end up
10819          * going through the rest of the queue to check for overlapped tags.
10820          *
10821          * XXX KDM check for other types of blockage first??
10822          */
10823         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10824                 return (CTL_ACTION_PASS);
10825
10826         /*
10827          * Ordered tags have to block until all items ahead of them
10828          * have completed.  If we get called with an ordered tag, we always
10829          * block, if something else is ahead of us in the queue.
10830          */
10831         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10832                 return (CTL_ACTION_BLOCK);
10833
10834         /*
10835          * Simple tags get blocked until all head of queue and ordered tags
10836          * ahead of them have completed.  I'm lumping untagged commands in
10837          * with simple tags here.  XXX KDM is that the right thing to do?
10838          */
10839         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10840           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10841          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10842           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10843                 return (CTL_ACTION_BLOCK);
10844
10845         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10846         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
10847             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
10848              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
10849              pending_io->scsiio.cdb[1], pending_io));
10850         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10851         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
10852                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
10853         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
10854             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
10855              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
10856              ooa_io->scsiio.cdb[1], ooa_io));
10857
10858         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10859
10860         switch (serialize_row[pending_entry->seridx]) {
10861         case CTL_SER_BLOCK:
10862                 return (CTL_ACTION_BLOCK);
10863         case CTL_SER_EXTENT:
10864                 return (ctl_extent_check(ooa_io, pending_io,
10865                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10866         case CTL_SER_EXTENTOPT:
10867                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10868                     SCP_QUEUE_ALG_UNRESTRICTED)
10869                         return (ctl_extent_check(ooa_io, pending_io,
10870                             (lun->be_lun &&
10871                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10872                 return (CTL_ACTION_PASS);
10873         case CTL_SER_EXTENTSEQ:
10874                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10875                         return (ctl_extent_check_seq(ooa_io, pending_io));
10876                 return (CTL_ACTION_PASS);
10877         case CTL_SER_PASS:
10878                 return (CTL_ACTION_PASS);
10879         case CTL_SER_BLOCKOPT:
10880                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10881                     SCP_QUEUE_ALG_UNRESTRICTED)
10882                         return (CTL_ACTION_BLOCK);
10883                 return (CTL_ACTION_PASS);
10884         case CTL_SER_SKIP:
10885                 return (CTL_ACTION_SKIP);
10886         default:
10887                 panic("%s: Invalid serialization value %d for %d => %d",
10888                     __func__, serialize_row[pending_entry->seridx],
10889                     pending_entry->seridx, ooa_entry->seridx);
10890         }
10891
10892         return (CTL_ACTION_ERROR);
10893 }
10894
10895 /*
10896  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10897  * Assumptions:
10898  * - pending_io is generally either incoming, or on the blocked queue
10899  * - starting I/O is the I/O we want to start the check with.
10900  */
10901 static ctl_action
10902 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10903               union ctl_io *starting_io)
10904 {
10905         union ctl_io *ooa_io;
10906         ctl_action action;
10907
10908         mtx_assert(&lun->lun_lock, MA_OWNED);
10909
10910         /*
10911          * Run back along the OOA queue, starting with the current
10912          * blocked I/O and going through every I/O before it on the
10913          * queue.  If starting_io is NULL, we'll just end up returning
10914          * CTL_ACTION_PASS.
10915          */
10916         for (ooa_io = starting_io; ooa_io != NULL;
10917              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10918              ooa_links)){
10919
10920                 /*
10921                  * This routine just checks to see whether
10922                  * cur_blocked is blocked by ooa_io, which is ahead
10923                  * of it in the queue.  It doesn't queue/dequeue
10924                  * cur_blocked.
10925                  */
10926                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10927                 switch (action) {
10928                 case CTL_ACTION_BLOCK:
10929                 case CTL_ACTION_OVERLAP:
10930                 case CTL_ACTION_OVERLAP_TAG:
10931                 case CTL_ACTION_SKIP:
10932                 case CTL_ACTION_ERROR:
10933                         return (action);
10934                         break; /* NOTREACHED */
10935                 case CTL_ACTION_PASS:
10936                         break;
10937                 default:
10938                         panic("%s: Invalid action %d\n", __func__, action);
10939                 }
10940         }
10941
10942         return (CTL_ACTION_PASS);
10943 }
10944
10945 /*
10946  * Assumptions:
10947  * - An I/O has just completed, and has been removed from the per-LUN OOA
10948  *   queue, so some items on the blocked queue may now be unblocked.
10949  */
10950 static int
10951 ctl_check_blocked(struct ctl_lun *lun)
10952 {
10953         struct ctl_softc *softc = lun->ctl_softc;
10954         union ctl_io *cur_blocked, *next_blocked;
10955
10956         mtx_assert(&lun->lun_lock, MA_OWNED);
10957
10958         /*
10959          * Run forward from the head of the blocked queue, checking each
10960          * entry against the I/Os prior to it on the OOA queue to see if
10961          * there is still any blockage.
10962          *
10963          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10964          * with our removing a variable on it while it is traversing the
10965          * list.
10966          */
10967         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10968              cur_blocked != NULL; cur_blocked = next_blocked) {
10969                 union ctl_io *prev_ooa;
10970                 ctl_action action;
10971
10972                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10973                                                           blocked_links);
10974
10975                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10976                                                       ctl_ooaq, ooa_links);
10977
10978                 /*
10979                  * If cur_blocked happens to be the first item in the OOA
10980                  * queue now, prev_ooa will be NULL, and the action
10981                  * returned will just be CTL_ACTION_PASS.
10982                  */
10983                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10984
10985                 switch (action) {
10986                 case CTL_ACTION_BLOCK:
10987                         /* Nothing to do here, still blocked */
10988                         break;
10989                 case CTL_ACTION_OVERLAP:
10990                 case CTL_ACTION_OVERLAP_TAG:
10991                         /*
10992                          * This shouldn't happen!  In theory we've already
10993                          * checked this command for overlap...
10994                          */
10995                         break;
10996                 case CTL_ACTION_PASS:
10997                 case CTL_ACTION_SKIP: {
10998                         const struct ctl_cmd_entry *entry;
10999
11000                         /*
11001                          * The skip case shouldn't happen, this transaction
11002                          * should have never made it onto the blocked queue.
11003                          */
11004                         /*
11005                          * This I/O is no longer blocked, we can remove it
11006                          * from the blocked queue.  Since this is a TAILQ
11007                          * (doubly linked list), we can do O(1) removals
11008                          * from any place on the list.
11009                          */
11010                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11011                                      blocked_links);
11012                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11013
11014                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11015                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11016                                 /*
11017                                  * Need to send IO back to original side to
11018                                  * run
11019                                  */
11020                                 union ctl_ha_msg msg_info;
11021
11022                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11023                                 msg_info.hdr.original_sc =
11024                                         cur_blocked->io_hdr.original_sc;
11025                                 msg_info.hdr.serializing_sc = cur_blocked;
11026                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11027                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11028                                     sizeof(msg_info.hdr), M_NOWAIT);
11029                                 break;
11030                         }
11031                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11032
11033                         /*
11034                          * Check this I/O for LUN state changes that may
11035                          * have happened while this command was blocked.
11036                          * The LUN state may have been changed by a command
11037                          * ahead of us in the queue, so we need to re-check
11038                          * for any states that can be caused by SCSI
11039                          * commands.
11040                          */
11041                         if (ctl_scsiio_lun_check(lun, entry,
11042                                                  &cur_blocked->scsiio) == 0) {
11043                                 cur_blocked->io_hdr.flags |=
11044                                                       CTL_FLAG_IS_WAS_ON_RTR;
11045                                 ctl_enqueue_rtr(cur_blocked);
11046                         } else
11047                                 ctl_done(cur_blocked);
11048                         break;
11049                 }
11050                 default:
11051                         /*
11052                          * This probably shouldn't happen -- we shouldn't
11053                          * get CTL_ACTION_ERROR, or anything else.
11054                          */
11055                         break;
11056                 }
11057         }
11058
11059         return (CTL_RETVAL_COMPLETE);
11060 }
11061
11062 /*
11063  * This routine (with one exception) checks LUN flags that can be set by
11064  * commands ahead of us in the OOA queue.  These flags have to be checked
11065  * when a command initially comes in, and when we pull a command off the
11066  * blocked queue and are preparing to execute it.  The reason we have to
11067  * check these flags for commands on the blocked queue is that the LUN
11068  * state may have been changed by a command ahead of us while we're on the
11069  * blocked queue.
11070  *
11071  * Ordering is somewhat important with these checks, so please pay
11072  * careful attention to the placement of any new checks.
11073  */
11074 static int
11075 ctl_scsiio_lun_check(struct ctl_lun *lun,
11076     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11077 {
11078         struct ctl_softc *softc = lun->ctl_softc;
11079         int retval;
11080         uint32_t residx;
11081
11082         retval = 0;
11083
11084         mtx_assert(&lun->lun_lock, MA_OWNED);
11085
11086         /*
11087          * If this shelf is a secondary shelf controller, we may have to
11088          * reject some commands disallowed by HA mode and link state.
11089          */
11090         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11091                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11092                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11093                         ctl_set_lun_unavail(ctsio);
11094                         retval = 1;
11095                         goto bailout;
11096                 }
11097                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11098                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11099                         ctl_set_lun_transit(ctsio);
11100                         retval = 1;
11101                         goto bailout;
11102                 }
11103                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11104                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11105                         ctl_set_lun_standby(ctsio);
11106                         retval = 1;
11107                         goto bailout;
11108                 }
11109
11110                 /* The rest of checks are only done on executing side */
11111                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11112                         goto bailout;
11113         }
11114
11115         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11116                 if (lun->be_lun &&
11117                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11118                         ctl_set_hw_write_protected(ctsio);
11119                         retval = 1;
11120                         goto bailout;
11121                 }
11122                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11123                         ctl_set_sense(ctsio, /*current_error*/ 1,
11124                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11125                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11126                         retval = 1;
11127                         goto bailout;
11128                 }
11129         }
11130
11131         /*
11132          * Check for a reservation conflict.  If this command isn't allowed
11133          * even on reserved LUNs, and if this initiator isn't the one who
11134          * reserved us, reject the command with a reservation conflict.
11135          */
11136         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11137         if ((lun->flags & CTL_LUN_RESERVED)
11138          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11139                 if (lun->res_idx != residx) {
11140                         ctl_set_reservation_conflict(ctsio);
11141                         retval = 1;
11142                         goto bailout;
11143                 }
11144         }
11145
11146         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11147             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11148                 /* No reservation or command is allowed. */;
11149         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11150             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11151              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11152              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11153                 /* The command is allowed for Write Exclusive resv. */;
11154         } else {
11155                 /*
11156                  * if we aren't registered or it's a res holder type
11157                  * reservation and this isn't the res holder then set a
11158                  * conflict.
11159                  */
11160                 if (ctl_get_prkey(lun, residx) == 0 ||
11161                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11162                         ctl_set_reservation_conflict(ctsio);
11163                         retval = 1;
11164                         goto bailout;
11165                 }
11166         }
11167
11168         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11169                 if (lun->flags & CTL_LUN_EJECTED)
11170                         ctl_set_lun_ejected(ctsio);
11171                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11172                         if (lun->flags & CTL_LUN_REMOVABLE)
11173                                 ctl_set_lun_no_media(ctsio);
11174                         else
11175                                 ctl_set_lun_int_reqd(ctsio);
11176                 } else if (lun->flags & CTL_LUN_STOPPED)
11177                         ctl_set_lun_stopped(ctsio);
11178                 else
11179                         goto bailout;
11180                 retval = 1;
11181                 goto bailout;
11182         }
11183
11184 bailout:
11185         return (retval);
11186 }
11187
11188 static void
11189 ctl_failover_io(union ctl_io *io, int have_lock)
11190 {
11191         ctl_set_busy(&io->scsiio);
11192         ctl_done(io);
11193 }
11194
11195 static void
11196 ctl_failover_lun(union ctl_io *rio)
11197 {
11198         struct ctl_softc *softc = CTL_SOFTC(rio);
11199         struct ctl_lun *lun;
11200         struct ctl_io_hdr *io, *next_io;
11201         uint32_t targ_lun;
11202
11203         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11204         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11205
11206         /* Find and lock the LUN. */
11207         mtx_lock(&softc->ctl_lock);
11208         if (targ_lun > CTL_MAX_LUNS ||
11209             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11210                 mtx_unlock(&softc->ctl_lock);
11211                 return;
11212         }
11213         mtx_lock(&lun->lun_lock);
11214         mtx_unlock(&softc->ctl_lock);
11215         if (lun->flags & CTL_LUN_DISABLED) {
11216                 mtx_unlock(&lun->lun_lock);
11217                 return;
11218         }
11219
11220         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11221                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11222                         /* We are master */
11223                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11224                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11225                                         io->flags |= CTL_FLAG_ABORT;
11226                                         io->flags |= CTL_FLAG_FAILOVER;
11227                                 } else { /* This can be only due to DATAMOVE */
11228                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11229                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11230                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11231                                         io->port_status = 31340;
11232                                         ctl_enqueue_isc((union ctl_io *)io);
11233                                 }
11234                         }
11235                         /* We are slave */
11236                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11237                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11238                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11239                                         io->flags |= CTL_FLAG_FAILOVER;
11240                                 } else {
11241                                         ctl_set_busy(&((union ctl_io *)io)->
11242                                             scsiio);
11243                                         ctl_done((union ctl_io *)io);
11244                                 }
11245                         }
11246                 }
11247         } else { /* SERIALIZE modes */
11248                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11249                     next_io) {
11250                         /* We are master */
11251                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11252                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11253                                     blocked_links);
11254                                 io->flags &= ~CTL_FLAG_BLOCKED;
11255                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11256                                 ctl_free_io((union ctl_io *)io);
11257                         }
11258                 }
11259                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11260                         /* We are master */
11261                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11262                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11263                                 ctl_free_io((union ctl_io *)io);
11264                         }
11265                         /* We are slave */
11266                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11267                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11268                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11269                                         ctl_set_busy(&((union ctl_io *)io)->
11270                                             scsiio);
11271                                         ctl_done((union ctl_io *)io);
11272                                 }
11273                         }
11274                 }
11275                 ctl_check_blocked(lun);
11276         }
11277         mtx_unlock(&lun->lun_lock);
11278 }
11279
11280 static int
11281 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11282 {
11283         struct ctl_lun *lun;
11284         const struct ctl_cmd_entry *entry;
11285         uint32_t initidx, targ_lun;
11286         int retval = 0;
11287
11288         lun = NULL;
11289         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11290         if (targ_lun < CTL_MAX_LUNS)
11291                 lun = softc->ctl_luns[targ_lun];
11292         if (lun) {
11293                 /*
11294                  * If the LUN is invalid, pretend that it doesn't exist.
11295                  * It will go away as soon as all pending I/O has been
11296                  * completed.
11297                  */
11298                 mtx_lock(&lun->lun_lock);
11299                 if (lun->flags & CTL_LUN_DISABLED) {
11300                         mtx_unlock(&lun->lun_lock);
11301                         lun = NULL;
11302                 }
11303         }
11304         CTL_LUN(ctsio) = lun;
11305         if (lun) {
11306                 CTL_BACKEND_LUN(ctsio) = lun->be_lun;
11307
11308                 /*
11309                  * Every I/O goes into the OOA queue for a particular LUN,
11310                  * and stays there until completion.
11311                  */
11312 #ifdef CTL_TIME_IO
11313                 if (TAILQ_EMPTY(&lun->ooa_queue))
11314                         lun->idle_time += getsbinuptime() - lun->last_busy;
11315 #endif
11316                 TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11317         }
11318
11319         /* Get command entry and return error if it is unsuppotyed. */
11320         entry = ctl_validate_command(ctsio);
11321         if (entry == NULL) {
11322                 if (lun)
11323                         mtx_unlock(&lun->lun_lock);
11324                 return (retval);
11325         }
11326
11327         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11328         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11329
11330         /*
11331          * Check to see whether we can send this command to LUNs that don't
11332          * exist.  This should pretty much only be the case for inquiry
11333          * and request sense.  Further checks, below, really require having
11334          * a LUN, so we can't really check the command anymore.  Just put
11335          * it on the rtr queue.
11336          */
11337         if (lun == NULL) {
11338                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11339                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11340                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11341                         return (retval);
11342                 }
11343
11344                 ctl_set_unsupported_lun(ctsio);
11345                 ctl_done((union ctl_io *)ctsio);
11346                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11347                 return (retval);
11348         } else {
11349                 /*
11350                  * Make sure we support this particular command on this LUN.
11351                  * e.g., we don't support writes to the control LUN.
11352                  */
11353                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11354                         mtx_unlock(&lun->lun_lock);
11355                         ctl_set_invalid_opcode(ctsio);
11356                         ctl_done((union ctl_io *)ctsio);
11357                         return (retval);
11358                 }
11359         }
11360
11361         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11362
11363 #ifdef CTL_WITH_CA
11364         /*
11365          * If we've got a request sense, it'll clear the contingent
11366          * allegiance condition.  Otherwise, if we have a CA condition for
11367          * this initiator, clear it, because it sent down a command other
11368          * than request sense.
11369          */
11370         if ((ctsio->cdb[0] != REQUEST_SENSE)
11371          && (ctl_is_set(lun->have_ca, initidx)))
11372                 ctl_clear_mask(lun->have_ca, initidx);
11373 #endif
11374
11375         /*
11376          * If the command has this flag set, it handles its own unit
11377          * attention reporting, we shouldn't do anything.  Otherwise we
11378          * check for any pending unit attentions, and send them back to the
11379          * initiator.  We only do this when a command initially comes in,
11380          * not when we pull it off the blocked queue.
11381          *
11382          * According to SAM-3, section 5.3.2, the order that things get
11383          * presented back to the host is basically unit attentions caused
11384          * by some sort of reset event, busy status, reservation conflicts
11385          * or task set full, and finally any other status.
11386          *
11387          * One issue here is that some of the unit attentions we report
11388          * don't fall into the "reset" category (e.g. "reported luns data
11389          * has changed").  So reporting it here, before the reservation
11390          * check, may be technically wrong.  I guess the only thing to do
11391          * would be to check for and report the reset events here, and then
11392          * check for the other unit attention types after we check for a
11393          * reservation conflict.
11394          *
11395          * XXX KDM need to fix this
11396          */
11397         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11398                 ctl_ua_type ua_type;
11399                 u_int sense_len = 0;
11400
11401                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11402                     &sense_len, SSD_TYPE_NONE);
11403                 if (ua_type != CTL_UA_NONE) {
11404                         mtx_unlock(&lun->lun_lock);
11405                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11406                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11407                         ctsio->sense_len = sense_len;
11408                         ctl_done((union ctl_io *)ctsio);
11409                         return (retval);
11410                 }
11411         }
11412
11413
11414         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11415                 mtx_unlock(&lun->lun_lock);
11416                 ctl_done((union ctl_io *)ctsio);
11417                 return (retval);
11418         }
11419
11420         /*
11421          * XXX CHD this is where we want to send IO to other side if
11422          * this LUN is secondary on this SC. We will need to make a copy
11423          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11424          * the copy we send as FROM_OTHER.
11425          * We also need to stuff the address of the original IO so we can
11426          * find it easily. Something similar will need be done on the other
11427          * side so when we are done we can find the copy.
11428          */
11429         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11430             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11431             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11432                 union ctl_ha_msg msg_info;
11433                 int isc_retval;
11434
11435                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11436                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11437                 mtx_unlock(&lun->lun_lock);
11438
11439                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11440                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11441                 msg_info.hdr.serializing_sc = NULL;
11442                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11443                 msg_info.scsi.tag_num = ctsio->tag_num;
11444                 msg_info.scsi.tag_type = ctsio->tag_type;
11445                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11446                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11447
11448                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11449                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11450                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11451                         ctl_set_busy(ctsio);
11452                         ctl_done((union ctl_io *)ctsio);
11453                         return (retval);
11454                 }
11455                 return (retval);
11456         }
11457
11458         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11459                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11460                               ctl_ooaq, ooa_links))) {
11461         case CTL_ACTION_BLOCK:
11462                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11463                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11464                                   blocked_links);
11465                 mtx_unlock(&lun->lun_lock);
11466                 return (retval);
11467         case CTL_ACTION_PASS:
11468         case CTL_ACTION_SKIP:
11469                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11470                 mtx_unlock(&lun->lun_lock);
11471                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11472                 break;
11473         case CTL_ACTION_OVERLAP:
11474                 mtx_unlock(&lun->lun_lock);
11475                 ctl_set_overlapped_cmd(ctsio);
11476                 ctl_done((union ctl_io *)ctsio);
11477                 break;
11478         case CTL_ACTION_OVERLAP_TAG:
11479                 mtx_unlock(&lun->lun_lock);
11480                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11481                 ctl_done((union ctl_io *)ctsio);
11482                 break;
11483         case CTL_ACTION_ERROR:
11484         default:
11485                 mtx_unlock(&lun->lun_lock);
11486                 ctl_set_internal_failure(ctsio,
11487                                          /*sks_valid*/ 0,
11488                                          /*retry_count*/ 0);
11489                 ctl_done((union ctl_io *)ctsio);
11490                 break;
11491         }
11492         return (retval);
11493 }
11494
11495 const struct ctl_cmd_entry *
11496 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11497 {
11498         const struct ctl_cmd_entry *entry;
11499         int service_action;
11500
11501         entry = &ctl_cmd_table[ctsio->cdb[0]];
11502         if (sa)
11503                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11504         if (entry->flags & CTL_CMD_FLAG_SA5) {
11505                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11506                 entry = &((const struct ctl_cmd_entry *)
11507                     entry->execute)[service_action];
11508         }
11509         return (entry);
11510 }
11511
11512 const struct ctl_cmd_entry *
11513 ctl_validate_command(struct ctl_scsiio *ctsio)
11514 {
11515         const struct ctl_cmd_entry *entry;
11516         int i, sa;
11517         uint8_t diff;
11518
11519         entry = ctl_get_cmd_entry(ctsio, &sa);
11520         if (entry->execute == NULL) {
11521                 if (sa)
11522                         ctl_set_invalid_field(ctsio,
11523                                               /*sks_valid*/ 1,
11524                                               /*command*/ 1,
11525                                               /*field*/ 1,
11526                                               /*bit_valid*/ 1,
11527                                               /*bit*/ 4);
11528                 else
11529                         ctl_set_invalid_opcode(ctsio);
11530                 ctl_done((union ctl_io *)ctsio);
11531                 return (NULL);
11532         }
11533         KASSERT(entry->length > 0,
11534             ("Not defined length for command 0x%02x/0x%02x",
11535              ctsio->cdb[0], ctsio->cdb[1]));
11536         for (i = 1; i < entry->length; i++) {
11537                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11538                 if (diff == 0)
11539                         continue;
11540                 ctl_set_invalid_field(ctsio,
11541                                       /*sks_valid*/ 1,
11542                                       /*command*/ 1,
11543                                       /*field*/ i,
11544                                       /*bit_valid*/ 1,
11545                                       /*bit*/ fls(diff) - 1);
11546                 ctl_done((union ctl_io *)ctsio);
11547                 return (NULL);
11548         }
11549         return (entry);
11550 }
11551
11552 static int
11553 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11554 {
11555
11556         switch (lun_type) {
11557         case T_DIRECT:
11558                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11559                         return (0);
11560                 break;
11561         case T_PROCESSOR:
11562                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11563                         return (0);
11564                 break;
11565         case T_CDROM:
11566                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11567                         return (0);
11568                 break;
11569         default:
11570                 return (0);
11571         }
11572         return (1);
11573 }
11574
11575 static int
11576 ctl_scsiio(struct ctl_scsiio *ctsio)
11577 {
11578         int retval;
11579         const struct ctl_cmd_entry *entry;
11580
11581         retval = CTL_RETVAL_COMPLETE;
11582
11583         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11584
11585         entry = ctl_get_cmd_entry(ctsio, NULL);
11586
11587         /*
11588          * If this I/O has been aborted, just send it straight to
11589          * ctl_done() without executing it.
11590          */
11591         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11592                 ctl_done((union ctl_io *)ctsio);
11593                 goto bailout;
11594         }
11595
11596         /*
11597          * All the checks should have been handled by ctl_scsiio_precheck().
11598          * We should be clear now to just execute the I/O.
11599          */
11600         retval = entry->execute(ctsio);
11601
11602 bailout:
11603         return (retval);
11604 }
11605
11606 /*
11607  * Since we only implement one target right now, a bus reset simply resets
11608  * our single target.
11609  */
11610 static int
11611 ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11612 {
11613         return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11614 }
11615
11616 static int
11617 ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11618                  ctl_ua_type ua_type)
11619 {
11620         struct ctl_port *port = CTL_PORT(io);
11621         struct ctl_lun *lun;
11622         int retval;
11623
11624         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11625                 union ctl_ha_msg msg_info;
11626
11627                 msg_info.hdr.nexus = io->io_hdr.nexus;
11628                 if (ua_type==CTL_UA_TARG_RESET)
11629                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11630                 else
11631                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11632                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11633                 msg_info.hdr.original_sc = NULL;
11634                 msg_info.hdr.serializing_sc = NULL;
11635                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11636                     sizeof(msg_info.task), M_WAITOK);
11637         }
11638         retval = 0;
11639
11640         mtx_lock(&softc->ctl_lock);
11641         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11642                 if (port != NULL &&
11643                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
11644                         continue;
11645                 retval += ctl_do_lun_reset(lun, io, ua_type);
11646         }
11647         mtx_unlock(&softc->ctl_lock);
11648         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11649         return (retval);
11650 }
11651
11652 /*
11653  * The LUN should always be set.  The I/O is optional, and is used to
11654  * distinguish between I/Os sent by this initiator, and by other
11655  * initiators.  We set unit attention for initiators other than this one.
11656  * SAM-3 is vague on this point.  It does say that a unit attention should
11657  * be established for other initiators when a LUN is reset (see section
11658  * 5.7.3), but it doesn't specifically say that the unit attention should
11659  * be established for this particular initiator when a LUN is reset.  Here
11660  * is the relevant text, from SAM-3 rev 8:
11661  *
11662  * 5.7.2 When a SCSI initiator port aborts its own tasks
11663  *
11664  * When a SCSI initiator port causes its own task(s) to be aborted, no
11665  * notification that the task(s) have been aborted shall be returned to
11666  * the SCSI initiator port other than the completion response for the
11667  * command or task management function action that caused the task(s) to
11668  * be aborted and notification(s) associated with related effects of the
11669  * action (e.g., a reset unit attention condition).
11670  *
11671  * XXX KDM for now, we're setting unit attention for all initiators.
11672  */
11673 static int
11674 ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11675 {
11676         union ctl_io *xio;
11677 #if 0
11678         uint32_t initidx;
11679 #endif
11680         int i;
11681
11682         mtx_lock(&lun->lun_lock);
11683         /*
11684          * Run through the OOA queue and abort each I/O.
11685          */
11686         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11687              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11688                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11689         }
11690
11691         /*
11692          * This version sets unit attention for every
11693          */
11694 #if 0
11695         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11696         ctl_est_ua_all(lun, initidx, ua_type);
11697 #else
11698         ctl_est_ua_all(lun, -1, ua_type);
11699 #endif
11700
11701         /*
11702          * A reset (any kind, really) clears reservations established with
11703          * RESERVE/RELEASE.  It does not clear reservations established
11704          * with PERSISTENT RESERVE OUT, but we don't support that at the
11705          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11706          * reservations made with the RESERVE/RELEASE commands, because
11707          * those commands are obsolete in SPC-3.
11708          */
11709         lun->flags &= ~CTL_LUN_RESERVED;
11710
11711 #ifdef CTL_WITH_CA
11712         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11713                 ctl_clear_mask(lun->have_ca, i);
11714 #endif
11715         lun->prevent_count = 0;
11716         if (lun->prevent) {
11717                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
11718                         ctl_clear_mask(lun->prevent, i);
11719         }
11720         mtx_unlock(&lun->lun_lock);
11721
11722         return (0);
11723 }
11724
11725 static int
11726 ctl_lun_reset(struct ctl_softc *softc, union ctl_io *io)
11727 {
11728         struct ctl_lun *lun;
11729         uint32_t targ_lun;
11730         int retval;
11731
11732         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11733         mtx_lock(&softc->ctl_lock);
11734         if (targ_lun >= CTL_MAX_LUNS ||
11735             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11736                 mtx_unlock(&softc->ctl_lock);
11737                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11738                 return (1);
11739         }
11740         retval = ctl_do_lun_reset(lun, io, CTL_UA_LUN_RESET);
11741         mtx_unlock(&softc->ctl_lock);
11742         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11743
11744         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11745                 union ctl_ha_msg msg_info;
11746
11747                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11748                 msg_info.hdr.nexus = io->io_hdr.nexus;
11749                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11750                 msg_info.hdr.original_sc = NULL;
11751                 msg_info.hdr.serializing_sc = NULL;
11752                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11753                     sizeof(msg_info.task), M_WAITOK);
11754         }
11755         return (retval);
11756 }
11757
11758 static void
11759 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11760     int other_sc)
11761 {
11762         union ctl_io *xio;
11763
11764         mtx_assert(&lun->lun_lock, MA_OWNED);
11765
11766         /*
11767          * Run through the OOA queue and attempt to find the given I/O.
11768          * The target port, initiator ID, tag type and tag number have to
11769          * match the values that we got from the initiator.  If we have an
11770          * untagged command to abort, simply abort the first untagged command
11771          * we come to.  We only allow one untagged command at a time of course.
11772          */
11773         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11774              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11775
11776                 if ((targ_port == UINT32_MAX ||
11777                      targ_port == xio->io_hdr.nexus.targ_port) &&
11778                     (init_id == UINT32_MAX ||
11779                      init_id == xio->io_hdr.nexus.initid)) {
11780                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11781                             init_id != xio->io_hdr.nexus.initid)
11782                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11783                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11784                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11785                                 union ctl_ha_msg msg_info;
11786
11787                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11788                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11789                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11790                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11791                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11792                                 msg_info.hdr.original_sc = NULL;
11793                                 msg_info.hdr.serializing_sc = NULL;
11794                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11795                                     sizeof(msg_info.task), M_NOWAIT);
11796                         }
11797                 }
11798         }
11799 }
11800
11801 static int
11802 ctl_abort_task_set(union ctl_io *io)
11803 {
11804         struct ctl_softc *softc = CTL_SOFTC(io);
11805         struct ctl_lun *lun;
11806         uint32_t targ_lun;
11807
11808         /*
11809          * Look up the LUN.
11810          */
11811         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11812         mtx_lock(&softc->ctl_lock);
11813         if (targ_lun >= CTL_MAX_LUNS ||
11814             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11815                 mtx_unlock(&softc->ctl_lock);
11816                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11817                 return (1);
11818         }
11819
11820         mtx_lock(&lun->lun_lock);
11821         mtx_unlock(&softc->ctl_lock);
11822         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11823                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11824                     io->io_hdr.nexus.initid,
11825                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11826         } else { /* CTL_TASK_CLEAR_TASK_SET */
11827                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11828                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11829         }
11830         mtx_unlock(&lun->lun_lock);
11831         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11832         return (0);
11833 }
11834
11835 static int
11836 ctl_i_t_nexus_reset(union ctl_io *io)
11837 {
11838         struct ctl_softc *softc = CTL_SOFTC(io);
11839         struct ctl_lun *lun;
11840         uint32_t initidx;
11841
11842         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11843                 union ctl_ha_msg msg_info;
11844
11845                 msg_info.hdr.nexus = io->io_hdr.nexus;
11846                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
11847                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11848                 msg_info.hdr.original_sc = NULL;
11849                 msg_info.hdr.serializing_sc = NULL;
11850                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11851                     sizeof(msg_info.task), M_WAITOK);
11852         }
11853
11854         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11855         mtx_lock(&softc->ctl_lock);
11856         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11857                 mtx_lock(&lun->lun_lock);
11858                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11859                     io->io_hdr.nexus.initid, 1);
11860 #ifdef CTL_WITH_CA
11861                 ctl_clear_mask(lun->have_ca, initidx);
11862 #endif
11863                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11864                         lun->flags &= ~CTL_LUN_RESERVED;
11865                 if (lun->prevent && ctl_is_set(lun->prevent, initidx)) {
11866                         ctl_clear_mask(lun->prevent, initidx);
11867                         lun->prevent_count--;
11868                 }
11869                 ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
11870                 mtx_unlock(&lun->lun_lock);
11871         }
11872         mtx_unlock(&softc->ctl_lock);
11873         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11874         return (0);
11875 }
11876
11877 static int
11878 ctl_abort_task(union ctl_io *io)
11879 {
11880         struct ctl_softc *softc = CTL_SOFTC(io);
11881         union ctl_io *xio;
11882         struct ctl_lun *lun;
11883 #if 0
11884         struct sbuf sb;
11885         char printbuf[128];
11886 #endif
11887         int found;
11888         uint32_t targ_lun;
11889
11890         found = 0;
11891
11892         /*
11893          * Look up the LUN.
11894          */
11895         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11896         mtx_lock(&softc->ctl_lock);
11897         if (targ_lun >= CTL_MAX_LUNS ||
11898             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11899                 mtx_unlock(&softc->ctl_lock);
11900                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11901                 return (1);
11902         }
11903
11904 #if 0
11905         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11906                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11907 #endif
11908
11909         mtx_lock(&lun->lun_lock);
11910         mtx_unlock(&softc->ctl_lock);
11911         /*
11912          * Run through the OOA queue and attempt to find the given I/O.
11913          * The target port, initiator ID, tag type and tag number have to
11914          * match the values that we got from the initiator.  If we have an
11915          * untagged command to abort, simply abort the first untagged command
11916          * we come to.  We only allow one untagged command at a time of course.
11917          */
11918         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11919              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11920 #if 0
11921                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11922
11923                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11924                             lun->lun, xio->scsiio.tag_num,
11925                             xio->scsiio.tag_type,
11926                             (xio->io_hdr.blocked_links.tqe_prev
11927                             == NULL) ? "" : " BLOCKED",
11928                             (xio->io_hdr.flags &
11929                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11930                             (xio->io_hdr.flags &
11931                             CTL_FLAG_ABORT) ? " ABORT" : "",
11932                             (xio->io_hdr.flags &
11933                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11934                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11935                 sbuf_finish(&sb);
11936                 printf("%s\n", sbuf_data(&sb));
11937 #endif
11938
11939                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11940                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11941                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11942                         continue;
11943
11944                 /*
11945                  * If the abort says that the task is untagged, the
11946                  * task in the queue must be untagged.  Otherwise,
11947                  * we just check to see whether the tag numbers
11948                  * match.  This is because the QLogic firmware
11949                  * doesn't pass back the tag type in an abort
11950                  * request.
11951                  */
11952 #if 0
11953                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11954                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11955                  || (xio->scsiio.tag_num == io->taskio.tag_num))
11956 #endif
11957                 /*
11958                  * XXX KDM we've got problems with FC, because it
11959                  * doesn't send down a tag type with aborts.  So we
11960                  * can only really go by the tag number...
11961                  * This may cause problems with parallel SCSI.
11962                  * Need to figure that out!!
11963                  */
11964                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
11965                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11966                         found = 1;
11967                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
11968                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11969                                 union ctl_ha_msg msg_info;
11970
11971                                 msg_info.hdr.nexus = io->io_hdr.nexus;
11972                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11973                                 msg_info.task.tag_num = io->taskio.tag_num;
11974                                 msg_info.task.tag_type = io->taskio.tag_type;
11975                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11976                                 msg_info.hdr.original_sc = NULL;
11977                                 msg_info.hdr.serializing_sc = NULL;
11978 #if 0
11979                                 printf("Sent Abort to other side\n");
11980 #endif
11981                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11982                                     sizeof(msg_info.task), M_NOWAIT);
11983                         }
11984 #if 0
11985                         printf("ctl_abort_task: found I/O to abort\n");
11986 #endif
11987                 }
11988         }
11989         mtx_unlock(&lun->lun_lock);
11990
11991         if (found == 0) {
11992                 /*
11993                  * This isn't really an error.  It's entirely possible for
11994                  * the abort and command completion to cross on the wire.
11995                  * This is more of an informative/diagnostic error.
11996                  */
11997 #if 0
11998                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11999                        "%u:%u:%u tag %d type %d\n",
12000                        io->io_hdr.nexus.initid,
12001                        io->io_hdr.nexus.targ_port,
12002                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12003                        io->taskio.tag_type);
12004 #endif
12005         }
12006         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12007         return (0);
12008 }
12009
12010 static int
12011 ctl_query_task(union ctl_io *io, int task_set)
12012 {
12013         struct ctl_softc *softc = CTL_SOFTC(io);
12014         union ctl_io *xio;
12015         struct ctl_lun *lun;
12016         int found = 0;
12017         uint32_t targ_lun;
12018
12019         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12020         mtx_lock(&softc->ctl_lock);
12021         if (targ_lun >= CTL_MAX_LUNS ||
12022             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12023                 mtx_unlock(&softc->ctl_lock);
12024                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12025                 return (1);
12026         }
12027         mtx_lock(&lun->lun_lock);
12028         mtx_unlock(&softc->ctl_lock);
12029         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12030              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12031
12032                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12033                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12034                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12035                         continue;
12036
12037                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12038                         found = 1;
12039                         break;
12040                 }
12041         }
12042         mtx_unlock(&lun->lun_lock);
12043         if (found)
12044                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12045         else
12046                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12047         return (0);
12048 }
12049
12050 static int
12051 ctl_query_async_event(union ctl_io *io)
12052 {
12053         struct ctl_softc *softc = CTL_SOFTC(io);
12054         struct ctl_lun *lun;
12055         ctl_ua_type ua;
12056         uint32_t targ_lun, initidx;
12057
12058         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12059         mtx_lock(&softc->ctl_lock);
12060         if (targ_lun >= CTL_MAX_LUNS ||
12061             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12062                 mtx_unlock(&softc->ctl_lock);
12063                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12064                 return (1);
12065         }
12066         mtx_lock(&lun->lun_lock);
12067         mtx_unlock(&softc->ctl_lock);
12068         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12069         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12070         mtx_unlock(&lun->lun_lock);
12071         if (ua != CTL_UA_NONE)
12072                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12073         else
12074                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12075         return (0);
12076 }
12077
12078 static void
12079 ctl_run_task(union ctl_io *io)
12080 {
12081         struct ctl_softc *softc = CTL_SOFTC(io);
12082         int retval = 1;
12083
12084         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12085         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12086             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12087         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12088         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12089         switch (io->taskio.task_action) {
12090         case CTL_TASK_ABORT_TASK:
12091                 retval = ctl_abort_task(io);
12092                 break;
12093         case CTL_TASK_ABORT_TASK_SET:
12094         case CTL_TASK_CLEAR_TASK_SET:
12095                 retval = ctl_abort_task_set(io);
12096                 break;
12097         case CTL_TASK_CLEAR_ACA:
12098                 break;
12099         case CTL_TASK_I_T_NEXUS_RESET:
12100                 retval = ctl_i_t_nexus_reset(io);
12101                 break;
12102         case CTL_TASK_LUN_RESET:
12103                 retval = ctl_lun_reset(softc, io);
12104                 break;
12105         case CTL_TASK_TARGET_RESET:
12106                 retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
12107                 break;
12108         case CTL_TASK_BUS_RESET:
12109                 retval = ctl_bus_reset(softc, io);
12110                 break;
12111         case CTL_TASK_PORT_LOGIN:
12112                 break;
12113         case CTL_TASK_PORT_LOGOUT:
12114                 break;
12115         case CTL_TASK_QUERY_TASK:
12116                 retval = ctl_query_task(io, 0);
12117                 break;
12118         case CTL_TASK_QUERY_TASK_SET:
12119                 retval = ctl_query_task(io, 1);
12120                 break;
12121         case CTL_TASK_QUERY_ASYNC_EVENT:
12122                 retval = ctl_query_async_event(io);
12123                 break;
12124         default:
12125                 printf("%s: got unknown task management event %d\n",
12126                        __func__, io->taskio.task_action);
12127                 break;
12128         }
12129         if (retval == 0)
12130                 io->io_hdr.status = CTL_SUCCESS;
12131         else
12132                 io->io_hdr.status = CTL_ERROR;
12133         ctl_done(io);
12134 }
12135
12136 /*
12137  * For HA operation.  Handle commands that come in from the other
12138  * controller.
12139  */
12140 static void
12141 ctl_handle_isc(union ctl_io *io)
12142 {
12143         struct ctl_softc *softc = CTL_SOFTC(io);
12144         struct ctl_lun *lun;
12145         const struct ctl_cmd_entry *entry;
12146         uint32_t targ_lun;
12147
12148         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12149         switch (io->io_hdr.msg_type) {
12150         case CTL_MSG_SERIALIZE:
12151                 ctl_serialize_other_sc_cmd(&io->scsiio);
12152                 break;
12153         case CTL_MSG_R2R:               /* Only used in SER_ONLY mode. */
12154                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12155                 if (targ_lun >= CTL_MAX_LUNS ||
12156                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12157                         ctl_done(io);
12158                         break;
12159                 }
12160                 mtx_lock(&lun->lun_lock);
12161                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
12162                         mtx_unlock(&lun->lun_lock);
12163                         ctl_done(io);
12164                         break;
12165                 }
12166                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12167                 mtx_unlock(&lun->lun_lock);
12168                 ctl_enqueue_rtr(io);
12169                 break;
12170         case CTL_MSG_FINISH_IO:
12171                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12172                         ctl_done(io);
12173                         break;
12174                 }
12175                 if (targ_lun >= CTL_MAX_LUNS ||
12176                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12177                         ctl_free_io(io);
12178                         break;
12179                 }
12180                 mtx_lock(&lun->lun_lock);
12181                 TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12182                 ctl_check_blocked(lun);
12183                 mtx_unlock(&lun->lun_lock);
12184                 ctl_free_io(io);
12185                 break;
12186         case CTL_MSG_PERS_ACTION:
12187                 ctl_hndl_per_res_out_on_other_sc(io);
12188                 ctl_free_io(io);
12189                 break;
12190         case CTL_MSG_BAD_JUJU:
12191                 ctl_done(io);
12192                 break;
12193         case CTL_MSG_DATAMOVE:          /* Only used in XFER mode */
12194                 ctl_datamove_remote(io);
12195                 break;
12196         case CTL_MSG_DATAMOVE_DONE:     /* Only used in XFER mode */
12197                 io->scsiio.be_move_done(io);
12198                 break;
12199         case CTL_MSG_FAILOVER:
12200                 ctl_failover_lun(io);
12201                 ctl_free_io(io);
12202                 break;
12203         default:
12204                 printf("%s: Invalid message type %d\n",
12205                        __func__, io->io_hdr.msg_type);
12206                 ctl_free_io(io);
12207                 break;
12208         }
12209
12210 }
12211
12212
12213 /*
12214  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12215  * there is no match.
12216  */
12217 static ctl_lun_error_pattern
12218 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12219 {
12220         const struct ctl_cmd_entry *entry;
12221         ctl_lun_error_pattern filtered_pattern, pattern;
12222
12223         pattern = desc->error_pattern;
12224
12225         /*
12226          * XXX KDM we need more data passed into this function to match a
12227          * custom pattern, and we actually need to implement custom pattern
12228          * matching.
12229          */
12230         if (pattern & CTL_LUN_PAT_CMD)
12231                 return (CTL_LUN_PAT_CMD);
12232
12233         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12234                 return (CTL_LUN_PAT_ANY);
12235
12236         entry = ctl_get_cmd_entry(ctsio, NULL);
12237
12238         filtered_pattern = entry->pattern & pattern;
12239
12240         /*
12241          * If the user requested specific flags in the pattern (e.g.
12242          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12243          * flags.
12244          *
12245          * If the user did not specify any flags, it doesn't matter whether
12246          * or not the command supports the flags.
12247          */
12248         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12249              (pattern & ~CTL_LUN_PAT_MASK))
12250                 return (CTL_LUN_PAT_NONE);
12251
12252         /*
12253          * If the user asked for a range check, see if the requested LBA
12254          * range overlaps with this command's LBA range.
12255          */
12256         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12257                 uint64_t lba1;
12258                 uint64_t len1;
12259                 ctl_action action;
12260                 int retval;
12261
12262                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12263                 if (retval != 0)
12264                         return (CTL_LUN_PAT_NONE);
12265
12266                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12267                                               desc->lba_range.len, FALSE);
12268                 /*
12269                  * A "pass" means that the LBA ranges don't overlap, so
12270                  * this doesn't match the user's range criteria.
12271                  */
12272                 if (action == CTL_ACTION_PASS)
12273                         return (CTL_LUN_PAT_NONE);
12274         }
12275
12276         return (filtered_pattern);
12277 }
12278
12279 static void
12280 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12281 {
12282         struct ctl_error_desc *desc, *desc2;
12283
12284         mtx_assert(&lun->lun_lock, MA_OWNED);
12285
12286         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12287                 ctl_lun_error_pattern pattern;
12288                 /*
12289                  * Check to see whether this particular command matches
12290                  * the pattern in the descriptor.
12291                  */
12292                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12293                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12294                         continue;
12295
12296                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12297                 case CTL_LUN_INJ_ABORTED:
12298                         ctl_set_aborted(&io->scsiio);
12299                         break;
12300                 case CTL_LUN_INJ_MEDIUM_ERR:
12301                         ctl_set_medium_error(&io->scsiio,
12302                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12303                              CTL_FLAG_DATA_OUT);
12304                         break;
12305                 case CTL_LUN_INJ_UA:
12306                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12307                          * OCCURRED */
12308                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12309                         break;
12310                 case CTL_LUN_INJ_CUSTOM:
12311                         /*
12312                          * We're assuming the user knows what he is doing.
12313                          * Just copy the sense information without doing
12314                          * checks.
12315                          */
12316                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12317                               MIN(sizeof(desc->custom_sense),
12318                                   sizeof(io->scsiio.sense_data)));
12319                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12320                         io->scsiio.sense_len = SSD_FULL_SIZE;
12321                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12322                         break;
12323                 case CTL_LUN_INJ_NONE:
12324                 default:
12325                         /*
12326                          * If this is an error injection type we don't know
12327                          * about, clear the continuous flag (if it is set)
12328                          * so it will get deleted below.
12329                          */
12330                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12331                         break;
12332                 }
12333                 /*
12334                  * By default, each error injection action is a one-shot
12335                  */
12336                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12337                         continue;
12338
12339                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12340
12341                 free(desc, M_CTL);
12342         }
12343 }
12344
12345 #ifdef CTL_IO_DELAY
12346 static void
12347 ctl_datamove_timer_wakeup(void *arg)
12348 {
12349         union ctl_io *io;
12350
12351         io = (union ctl_io *)arg;
12352
12353         ctl_datamove(io);
12354 }
12355 #endif /* CTL_IO_DELAY */
12356
12357 void
12358 ctl_datamove(union ctl_io *io)
12359 {
12360         void (*fe_datamove)(union ctl_io *io);
12361
12362         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12363
12364         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12365
12366         /* No data transferred yet.  Frontend must update this when done. */
12367         io->scsiio.kern_data_resid = io->scsiio.kern_data_len;
12368
12369 #ifdef CTL_TIME_IO
12370         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12371                 char str[256];
12372                 char path_str[64];
12373                 struct sbuf sb;
12374
12375                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12376                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12377
12378                 sbuf_cat(&sb, path_str);
12379                 switch (io->io_hdr.io_type) {
12380                 case CTL_IO_SCSI:
12381                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12382                         sbuf_printf(&sb, "\n");
12383                         sbuf_cat(&sb, path_str);
12384                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12385                                     io->scsiio.tag_num, io->scsiio.tag_type);
12386                         break;
12387                 case CTL_IO_TASK:
12388                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12389                                     "Tag Type: %d\n", io->taskio.task_action,
12390                                     io->taskio.tag_num, io->taskio.tag_type);
12391                         break;
12392                 default:
12393                         panic("%s: Invalid CTL I/O type %d\n",
12394                             __func__, io->io_hdr.io_type);
12395                 }
12396                 sbuf_cat(&sb, path_str);
12397                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12398                             (intmax_t)time_uptime - io->io_hdr.start_time);
12399                 sbuf_finish(&sb);
12400                 printf("%s", sbuf_data(&sb));
12401         }
12402 #endif /* CTL_TIME_IO */
12403
12404 #ifdef CTL_IO_DELAY
12405         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12406                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12407         } else {
12408                 if ((lun != NULL)
12409                  && (lun->delay_info.datamove_delay > 0)) {
12410
12411                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12412                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12413                         callout_reset(&io->io_hdr.delay_callout,
12414                                       lun->delay_info.datamove_delay * hz,
12415                                       ctl_datamove_timer_wakeup, io);
12416                         if (lun->delay_info.datamove_type ==
12417                             CTL_DELAY_TYPE_ONESHOT)
12418                                 lun->delay_info.datamove_delay = 0;
12419                         return;
12420                 }
12421         }
12422 #endif
12423
12424         /*
12425          * This command has been aborted.  Set the port status, so we fail
12426          * the data move.
12427          */
12428         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12429                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12430                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12431                        io->io_hdr.nexus.targ_port,
12432                        io->io_hdr.nexus.targ_lun);
12433                 io->io_hdr.port_status = 31337;
12434                 /*
12435                  * Note that the backend, in this case, will get the
12436                  * callback in its context.  In other cases it may get
12437                  * called in the frontend's interrupt thread context.
12438                  */
12439                 io->scsiio.be_move_done(io);
12440                 return;
12441         }
12442
12443         /* Don't confuse frontend with zero length data move. */
12444         if (io->scsiio.kern_data_len == 0) {
12445                 io->scsiio.be_move_done(io);
12446                 return;
12447         }
12448
12449         fe_datamove = CTL_PORT(io)->fe_datamove;
12450         fe_datamove(io);
12451 }
12452
12453 static void
12454 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12455 {
12456         union ctl_ha_msg msg;
12457 #ifdef CTL_TIME_IO
12458         struct bintime cur_bt;
12459 #endif
12460
12461         memset(&msg, 0, sizeof(msg));
12462         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12463         msg.hdr.original_sc = io;
12464         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12465         msg.hdr.nexus = io->io_hdr.nexus;
12466         msg.hdr.status = io->io_hdr.status;
12467         msg.scsi.kern_data_resid = io->scsiio.kern_data_resid;
12468         msg.scsi.tag_num = io->scsiio.tag_num;
12469         msg.scsi.tag_type = io->scsiio.tag_type;
12470         msg.scsi.scsi_status = io->scsiio.scsi_status;
12471         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12472                io->scsiio.sense_len);
12473         msg.scsi.sense_len = io->scsiio.sense_len;
12474         msg.scsi.port_status = io->io_hdr.port_status;
12475         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12476         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12477                 ctl_failover_io(io, /*have_lock*/ have_lock);
12478                 return;
12479         }
12480         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12481             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12482             msg.scsi.sense_len, M_WAITOK);
12483
12484 #ifdef CTL_TIME_IO
12485         getbinuptime(&cur_bt);
12486         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12487         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12488 #endif
12489         io->io_hdr.num_dmas++;
12490 }
12491
12492 /*
12493  * The DMA to the remote side is done, now we need to tell the other side
12494  * we're done so it can continue with its data movement.
12495  */
12496 static void
12497 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12498 {
12499         union ctl_io *io;
12500         uint32_t i;
12501
12502         io = rq->context;
12503
12504         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12505                 printf("%s: ISC DMA write failed with error %d", __func__,
12506                        rq->ret);
12507                 ctl_set_internal_failure(&io->scsiio,
12508                                          /*sks_valid*/ 1,
12509                                          /*retry_count*/ rq->ret);
12510         }
12511
12512         ctl_dt_req_free(rq);
12513
12514         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12515                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12516         free(io->io_hdr.remote_sglist, M_CTL);
12517         io->io_hdr.remote_sglist = NULL;
12518         io->io_hdr.local_sglist = NULL;
12519
12520         /*
12521          * The data is in local and remote memory, so now we need to send
12522          * status (good or back) back to the other side.
12523          */
12524         ctl_send_datamove_done(io, /*have_lock*/ 0);
12525 }
12526
12527 /*
12528  * We've moved the data from the host/controller into local memory.  Now we
12529  * need to push it over to the remote controller's memory.
12530  */
12531 static int
12532 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12533 {
12534         int retval;
12535
12536         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12537                                           ctl_datamove_remote_write_cb);
12538         return (retval);
12539 }
12540
12541 static void
12542 ctl_datamove_remote_write(union ctl_io *io)
12543 {
12544         int retval;
12545         void (*fe_datamove)(union ctl_io *io);
12546
12547         /*
12548          * - Get the data from the host/HBA into local memory.
12549          * - DMA memory from the local controller to the remote controller.
12550          * - Send status back to the remote controller.
12551          */
12552
12553         retval = ctl_datamove_remote_sgl_setup(io);
12554         if (retval != 0)
12555                 return;
12556
12557         /* Switch the pointer over so the FETD knows what to do */
12558         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12559
12560         /*
12561          * Use a custom move done callback, since we need to send completion
12562          * back to the other controller, not to the backend on this side.
12563          */
12564         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12565
12566         fe_datamove = CTL_PORT(io)->fe_datamove;
12567         fe_datamove(io);
12568 }
12569
12570 static int
12571 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12572 {
12573 #if 0
12574         char str[256];
12575         char path_str[64];
12576         struct sbuf sb;
12577 #endif
12578         uint32_t i;
12579
12580         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12581                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12582         free(io->io_hdr.remote_sglist, M_CTL);
12583         io->io_hdr.remote_sglist = NULL;
12584         io->io_hdr.local_sglist = NULL;
12585
12586 #if 0
12587         scsi_path_string(io, path_str, sizeof(path_str));
12588         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12589         sbuf_cat(&sb, path_str);
12590         scsi_command_string(&io->scsiio, NULL, &sb);
12591         sbuf_printf(&sb, "\n");
12592         sbuf_cat(&sb, path_str);
12593         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12594                     io->scsiio.tag_num, io->scsiio.tag_type);
12595         sbuf_cat(&sb, path_str);
12596         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12597                     io->io_hdr.flags, io->io_hdr.status);
12598         sbuf_finish(&sb);
12599         printk("%s", sbuf_data(&sb));
12600 #endif
12601
12602
12603         /*
12604          * The read is done, now we need to send status (good or bad) back
12605          * to the other side.
12606          */
12607         ctl_send_datamove_done(io, /*have_lock*/ 0);
12608
12609         return (0);
12610 }
12611
12612 static void
12613 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12614 {
12615         union ctl_io *io;
12616         void (*fe_datamove)(union ctl_io *io);
12617
12618         io = rq->context;
12619
12620         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12621                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12622                        rq->ret);
12623                 ctl_set_internal_failure(&io->scsiio,
12624                                          /*sks_valid*/ 1,
12625                                          /*retry_count*/ rq->ret);
12626         }
12627
12628         ctl_dt_req_free(rq);
12629
12630         /* Switch the pointer over so the FETD knows what to do */
12631         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12632
12633         /*
12634          * Use a custom move done callback, since we need to send completion
12635          * back to the other controller, not to the backend on this side.
12636          */
12637         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12638
12639         /* XXX KDM add checks like the ones in ctl_datamove? */
12640
12641         fe_datamove = CTL_PORT(io)->fe_datamove;
12642         fe_datamove(io);
12643 }
12644
12645 static int
12646 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12647 {
12648         struct ctl_sg_entry *local_sglist;
12649         uint32_t len_to_go;
12650         int retval;
12651         int i;
12652
12653         retval = 0;
12654         local_sglist = io->io_hdr.local_sglist;
12655         len_to_go = io->scsiio.kern_data_len;
12656
12657         /*
12658          * The difficult thing here is that the size of the various
12659          * S/G segments may be different than the size from the
12660          * remote controller.  That'll make it harder when DMAing
12661          * the data back to the other side.
12662          */
12663         for (i = 0; len_to_go > 0; i++) {
12664                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12665                 local_sglist[i].addr =
12666                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12667
12668                 len_to_go -= local_sglist[i].len;
12669         }
12670         /*
12671          * Reset the number of S/G entries accordingly.  The original
12672          * number of S/G entries is available in rem_sg_entries.
12673          */
12674         io->scsiio.kern_sg_entries = i;
12675
12676 #if 0
12677         printf("%s: kern_sg_entries = %d\n", __func__,
12678                io->scsiio.kern_sg_entries);
12679         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12680                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12681                        local_sglist[i].addr, local_sglist[i].len);
12682 #endif
12683
12684         return (retval);
12685 }
12686
12687 static int
12688 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12689                          ctl_ha_dt_cb callback)
12690 {
12691         struct ctl_ha_dt_req *rq;
12692         struct ctl_sg_entry *remote_sglist, *local_sglist;
12693         uint32_t local_used, remote_used, total_used;
12694         int i, j, isc_ret;
12695
12696         rq = ctl_dt_req_alloc();
12697
12698         /*
12699          * If we failed to allocate the request, and if the DMA didn't fail
12700          * anyway, set busy status.  This is just a resource allocation
12701          * failure.
12702          */
12703         if ((rq == NULL)
12704          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12705              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12706                 ctl_set_busy(&io->scsiio);
12707
12708         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12709             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12710
12711                 if (rq != NULL)
12712                         ctl_dt_req_free(rq);
12713
12714                 /*
12715                  * The data move failed.  We need to return status back
12716                  * to the other controller.  No point in trying to DMA
12717                  * data to the remote controller.
12718                  */
12719
12720                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12721
12722                 return (1);
12723         }
12724
12725         local_sglist = io->io_hdr.local_sglist;
12726         remote_sglist = io->io_hdr.remote_sglist;
12727         local_used = 0;
12728         remote_used = 0;
12729         total_used = 0;
12730
12731         /*
12732          * Pull/push the data over the wire from/to the other controller.
12733          * This takes into account the possibility that the local and
12734          * remote sglists may not be identical in terms of the size of
12735          * the elements and the number of elements.
12736          *
12737          * One fundamental assumption here is that the length allocated for
12738          * both the local and remote sglists is identical.  Otherwise, we've
12739          * essentially got a coding error of some sort.
12740          */
12741         isc_ret = CTL_HA_STATUS_SUCCESS;
12742         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12743                 uint32_t cur_len;
12744                 uint8_t *tmp_ptr;
12745
12746                 rq->command = command;
12747                 rq->context = io;
12748
12749                 /*
12750                  * Both pointers should be aligned.  But it is possible
12751                  * that the allocation length is not.  They should both
12752                  * also have enough slack left over at the end, though,
12753                  * to round up to the next 8 byte boundary.
12754                  */
12755                 cur_len = MIN(local_sglist[i].len - local_used,
12756                               remote_sglist[j].len - remote_used);
12757                 rq->size = cur_len;
12758
12759                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12760                 tmp_ptr += local_used;
12761
12762 #if 0
12763                 /* Use physical addresses when talking to ISC hardware */
12764                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12765                         /* XXX KDM use busdma */
12766                         rq->local = vtophys(tmp_ptr);
12767                 } else
12768                         rq->local = tmp_ptr;
12769 #else
12770                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12771                     ("HA does not support BUS_ADDR"));
12772                 rq->local = tmp_ptr;
12773 #endif
12774
12775                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12776                 tmp_ptr += remote_used;
12777                 rq->remote = tmp_ptr;
12778
12779                 rq->callback = NULL;
12780
12781                 local_used += cur_len;
12782                 if (local_used >= local_sglist[i].len) {
12783                         i++;
12784                         local_used = 0;
12785                 }
12786
12787                 remote_used += cur_len;
12788                 if (remote_used >= remote_sglist[j].len) {
12789                         j++;
12790                         remote_used = 0;
12791                 }
12792                 total_used += cur_len;
12793
12794                 if (total_used >= io->scsiio.kern_data_len)
12795                         rq->callback = callback;
12796
12797 #if 0
12798                 printf("%s: %s: local %p remote %p size %d\n", __func__,
12799                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12800                        rq->local, rq->remote, rq->size);
12801 #endif
12802
12803                 isc_ret = ctl_dt_single(rq);
12804                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
12805                         break;
12806         }
12807         if (isc_ret != CTL_HA_STATUS_WAIT) {
12808                 rq->ret = isc_ret;
12809                 callback(rq);
12810         }
12811
12812         return (0);
12813 }
12814
12815 static void
12816 ctl_datamove_remote_read(union ctl_io *io)
12817 {
12818         int retval;
12819         uint32_t i;
12820
12821         /*
12822          * This will send an error to the other controller in the case of a
12823          * failure.
12824          */
12825         retval = ctl_datamove_remote_sgl_setup(io);
12826         if (retval != 0)
12827                 return;
12828
12829         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12830                                           ctl_datamove_remote_read_cb);
12831         if (retval != 0) {
12832                 /*
12833                  * Make sure we free memory if there was an error..  The
12834                  * ctl_datamove_remote_xfer() function will send the
12835                  * datamove done message, or call the callback with an
12836                  * error if there is a problem.
12837                  */
12838                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12839                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12840                 free(io->io_hdr.remote_sglist, M_CTL);
12841                 io->io_hdr.remote_sglist = NULL;
12842                 io->io_hdr.local_sglist = NULL;
12843         }
12844 }
12845
12846 /*
12847  * Process a datamove request from the other controller.  This is used for
12848  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12849  * first.  Once that is complete, the data gets DMAed into the remote
12850  * controller's memory.  For reads, we DMA from the remote controller's
12851  * memory into our memory first, and then move it out to the FETD.
12852  */
12853 static void
12854 ctl_datamove_remote(union ctl_io *io)
12855 {
12856
12857         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12858
12859         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12860                 ctl_failover_io(io, /*have_lock*/ 0);
12861                 return;
12862         }
12863
12864         /*
12865          * Note that we look for an aborted I/O here, but don't do some of
12866          * the other checks that ctl_datamove() normally does.
12867          * We don't need to run the datamove delay code, since that should
12868          * have been done if need be on the other controller.
12869          */
12870         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12871                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
12872                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12873                        io->io_hdr.nexus.targ_port,
12874                        io->io_hdr.nexus.targ_lun);
12875                 io->io_hdr.port_status = 31338;
12876                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12877                 return;
12878         }
12879
12880         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
12881                 ctl_datamove_remote_write(io);
12882         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
12883                 ctl_datamove_remote_read(io);
12884         else {
12885                 io->io_hdr.port_status = 31339;
12886                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12887         }
12888 }
12889
12890 static void
12891 ctl_process_done(union ctl_io *io)
12892 {
12893         struct ctl_softc *softc = CTL_SOFTC(io);
12894         struct ctl_port *port = CTL_PORT(io);
12895         struct ctl_lun *lun = CTL_LUN(io);
12896         void (*fe_done)(union ctl_io *io);
12897         union ctl_ha_msg msg;
12898
12899         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12900         fe_done = port->fe_done;
12901
12902 #ifdef CTL_TIME_IO
12903         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12904                 char str[256];
12905                 char path_str[64];
12906                 struct sbuf sb;
12907
12908                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12909                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12910
12911                 sbuf_cat(&sb, path_str);
12912                 switch (io->io_hdr.io_type) {
12913                 case CTL_IO_SCSI:
12914                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12915                         sbuf_printf(&sb, "\n");
12916                         sbuf_cat(&sb, path_str);
12917                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12918                                     io->scsiio.tag_num, io->scsiio.tag_type);
12919                         break;
12920                 case CTL_IO_TASK:
12921                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12922                                     "Tag Type: %d\n", io->taskio.task_action,
12923                                     io->taskio.tag_num, io->taskio.tag_type);
12924                         break;
12925                 default:
12926                         panic("%s: Invalid CTL I/O type %d\n",
12927                             __func__, io->io_hdr.io_type);
12928                 }
12929                 sbuf_cat(&sb, path_str);
12930                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12931                             (intmax_t)time_uptime - io->io_hdr.start_time);
12932                 sbuf_finish(&sb);
12933                 printf("%s", sbuf_data(&sb));
12934         }
12935 #endif /* CTL_TIME_IO */
12936
12937         switch (io->io_hdr.io_type) {
12938         case CTL_IO_SCSI:
12939                 break;
12940         case CTL_IO_TASK:
12941                 if (ctl_debug & CTL_DEBUG_INFO)
12942                         ctl_io_error_print(io, NULL);
12943                 fe_done(io);
12944                 return;
12945         default:
12946                 panic("%s: Invalid CTL I/O type %d\n",
12947                     __func__, io->io_hdr.io_type);
12948         }
12949
12950         if (lun == NULL) {
12951                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12952                                  io->io_hdr.nexus.targ_mapped_lun));
12953                 goto bailout;
12954         }
12955
12956         mtx_lock(&lun->lun_lock);
12957
12958         /*
12959          * Check to see if we have any informational exception and status
12960          * of this command can be modified to report it in form of either
12961          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
12962          */
12963         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
12964             io->io_hdr.status == CTL_SUCCESS &&
12965             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
12966                 uint8_t mrie = lun->MODE_IE.mrie;
12967                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
12968                     (lun->MODE_VER.byte3 & SMS_VER_PER));
12969                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
12970                      mrie == SIEP_MRIE_REC_UNCOND ||
12971                      mrie == SIEP_MRIE_NO_SENSE) &&
12972                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
12973                      CTL_CMD_FLAG_NO_SENSE) == 0) {
12974                         ctl_set_sense(&io->scsiio,
12975                               /*current_error*/ 1,
12976                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
12977                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
12978                               /*asc*/ lun->ie_asc,
12979                               /*ascq*/ lun->ie_ascq,
12980                               SSD_ELEM_NONE);
12981                         lun->ie_reported = 1;
12982                 }
12983         } else if (lun->ie_reported < 0)
12984                 lun->ie_reported = 0;
12985
12986         /*
12987          * Check to see if we have any errors to inject here.  We only
12988          * inject errors for commands that don't already have errors set.
12989          */
12990         if (!STAILQ_EMPTY(&lun->error_list) &&
12991             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
12992             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
12993                 ctl_inject_error(lun, io);
12994
12995         /*
12996          * XXX KDM how do we treat commands that aren't completed
12997          * successfully?
12998          *
12999          * XXX KDM should we also track I/O latency?
13000          */
13001         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13002             io->io_hdr.io_type == CTL_IO_SCSI) {
13003                 int type;
13004 #ifdef CTL_TIME_IO
13005                 struct bintime bt;
13006
13007                 getbinuptime(&bt);
13008                 bintime_sub(&bt, &io->io_hdr.start_bt);
13009 #endif
13010                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13011                     CTL_FLAG_DATA_IN)
13012                         type = CTL_STATS_READ;
13013                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13014                     CTL_FLAG_DATA_OUT)
13015                         type = CTL_STATS_WRITE;
13016                 else
13017                         type = CTL_STATS_NO_IO;
13018
13019 #ifdef CTL_LEGACY_STATS
13020                 uint32_t targ_port = port->targ_port;
13021                 lun->legacy_stats.ports[targ_port].bytes[type] +=
13022                     io->scsiio.kern_total_len;
13023                 lun->legacy_stats.ports[targ_port].operations[type] ++;
13024                 lun->legacy_stats.ports[targ_port].num_dmas[type] +=
13025                     io->io_hdr.num_dmas;
13026 #ifdef CTL_TIME_IO
13027                 bintime_add(&lun->legacy_stats.ports[targ_port].dma_time[type],
13028                    &io->io_hdr.dma_bt);
13029                 bintime_add(&lun->legacy_stats.ports[targ_port].time[type],
13030                     &bt);
13031 #endif
13032 #endif /* CTL_LEGACY_STATS */
13033
13034                 lun->stats.bytes[type] += io->scsiio.kern_total_len;
13035                 lun->stats.operations[type] ++;
13036                 lun->stats.dmas[type] += io->io_hdr.num_dmas;
13037 #ifdef CTL_TIME_IO
13038                 bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt);
13039                 bintime_add(&lun->stats.time[type], &bt);
13040 #endif
13041
13042                 mtx_lock(&port->port_lock);
13043                 port->stats.bytes[type] += io->scsiio.kern_total_len;
13044                 port->stats.operations[type] ++;
13045                 port->stats.dmas[type] += io->io_hdr.num_dmas;
13046 #ifdef CTL_TIME_IO
13047                 bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt);
13048                 bintime_add(&port->stats.time[type], &bt);
13049 #endif
13050                 mtx_unlock(&port->port_lock);
13051         }
13052
13053         /*
13054          * Remove this from the OOA queue.
13055          */
13056         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13057 #ifdef CTL_TIME_IO
13058         if (TAILQ_EMPTY(&lun->ooa_queue))
13059                 lun->last_busy = getsbinuptime();
13060 #endif
13061
13062         /*
13063          * Run through the blocked queue on this LUN and see if anything
13064          * has become unblocked, now that this transaction is done.
13065          */
13066         ctl_check_blocked(lun);
13067
13068         /*
13069          * If the LUN has been invalidated, free it if there is nothing
13070          * left on its OOA queue.
13071          */
13072         if ((lun->flags & CTL_LUN_INVALID)
13073          && TAILQ_EMPTY(&lun->ooa_queue)) {
13074                 mtx_unlock(&lun->lun_lock);
13075                 mtx_lock(&softc->ctl_lock);
13076                 ctl_free_lun(lun);
13077                 mtx_unlock(&softc->ctl_lock);
13078         } else
13079                 mtx_unlock(&lun->lun_lock);
13080
13081 bailout:
13082
13083         /*
13084          * If this command has been aborted, make sure we set the status
13085          * properly.  The FETD is responsible for freeing the I/O and doing
13086          * whatever it needs to do to clean up its state.
13087          */
13088         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13089                 ctl_set_task_aborted(&io->scsiio);
13090
13091         /*
13092          * If enabled, print command error status.
13093          */
13094         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13095             (ctl_debug & CTL_DEBUG_INFO) != 0)
13096                 ctl_io_error_print(io, NULL);
13097
13098         /*
13099          * Tell the FETD or the other shelf controller we're done with this
13100          * command.  Note that only SCSI commands get to this point.  Task
13101          * management commands are completed above.
13102          */
13103         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13104             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13105                 memset(&msg, 0, sizeof(msg));
13106                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13107                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13108                 msg.hdr.nexus = io->io_hdr.nexus;
13109                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13110                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13111                     M_WAITOK);
13112         }
13113
13114         fe_done(io);
13115 }
13116
13117 #ifdef CTL_WITH_CA
13118 /*
13119  * Front end should call this if it doesn't do autosense.  When the request
13120  * sense comes back in from the initiator, we'll dequeue this and send it.
13121  */
13122 int
13123 ctl_queue_sense(union ctl_io *io)
13124 {
13125         struct ctl_softc *softc = CTL_SOFTC(io);
13126         struct ctl_port *port = CTL_PORT(io);
13127         struct ctl_lun *lun;
13128         uint32_t initidx, targ_lun;
13129
13130         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13131
13132         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13133
13134         /*
13135          * LUN lookup will likely move to the ctl_work_thread() once we
13136          * have our new queueing infrastructure (that doesn't put things on
13137          * a per-LUN queue initially).  That is so that we can handle
13138          * things like an INQUIRY to a LUN that we don't have enabled.  We
13139          * can't deal with that right now.
13140          * If we don't have a LUN for this, just toss the sense information.
13141          */
13142         mtx_lock(&softc->ctl_lock);
13143         if (targ_lun >= CTL_MAX_LUNS ||
13144             (lun = softc->ctl_luns[targ_lun]) == NULL) {
13145                 mtx_unlock(&softc->ctl_lock);
13146                 goto bailout;
13147         }
13148         mtx_lock(&lun->lun_lock);
13149         mtx_unlock(&softc->ctl_lock);
13150
13151         /*
13152          * Already have CA set for this LUN...toss the sense information.
13153          */
13154         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13155         if (ctl_is_set(lun->have_ca, initidx)) {
13156                 mtx_unlock(&lun->lun_lock);
13157                 goto bailout;
13158         }
13159
13160         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13161                MIN(sizeof(lun->pending_sense[initidx]),
13162                sizeof(io->scsiio.sense_data)));
13163         ctl_set_mask(lun->have_ca, initidx);
13164         mtx_unlock(&lun->lun_lock);
13165
13166 bailout:
13167         ctl_free_io(io);
13168         return (CTL_RETVAL_COMPLETE);
13169 }
13170 #endif
13171
13172 /*
13173  * Primary command inlet from frontend ports.  All SCSI and task I/O
13174  * requests must go through this function.
13175  */
13176 int
13177 ctl_queue(union ctl_io *io)
13178 {
13179         struct ctl_port *port = CTL_PORT(io);
13180
13181         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13182
13183 #ifdef CTL_TIME_IO
13184         io->io_hdr.start_time = time_uptime;
13185         getbinuptime(&io->io_hdr.start_bt);
13186 #endif /* CTL_TIME_IO */
13187
13188         /* Map FE-specific LUN ID into global one. */
13189         io->io_hdr.nexus.targ_mapped_lun =
13190             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13191
13192         switch (io->io_hdr.io_type) {
13193         case CTL_IO_SCSI:
13194         case CTL_IO_TASK:
13195                 if (ctl_debug & CTL_DEBUG_CDB)
13196                         ctl_io_print(io);
13197                 ctl_enqueue_incoming(io);
13198                 break;
13199         default:
13200                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13201                 return (EINVAL);
13202         }
13203
13204         return (CTL_RETVAL_COMPLETE);
13205 }
13206
13207 #ifdef CTL_IO_DELAY
13208 static void
13209 ctl_done_timer_wakeup(void *arg)
13210 {
13211         union ctl_io *io;
13212
13213         io = (union ctl_io *)arg;
13214         ctl_done(io);
13215 }
13216 #endif /* CTL_IO_DELAY */
13217
13218 void
13219 ctl_serseq_done(union ctl_io *io)
13220 {
13221         struct ctl_lun *lun = CTL_LUN(io);;
13222
13223         if (lun->be_lun == NULL ||
13224             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13225                 return;
13226         mtx_lock(&lun->lun_lock);
13227         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13228         ctl_check_blocked(lun);
13229         mtx_unlock(&lun->lun_lock);
13230 }
13231
13232 void
13233 ctl_done(union ctl_io *io)
13234 {
13235
13236         /*
13237          * Enable this to catch duplicate completion issues.
13238          */
13239 #if 0
13240         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13241                 printf("%s: type %d msg %d cdb %x iptl: "
13242                        "%u:%u:%u tag 0x%04x "
13243                        "flag %#x status %x\n",
13244                         __func__,
13245                         io->io_hdr.io_type,
13246                         io->io_hdr.msg_type,
13247                         io->scsiio.cdb[0],
13248                         io->io_hdr.nexus.initid,
13249                         io->io_hdr.nexus.targ_port,
13250                         io->io_hdr.nexus.targ_lun,
13251                         (io->io_hdr.io_type ==
13252                         CTL_IO_TASK) ?
13253                         io->taskio.tag_num :
13254                         io->scsiio.tag_num,
13255                         io->io_hdr.flags,
13256                         io->io_hdr.status);
13257         } else
13258                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13259 #endif
13260
13261         /*
13262          * This is an internal copy of an I/O, and should not go through
13263          * the normal done processing logic.
13264          */
13265         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13266                 return;
13267
13268 #ifdef CTL_IO_DELAY
13269         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13270                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13271         } else {
13272                 struct ctl_lun *lun = CTL_LUN(io);
13273
13274                 if ((lun != NULL)
13275                  && (lun->delay_info.done_delay > 0)) {
13276
13277                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13278                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13279                         callout_reset(&io->io_hdr.delay_callout,
13280                                       lun->delay_info.done_delay * hz,
13281                                       ctl_done_timer_wakeup, io);
13282                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13283                                 lun->delay_info.done_delay = 0;
13284                         return;
13285                 }
13286         }
13287 #endif /* CTL_IO_DELAY */
13288
13289         ctl_enqueue_done(io);
13290 }
13291
13292 static void
13293 ctl_work_thread(void *arg)
13294 {
13295         struct ctl_thread *thr = (struct ctl_thread *)arg;
13296         struct ctl_softc *softc = thr->ctl_softc;
13297         union ctl_io *io;
13298         int retval;
13299
13300         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13301
13302         while (!softc->shutdown) {
13303                 /*
13304                  * We handle the queues in this order:
13305                  * - ISC
13306                  * - done queue (to free up resources, unblock other commands)
13307                  * - RtR queue
13308                  * - incoming queue
13309                  *
13310                  * If those queues are empty, we break out of the loop and
13311                  * go to sleep.
13312                  */
13313                 mtx_lock(&thr->queue_lock);
13314                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13315                 if (io != NULL) {
13316                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13317                         mtx_unlock(&thr->queue_lock);
13318                         ctl_handle_isc(io);
13319                         continue;
13320                 }
13321                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13322                 if (io != NULL) {
13323                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13324                         /* clear any blocked commands, call fe_done */
13325                         mtx_unlock(&thr->queue_lock);
13326                         ctl_process_done(io);
13327                         continue;
13328                 }
13329                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13330                 if (io != NULL) {
13331                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13332                         mtx_unlock(&thr->queue_lock);
13333                         if (io->io_hdr.io_type == CTL_IO_TASK)
13334                                 ctl_run_task(io);
13335                         else
13336                                 ctl_scsiio_precheck(softc, &io->scsiio);
13337                         continue;
13338                 }
13339                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13340                 if (io != NULL) {
13341                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13342                         mtx_unlock(&thr->queue_lock);
13343                         retval = ctl_scsiio(&io->scsiio);
13344                         if (retval != CTL_RETVAL_COMPLETE)
13345                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13346                         continue;
13347                 }
13348
13349                 /* Sleep until we have something to do. */
13350                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13351         }
13352         thr->thread = NULL;
13353         kthread_exit();
13354 }
13355
13356 static void
13357 ctl_lun_thread(void *arg)
13358 {
13359         struct ctl_softc *softc = (struct ctl_softc *)arg;
13360         struct ctl_be_lun *be_lun;
13361
13362         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13363
13364         while (!softc->shutdown) {
13365                 mtx_lock(&softc->ctl_lock);
13366                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13367                 if (be_lun != NULL) {
13368                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13369                         mtx_unlock(&softc->ctl_lock);
13370                         ctl_create_lun(be_lun);
13371                         continue;
13372                 }
13373
13374                 /* Sleep until we have something to do. */
13375                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13376                     PDROP | PRIBIO, "-", 0);
13377         }
13378         softc->lun_thread = NULL;
13379         kthread_exit();
13380 }
13381
13382 static void
13383 ctl_thresh_thread(void *arg)
13384 {
13385         struct ctl_softc *softc = (struct ctl_softc *)arg;
13386         struct ctl_lun *lun;
13387         struct ctl_logical_block_provisioning_page *page;
13388         const char *attr;
13389         union ctl_ha_msg msg;
13390         uint64_t thres, val;
13391         int i, e, set;
13392
13393         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13394
13395         while (!softc->shutdown) {
13396                 mtx_lock(&softc->ctl_lock);
13397                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13398                         if ((lun->flags & CTL_LUN_DISABLED) ||
13399                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13400                             lun->backend->lun_attr == NULL)
13401                                 continue;
13402                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13403                             softc->ha_mode == CTL_HA_MODE_XFER)
13404                                 continue;
13405                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13406                                 continue;
13407                         e = 0;
13408                         page = &lun->MODE_LBP;
13409                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13410                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13411                                         continue;
13412                                 thres = scsi_4btoul(page->descr[i].count);
13413                                 thres <<= CTL_LBP_EXPONENT;
13414                                 switch (page->descr[i].resource) {
13415                                 case 0x01:
13416                                         attr = "blocksavail";
13417                                         break;
13418                                 case 0x02:
13419                                         attr = "blocksused";
13420                                         break;
13421                                 case 0xf1:
13422                                         attr = "poolblocksavail";
13423                                         break;
13424                                 case 0xf2:
13425                                         attr = "poolblocksused";
13426                                         break;
13427                                 default:
13428                                         continue;
13429                                 }
13430                                 mtx_unlock(&softc->ctl_lock); // XXX
13431                                 val = lun->backend->lun_attr(
13432                                     lun->be_lun->be_lun, attr);
13433                                 mtx_lock(&softc->ctl_lock);
13434                                 if (val == UINT64_MAX)
13435                                         continue;
13436                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13437                                     == SLBPPD_ARMING_INC)
13438                                         e = (val >= thres);
13439                                 else
13440                                         e = (val <= thres);
13441                                 if (e)
13442                                         break;
13443                         }
13444                         mtx_lock(&lun->lun_lock);
13445                         if (e) {
13446                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13447                                     (uint8_t *)page, lun->ua_tpt_info);
13448                                 if (lun->lasttpt == 0 ||
13449                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13450                                         lun->lasttpt = time_uptime;
13451                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13452                                         set = 1;
13453                                 } else
13454                                         set = 0;
13455                         } else {
13456                                 lun->lasttpt = 0;
13457                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13458                                 set = -1;
13459                         }
13460                         mtx_unlock(&lun->lun_lock);
13461                         if (set != 0 &&
13462                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13463                                 /* Send msg to other side. */
13464                                 bzero(&msg.ua, sizeof(msg.ua));
13465                                 msg.hdr.msg_type = CTL_MSG_UA;
13466                                 msg.hdr.nexus.initid = -1;
13467                                 msg.hdr.nexus.targ_port = -1;
13468                                 msg.hdr.nexus.targ_lun = lun->lun;
13469                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13470                                 msg.ua.ua_all = 1;
13471                                 msg.ua.ua_set = (set > 0);
13472                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13473                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13474                                 mtx_unlock(&softc->ctl_lock); // XXX
13475                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13476                                     sizeof(msg.ua), M_WAITOK);
13477                                 mtx_lock(&softc->ctl_lock);
13478                         }
13479                 }
13480                 mtx_sleep(&softc->thresh_thread, &softc->ctl_lock,
13481                     PDROP | PRIBIO, "-", CTL_LBP_PERIOD * hz);
13482         }
13483         softc->thresh_thread = NULL;
13484         kthread_exit();
13485 }
13486
13487 static void
13488 ctl_enqueue_incoming(union ctl_io *io)
13489 {
13490         struct ctl_softc *softc = CTL_SOFTC(io);
13491         struct ctl_thread *thr;
13492         u_int idx;
13493
13494         idx = (io->io_hdr.nexus.targ_port * 127 +
13495                io->io_hdr.nexus.initid) % worker_threads;
13496         thr = &softc->threads[idx];
13497         mtx_lock(&thr->queue_lock);
13498         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13499         mtx_unlock(&thr->queue_lock);
13500         wakeup(thr);
13501 }
13502
13503 static void
13504 ctl_enqueue_rtr(union ctl_io *io)
13505 {
13506         struct ctl_softc *softc = CTL_SOFTC(io);
13507         struct ctl_thread *thr;
13508
13509         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13510         mtx_lock(&thr->queue_lock);
13511         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13512         mtx_unlock(&thr->queue_lock);
13513         wakeup(thr);
13514 }
13515
13516 static void
13517 ctl_enqueue_done(union ctl_io *io)
13518 {
13519         struct ctl_softc *softc = CTL_SOFTC(io);
13520         struct ctl_thread *thr;
13521
13522         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13523         mtx_lock(&thr->queue_lock);
13524         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13525         mtx_unlock(&thr->queue_lock);
13526         wakeup(thr);
13527 }
13528
13529 static void
13530 ctl_enqueue_isc(union ctl_io *io)
13531 {
13532         struct ctl_softc *softc = CTL_SOFTC(io);
13533         struct ctl_thread *thr;
13534
13535         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13536         mtx_lock(&thr->queue_lock);
13537         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13538         mtx_unlock(&thr->queue_lock);
13539         wakeup(thr);
13540 }
13541
13542 /*
13543  *  vim: ts=8
13544  */