sys/cam/ctl/ctl.c

   1 /*-
   2  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
   3  * Copyright (c) 2012 The FreeBSD Foundation
   4  * Copyright (c) 2014-2017 Alexander Motin <mav@FreeBSD.org>
   5  * All rights reserved.
   6  *
   7  * Portions of this software were developed by Edward Tomasz Napierala
   8  * under sponsorship from the FreeBSD Foundation.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions, and the following disclaimer,
  15  *    without modification.
  16  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
  17  *    substantially similar to the "NO WARRANTY" disclaimer below
  18  *    ("Disclaimer") and any redistribution must be conditioned upon
  19  *    including a substantially similar Disclaimer requirement for further
  20  *    binary redistribution.
  21  *
  22  * NO WARRANTY
  23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
  26  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  27  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  31  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
  32  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  33  * POSSIBILITY OF SUCH DAMAGES.
  34  *
  35  * $Id$
  36  */
  37 /*
  38  * CAM Target Layer, a SCSI device emulation subsystem.
  39  *
  40  * Author: Ken Merry <ken@FreeBSD.org>
  41  */
  42
  43 #define _CTL_C
  44
  45 #include <sys/cdefs.h>
  46 __FBSDID("$FreeBSD$");
  47
  48 #include <sys/param.h>
  49 #include <sys/systm.h>
  50 #include <sys/ctype.h>
  51 #include <sys/kernel.h>
  52 #include <sys/types.h>
  53 #include <sys/kthread.h>
  54 #include <sys/bio.h>
  55 #include <sys/fcntl.h>
  56 #include <sys/lock.h>
  57 #include <sys/module.h>
  58 #include <sys/mutex.h>
  59 #include <sys/condvar.h>
  60 #include <sys/malloc.h>
  61 #include <sys/conf.h>
  62 #include <sys/ioccom.h>
  63 #include <sys/queue.h>
  64 #include <sys/sbuf.h>
  65 #include <sys/smp.h>
  66 #include <sys/endian.h>
  67 #include <sys/sysctl.h>
  68 #include <vm/uma.h>
  69
  70 #include <cam/cam.h>
  71 #include <cam/scsi/scsi_all.h>
  72 #include <cam/scsi/scsi_cd.h>
  73 #include <cam/scsi/scsi_da.h>
  74 #include <cam/ctl/ctl_io.h>
  75 #include <cam/ctl/ctl.h>
  76 #include <cam/ctl/ctl_frontend.h>
  77 #include <cam/ctl/ctl_util.h>
  78 #include <cam/ctl/ctl_backend.h>
  79 #include <cam/ctl/ctl_ioctl.h>
  80 #include <cam/ctl/ctl_ha.h>
  81 #include <cam/ctl/ctl_private.h>
  82 #include <cam/ctl/ctl_debug.h>
  83 #include <cam/ctl/ctl_scsi_all.h>
  84 #include <cam/ctl/ctl_error.h>
  85
  86 struct ctl_softc *control_softc = NULL;
  87
  88 /*
  89  * Template mode pages.
  90  */
  91
  92 /*
  93  * Note that these are default values only.  The actual values will be
  94  * filled in when the user does a mode sense.
  95  */
  96 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
  97         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
  98         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
  99         /*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
 100         /*read_retry_count*/0,
 101         /*correction_span*/0,
 102         /*head_offset_count*/0,
 103         /*data_strobe_offset_cnt*/0,
 104         /*byte8*/SMS_RWER_LBPERE,
 105         /*write_retry_count*/0,
 106         /*reserved2*/0,
 107         /*recovery_time_limit*/{0, 0},
 108 };
 109
 110 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
 111         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 112         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 113         /*byte3*/SMS_RWER_PER,
 114         /*read_retry_count*/0,
 115         /*correction_span*/0,
 116         /*head_offset_count*/0,
 117         /*data_strobe_offset_cnt*/0,
 118         /*byte8*/SMS_RWER_LBPERE,
 119         /*write_retry_count*/0,
 120         /*reserved2*/0,
 121         /*recovery_time_limit*/{0, 0},
 122 };
 123
 124 const static struct scsi_format_page format_page_default = {
 125         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 126         /*page_length*/sizeof(struct scsi_format_page) - 2,
 127         /*tracks_per_zone*/ {0, 0},
 128         /*alt_sectors_per_zone*/ {0, 0},
 129         /*alt_tracks_per_zone*/ {0, 0},
 130         /*alt_tracks_per_lun*/ {0, 0},
 131         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
 132                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
 133         /*bytes_per_sector*/ {0, 0},
 134         /*interleave*/ {0, 0},
 135         /*track_skew*/ {0, 0},
 136         /*cylinder_skew*/ {0, 0},
 137         /*flags*/ SFP_HSEC,
 138         /*reserved*/ {0, 0, 0}
 139 };
 140
 141 const static struct scsi_format_page format_page_changeable = {
 142         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 143         /*page_length*/sizeof(struct scsi_format_page) - 2,
 144         /*tracks_per_zone*/ {0, 0},
 145         /*alt_sectors_per_zone*/ {0, 0},
 146         /*alt_tracks_per_zone*/ {0, 0},
 147         /*alt_tracks_per_lun*/ {0, 0},
 148         /*sectors_per_track*/ {0, 0},
 149         /*bytes_per_sector*/ {0, 0},
 150         /*interleave*/ {0, 0},
 151         /*track_skew*/ {0, 0},
 152         /*cylinder_skew*/ {0, 0},
 153         /*flags*/ 0,
 154         /*reserved*/ {0, 0, 0}
 155 };
 156
 157 const static struct scsi_rigid_disk_page rigid_disk_page_default = {
 158         /*page_code*/SMS_RIGID_DISK_PAGE,
 159         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 160         /*cylinders*/ {0, 0, 0},
 161         /*heads*/ CTL_DEFAULT_HEADS,
 162         /*start_write_precomp*/ {0, 0, 0},
 163         /*start_reduced_current*/ {0, 0, 0},
 164         /*step_rate*/ {0, 0},
 165         /*landing_zone_cylinder*/ {0, 0, 0},
 166         /*rpl*/ SRDP_RPL_DISABLED,
 167         /*rotational_offset*/ 0,
 168         /*reserved1*/ 0,
 169         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
 170                            CTL_DEFAULT_ROTATION_RATE & 0xff},
 171         /*reserved2*/ {0, 0}
 172 };
 173
 174 const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
 175         /*page_code*/SMS_RIGID_DISK_PAGE,
 176         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 177         /*cylinders*/ {0, 0, 0},
 178         /*heads*/ 0,
 179         /*start_write_precomp*/ {0, 0, 0},
 180         /*start_reduced_current*/ {0, 0, 0},
 181         /*step_rate*/ {0, 0},
 182         /*landing_zone_cylinder*/ {0, 0, 0},
 183         /*rpl*/ 0,
 184         /*rotational_offset*/ 0,
 185         /*reserved1*/ 0,
 186         /*rotation_rate*/ {0, 0},
 187         /*reserved2*/ {0, 0}
 188 };
 189
 190 const static struct scsi_da_verify_recovery_page verify_er_page_default = {
 191         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 192         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 193         /*byte3*/0,
 194         /*read_retry_count*/0,
 195         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 196         /*recovery_time_limit*/{0, 0},
 197 };
 198
 199 const static struct scsi_da_verify_recovery_page verify_er_page_changeable = {
 200         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 201         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 202         /*byte3*/SMS_VER_PER,
 203         /*read_retry_count*/0,
 204         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 205         /*recovery_time_limit*/{0, 0},
 206 };
 207
 208 const static struct scsi_caching_page caching_page_default = {
 209         /*page_code*/SMS_CACHING_PAGE,
 210         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 211         /*flags1*/ SCP_DISC | SCP_WCE,
 212         /*ret_priority*/ 0,
 213         /*disable_pf_transfer_len*/ {0xff, 0xff},
 214         /*min_prefetch*/ {0, 0},
 215         /*max_prefetch*/ {0xff, 0xff},
 216         /*max_pf_ceiling*/ {0xff, 0xff},
 217         /*flags2*/ 0,
 218         /*cache_segments*/ 0,
 219         /*cache_seg_size*/ {0, 0},
 220         /*reserved*/ 0,
 221         /*non_cache_seg_size*/ {0, 0, 0}
 222 };
 223
 224 const static struct scsi_caching_page caching_page_changeable = {
 225         /*page_code*/SMS_CACHING_PAGE,
 226         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 227         /*flags1*/ SCP_WCE | SCP_RCD,
 228         /*ret_priority*/ 0,
 229         /*disable_pf_transfer_len*/ {0, 0},
 230         /*min_prefetch*/ {0, 0},
 231         /*max_prefetch*/ {0, 0},
 232         /*max_pf_ceiling*/ {0, 0},
 233         /*flags2*/ 0,
 234         /*cache_segments*/ 0,
 235         /*cache_seg_size*/ {0, 0},
 236         /*reserved*/ 0,
 237         /*non_cache_seg_size*/ {0, 0, 0}
 238 };
 239
 240 const static struct scsi_control_page control_page_default = {
 241         /*page_code*/SMS_CONTROL_MODE_PAGE,
 242         /*page_length*/sizeof(struct scsi_control_page) - 2,
 243         /*rlec*/0,
 244         /*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
 245         /*eca_and_aen*/0,
 246         /*flags4*/SCP_TAS,
 247         /*aen_holdoff_period*/{0, 0},
 248         /*busy_timeout_period*/{0, 0},
 249         /*extended_selftest_completion_time*/{0, 0}
 250 };
 251
 252 const static struct scsi_control_page control_page_changeable = {
 253         /*page_code*/SMS_CONTROL_MODE_PAGE,
 254         /*page_length*/sizeof(struct scsi_control_page) - 2,
 255         /*rlec*/SCP_DSENSE,
 256         /*queue_flags*/SCP_QUEUE_ALG_MASK | SCP_NUAR,
 257         /*eca_and_aen*/SCP_SWP,
 258         /*flags4*/0,
 259         /*aen_holdoff_period*/{0, 0},
 260         /*busy_timeout_period*/{0, 0},
 261         /*extended_selftest_completion_time*/{0, 0}
 262 };
 263
 264 #define CTL_CEM_LEN     (sizeof(struct scsi_control_ext_page) - 4)
 265
 266 const static struct scsi_control_ext_page control_ext_page_default = {
 267         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 268         /*subpage_code*/0x01,
 269         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 270         /*flags*/0,
 271         /*prio*/0,
 272         /*max_sense*/0
 273 };
 274
 275 const static struct scsi_control_ext_page control_ext_page_changeable = {
 276         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 277         /*subpage_code*/0x01,
 278         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 279         /*flags*/0,
 280         /*prio*/0,
 281         /*max_sense*/0xff
 282 };
 283
 284 const static struct scsi_info_exceptions_page ie_page_default = {
 285         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 286         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 287         /*info_flags*/SIEP_FLAGS_EWASC,
 288         /*mrie*/SIEP_MRIE_NO,
 289         /*interval_timer*/{0, 0, 0, 0},
 290         /*report_count*/{0, 0, 0, 1}
 291 };
 292
 293 const static struct scsi_info_exceptions_page ie_page_changeable = {
 294         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 295         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 296         /*info_flags*/SIEP_FLAGS_EWASC | SIEP_FLAGS_DEXCPT | SIEP_FLAGS_TEST |
 297             SIEP_FLAGS_LOGERR,
 298         /*mrie*/0x0f,
 299         /*interval_timer*/{0xff, 0xff, 0xff, 0xff},
 300         /*report_count*/{0xff, 0xff, 0xff, 0xff}
 301 };
 302
 303 #define CTL_LBPM_LEN    (sizeof(struct ctl_logical_block_provisioning_page) - 4)
 304
 305 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
 306         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 307         /*subpage_code*/0x02,
 308         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 309         /*flags*/0,
 310         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 311         /*descr*/{}},
 312         {{/*flags*/0,
 313           /*resource*/0x01,
 314           /*reserved*/{0, 0},
 315           /*count*/{0, 0, 0, 0}},
 316          {/*flags*/0,
 317           /*resource*/0x02,
 318           /*reserved*/{0, 0},
 319           /*count*/{0, 0, 0, 0}},
 320          {/*flags*/0,
 321           /*resource*/0xf1,
 322           /*reserved*/{0, 0},
 323           /*count*/{0, 0, 0, 0}},
 324          {/*flags*/0,
 325           /*resource*/0xf2,
 326           /*reserved*/{0, 0},
 327           /*count*/{0, 0, 0, 0}}
 328         }
 329 };
 330
 331 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
 332         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 333         /*subpage_code*/0x02,
 334         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 335         /*flags*/SLBPP_SITUA,
 336         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 337         /*descr*/{}},
 338         {{/*flags*/0,
 339           /*resource*/0,
 340           /*reserved*/{0, 0},
 341           /*count*/{0, 0, 0, 0}},
 342          {/*flags*/0,
 343           /*resource*/0,
 344           /*reserved*/{0, 0},
 345           /*count*/{0, 0, 0, 0}},
 346          {/*flags*/0,
 347           /*resource*/0,
 348           /*reserved*/{0, 0},
 349           /*count*/{0, 0, 0, 0}},
 350          {/*flags*/0,
 351           /*resource*/0,
 352           /*reserved*/{0, 0},
 353           /*count*/{0, 0, 0, 0}}
 354         }
 355 };
 356
 357 const static struct scsi_cddvd_capabilities_page cddvd_page_default = {
 358         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 359         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 360         /*caps1*/0x3f,
 361         /*caps2*/0x00,
 362         /*caps3*/0xf0,
 363         /*caps4*/0x00,
 364         /*caps5*/0x29,
 365         /*caps6*/0x00,
 366         /*obsolete*/{0, 0},
 367         /*nvol_levels*/{0, 0},
 368         /*buffer_size*/{8, 0},
 369         /*obsolete2*/{0, 0},
 370         /*reserved*/0,
 371         /*digital*/0,
 372         /*obsolete3*/0,
 373         /*copy_management*/0,
 374         /*reserved2*/0,
 375         /*rotation_control*/0,
 376         /*cur_write_speed*/0,
 377         /*num_speed_descr*/0,
 378 };
 379
 380 const static struct scsi_cddvd_capabilities_page cddvd_page_changeable = {
 381         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 382         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 383         /*caps1*/0,
 384         /*caps2*/0,
 385         /*caps3*/0,
 386         /*caps4*/0,
 387         /*caps5*/0,
 388         /*caps6*/0,
 389         /*obsolete*/{0, 0},
 390         /*nvol_levels*/{0, 0},
 391         /*buffer_size*/{0, 0},
 392         /*obsolete2*/{0, 0},
 393         /*reserved*/0,
 394         /*digital*/0,
 395         /*obsolete3*/0,
 396         /*copy_management*/0,
 397         /*reserved2*/0,
 398         /*rotation_control*/0,
 399         /*cur_write_speed*/0,
 400         /*num_speed_descr*/0,
 401 };
 402
 403 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
 404 static int worker_threads = -1;
 405 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 406     &worker_threads, 1, "Number of worker threads");
 407 static int ctl_debug = CTL_DEBUG_NONE;
 408 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
 409     &ctl_debug, 0, "Enabled debug flags");
 410 static int ctl_lun_map_size = 1024;
 411 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, lun_map_size, CTLFLAG_RWTUN,
 412     &ctl_lun_map_size, 0, "Size of per-port LUN map (max LUN + 1)");
 413
 414 /*
 415  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 416  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 417  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
 418  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
 419  */
 420 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
 421
 422 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 423                                   int param);
 424 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 425 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
 426 static int ctl_init(void);
 427 static int ctl_shutdown(void);
 428 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 429 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 430 static void ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 431 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 432                               struct ctl_ooa *ooa_hdr,
 433                               struct ctl_ooa_entry *kern_entries);
 434 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 435                      struct thread *td);
 436 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 437                          struct ctl_be_lun *be_lun);
 438 static int ctl_free_lun(struct ctl_lun *lun);
 439 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 440
 441 static int ctl_do_mode_select(union ctl_io *io);
 442 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 443                            uint64_t res_key, uint64_t sa_res_key,
 444                            uint8_t type, uint32_t residx,
 445                            struct ctl_scsiio *ctsio,
 446                            struct scsi_per_res_out *cdb,
 447                            struct scsi_per_res_out_parms* param);
 448 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 449                                   union ctl_ha_msg *msg);
 450 static void ctl_hndl_per_res_out_on_other_sc(union ctl_io *io);
 451 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 452 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 453 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 454 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 455 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 456 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 457                                          int alloc_len);
 458 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 459                                          int alloc_len);
 460 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 461 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 462 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 463 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 464 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 465 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
 466     bool seq);
 467 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
 468 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 469     union ctl_io *pending_io, union ctl_io *ooa_io);
 470 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 471                                 union ctl_io *starting_io);
 472 static int ctl_check_blocked(struct ctl_lun *lun);
 473 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
 474                                 const struct ctl_cmd_entry *entry,
 475                                 struct ctl_scsiio *ctsio);
 476 static void ctl_failover_lun(union ctl_io *io);
 477 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 478                                struct ctl_scsiio *ctsio);
 479 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 480
 481 static int ctl_target_reset(union ctl_io *io);
 482 static void ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx,
 483                          ctl_ua_type ua_type);
 484 static int ctl_lun_reset(union ctl_io *io);
 485 static int ctl_abort_task(union ctl_io *io);
 486 static int ctl_abort_task_set(union ctl_io *io);
 487 static int ctl_query_task(union ctl_io *io, int task_set);
 488 static void ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
 489                               ctl_ua_type ua_type);
 490 static int ctl_i_t_nexus_reset(union ctl_io *io);
 491 static int ctl_query_async_event(union ctl_io *io);
 492 static void ctl_run_task(union ctl_io *io);
 493 #ifdef CTL_IO_DELAY
 494 static void ctl_datamove_timer_wakeup(void *arg);
 495 static void ctl_done_timer_wakeup(void *arg);
 496 #endif /* CTL_IO_DELAY */
 497
 498 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 499 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 500 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 501 static void ctl_datamove_remote_write(union ctl_io *io);
 502 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 503 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 504 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 505 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 506                                     ctl_ha_dt_cb callback);
 507 static void ctl_datamove_remote_read(union ctl_io *io);
 508 static void ctl_datamove_remote(union ctl_io *io);
 509 static void ctl_process_done(union ctl_io *io);
 510 static void ctl_lun_thread(void *arg);
 511 static void ctl_thresh_thread(void *arg);
 512 static void ctl_work_thread(void *arg);
 513 static void ctl_enqueue_incoming(union ctl_io *io);
 514 static void ctl_enqueue_rtr(union ctl_io *io);
 515 static void ctl_enqueue_done(union ctl_io *io);
 516 static void ctl_enqueue_isc(union ctl_io *io);
 517 static const struct ctl_cmd_entry *
 518     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 519 static const struct ctl_cmd_entry *
 520     ctl_validate_command(struct ctl_scsiio *ctsio);
 521 static int ctl_cmd_applicable(uint8_t lun_type,
 522     const struct ctl_cmd_entry *entry);
 523 static int ctl_ha_init(void);
 524 static int ctl_ha_shutdown(void);
 525
 526 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
 527 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
 528 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
 529 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
 530
 531 /*
 532  * Load the serialization table.  This isn't very pretty, but is probably
 533  * the easiest way to do it.
 534  */
 535 #include "ctl_ser_table.c"
 536
 537 /*
 538  * We only need to define open, close and ioctl routines for this driver.
 539  */
 540 static struct cdevsw ctl_cdevsw = {
 541         .d_version =    D_VERSION,
 542         .d_flags =      0,
 543         .d_open =       ctl_open,
 544         .d_close =      ctl_close,
 545         .d_ioctl =      ctl_ioctl,
 546         .d_name =       "ctl",
 547 };
 548
 549
 550 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 551
 552 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 553
 554 static moduledata_t ctl_moduledata = {
 555         "ctl",
 556         ctl_module_event_handler,
 557         NULL
 558 };
 559
 560 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 561 MODULE_VERSION(ctl, 1);
 562
 563 static struct ctl_frontend ha_frontend =
 564 {
 565         .name = "ha",
 566         .init = ctl_ha_init,
 567         .shutdown = ctl_ha_shutdown,
 568 };
 569
 570 static int
 571 ctl_ha_init(void)
 572 {
 573         struct ctl_softc *softc = control_softc;
 574
 575         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
 576                             &softc->othersc_pool) != 0)
 577                 return (ENOMEM);
 578         if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
 579                 ctl_pool_free(softc->othersc_pool);
 580                 return (EIO);
 581         }
 582         if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
 583             != CTL_HA_STATUS_SUCCESS) {
 584                 ctl_ha_msg_destroy(softc);
 585                 ctl_pool_free(softc->othersc_pool);
 586                 return (EIO);
 587         }
 588         return (0);
 589 };
 590
 591 static int
 592 ctl_ha_shutdown(void)
 593 {
 594         struct ctl_softc *softc = control_softc;
 595         struct ctl_port *port;
 596
 597         ctl_ha_msg_shutdown(softc);
 598         if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL) != CTL_HA_STATUS_SUCCESS)
 599                 return (EIO);
 600         if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
 601                 return (EIO);
 602         ctl_pool_free(softc->othersc_pool);
 603         while ((port = STAILQ_FIRST(&ha_frontend.port_list)) != NULL) {
 604                 ctl_port_deregister(port);
 605                 free(port->port_name, M_CTL);
 606                 free(port, M_CTL);
 607         }
 608         return (0);
 609 };
 610
 611 static void
 612 ctl_ha_datamove(union ctl_io *io)
 613 {
 614         struct ctl_lun *lun = CTL_LUN(io);
 615         struct ctl_sg_entry *sgl;
 616         union ctl_ha_msg msg;
 617         uint32_t sg_entries_sent;
 618         int do_sg_copy, i, j;
 619
 620         memset(&msg.dt, 0, sizeof(msg.dt));
 621         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
 622         msg.hdr.original_sc = io->io_hdr.original_sc;
 623         msg.hdr.serializing_sc = io;
 624         msg.hdr.nexus = io->io_hdr.nexus;
 625         msg.hdr.status = io->io_hdr.status;
 626         msg.dt.flags = io->io_hdr.flags;
 627
 628         /*
 629          * We convert everything into a S/G list here.  We can't
 630          * pass by reference, only by value between controllers.
 631          * So we can't pass a pointer to the S/G list, only as many
 632          * S/G entries as we can fit in here.  If it's possible for
 633          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
 634          * then we need to break this up into multiple transfers.
 635          */
 636         if (io->scsiio.kern_sg_entries == 0) {
 637                 msg.dt.kern_sg_entries = 1;
 638 #if 0
 639                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 640                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 641                 } else {
 642                         /* XXX KDM use busdma here! */
 643                         msg.dt.sg_list[0].addr =
 644                             (void *)vtophys(io->scsiio.kern_data_ptr);
 645                 }
 646 #else
 647                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
 648                     ("HA does not support BUS_ADDR"));
 649                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 650 #endif
 651                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
 652                 do_sg_copy = 0;
 653         } else {
 654                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
 655                 do_sg_copy = 1;
 656         }
 657
 658         msg.dt.kern_data_len = io->scsiio.kern_data_len;
 659         msg.dt.kern_total_len = io->scsiio.kern_total_len;
 660         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
 661         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
 662         msg.dt.sg_sequence = 0;
 663
 664         /*
 665          * Loop until we've sent all of the S/G entries.  On the
 666          * other end, we'll recompose these S/G entries into one
 667          * contiguous list before processing.
 668          */
 669         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
 670             msg.dt.sg_sequence++) {
 671                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
 672                     sizeof(msg.dt.sg_list[0])),
 673                     msg.dt.kern_sg_entries - sg_entries_sent);
 674                 if (do_sg_copy != 0) {
 675                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
 676                         for (i = sg_entries_sent, j = 0;
 677                              i < msg.dt.cur_sg_entries; i++, j++) {
 678 #if 0
 679                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 680                                         msg.dt.sg_list[j].addr = sgl[i].addr;
 681                                 } else {
 682                                         /* XXX KDM use busdma here! */
 683                                         msg.dt.sg_list[j].addr =
 684                                             (void *)vtophys(sgl[i].addr);
 685                                 }
 686 #else
 687                                 KASSERT((io->io_hdr.flags &
 688                                     CTL_FLAG_BUS_ADDR) == 0,
 689                                     ("HA does not support BUS_ADDR"));
 690                                 msg.dt.sg_list[j].addr = sgl[i].addr;
 691 #endif
 692                                 msg.dt.sg_list[j].len = sgl[i].len;
 693                         }
 694                 }
 695
 696                 sg_entries_sent += msg.dt.cur_sg_entries;
 697                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
 698                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 699                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
 700                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
 701                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
 702                         io->io_hdr.port_status = 31341;
 703                         io->scsiio.be_move_done(io);
 704                         return;
 705                 }
 706                 msg.dt.sent_sg_entries = sg_entries_sent;
 707         }
 708
 709         /*
 710          * Officially handover the request from us to peer.
 711          * If failover has just happened, then we must return error.
 712          * If failover happen just after, then it is not our problem.
 713          */
 714         if (lun)
 715                 mtx_lock(&lun->lun_lock);
 716         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
 717                 if (lun)
 718                         mtx_unlock(&lun->lun_lock);
 719                 io->io_hdr.port_status = 31342;
 720                 io->scsiio.be_move_done(io);
 721                 return;
 722         }
 723         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
 724         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
 725         if (lun)
 726                 mtx_unlock(&lun->lun_lock);
 727 }
 728
 729 static void
 730 ctl_ha_done(union ctl_io *io)
 731 {
 732         union ctl_ha_msg msg;
 733
 734         if (io->io_hdr.io_type == CTL_IO_SCSI) {
 735                 memset(&msg, 0, sizeof(msg));
 736                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
 737                 msg.hdr.original_sc = io->io_hdr.original_sc;
 738                 msg.hdr.nexus = io->io_hdr.nexus;
 739                 msg.hdr.status = io->io_hdr.status;
 740                 msg.scsi.scsi_status = io->scsiio.scsi_status;
 741                 msg.scsi.tag_num = io->scsiio.tag_num;
 742                 msg.scsi.tag_type = io->scsiio.tag_type;
 743                 msg.scsi.sense_len = io->scsiio.sense_len;
 744                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
 745                     io->scsiio.sense_len);
 746                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 747                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
 748                     msg.scsi.sense_len, M_WAITOK);
 749         }
 750         ctl_free_io(io);
 751 }
 752
 753 static void
 754 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 755                             union ctl_ha_msg *msg_info)
 756 {
 757         struct ctl_scsiio *ctsio;
 758
 759         if (msg_info->hdr.original_sc == NULL) {
 760                 printf("%s: original_sc == NULL!\n", __func__);
 761                 /* XXX KDM now what? */
 762                 return;
 763         }
 764
 765         ctsio = &msg_info->hdr.original_sc->scsiio;
 766         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 767         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 768         ctsio->io_hdr.status = msg_info->hdr.status;
 769         ctsio->scsi_status = msg_info->scsi.scsi_status;
 770         ctsio->sense_len = msg_info->scsi.sense_len;
 771         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 772                msg_info->scsi.sense_len);
 773         ctl_enqueue_isc((union ctl_io *)ctsio);
 774 }
 775
 776 static void
 777 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 778                                 union ctl_ha_msg *msg_info)
 779 {
 780         struct ctl_scsiio *ctsio;
 781
 782         if (msg_info->hdr.serializing_sc == NULL) {
 783                 printf("%s: serializing_sc == NULL!\n", __func__);
 784                 /* XXX KDM now what? */
 785                 return;
 786         }
 787
 788         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 789         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 790         ctl_enqueue_isc((union ctl_io *)ctsio);
 791 }
 792
 793 void
 794 ctl_isc_announce_lun(struct ctl_lun *lun)
 795 {
 796         struct ctl_softc *softc = lun->ctl_softc;
 797         union ctl_ha_msg *msg;
 798         struct ctl_ha_msg_lun_pr_key pr_key;
 799         int i, k;
 800
 801         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 802                 return;
 803         mtx_lock(&lun->lun_lock);
 804         i = sizeof(msg->lun);
 805         if (lun->lun_devid)
 806                 i += lun->lun_devid->len;
 807         i += sizeof(pr_key) * lun->pr_key_count;
 808 alloc:
 809         mtx_unlock(&lun->lun_lock);
 810         msg = malloc(i, M_CTL, M_WAITOK);
 811         mtx_lock(&lun->lun_lock);
 812         k = sizeof(msg->lun);
 813         if (lun->lun_devid)
 814                 k += lun->lun_devid->len;
 815         k += sizeof(pr_key) * lun->pr_key_count;
 816         if (i < k) {
 817                 free(msg, M_CTL);
 818                 i = k;
 819                 goto alloc;
 820         }
 821         bzero(&msg->lun, sizeof(msg->lun));
 822         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 823         msg->hdr.nexus.targ_lun = lun->lun;
 824         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 825         msg->lun.flags = lun->flags;
 826         msg->lun.pr_generation = lun->pr_generation;
 827         msg->lun.pr_res_idx = lun->pr_res_idx;
 828         msg->lun.pr_res_type = lun->pr_res_type;
 829         msg->lun.pr_key_count = lun->pr_key_count;
 830         i = 0;
 831         if (lun->lun_devid) {
 832                 msg->lun.lun_devid_len = lun->lun_devid->len;
 833                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 834                     msg->lun.lun_devid_len);
 835                 i += msg->lun.lun_devid_len;
 836         }
 837         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 838                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 839                         continue;
 840                 pr_key.pr_iid = k;
 841                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 842                 i += sizeof(pr_key);
 843         }
 844         mtx_unlock(&lun->lun_lock);
 845         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 846             M_WAITOK);
 847         free(msg, M_CTL);
 848
 849         if (lun->flags & CTL_LUN_PRIMARY_SC) {
 850                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 851                         ctl_isc_announce_mode(lun, -1,
 852                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
 853                             lun->mode_pages.index[i].subpage);
 854                 }
 855         }
 856 }
 857
 858 void
 859 ctl_isc_announce_port(struct ctl_port *port)
 860 {
 861         struct ctl_softc *softc = port->ctl_softc;
 862         union ctl_ha_msg *msg;
 863         int i;
 864
 865         if (port->targ_port < softc->port_min ||
 866             port->targ_port >= softc->port_max ||
 867             softc->ha_link != CTL_HA_LINK_ONLINE)
 868                 return;
 869         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 870         if (port->lun_map)
 871                 i += port->lun_map_size * sizeof(uint32_t);
 872         if (port->port_devid)
 873                 i += port->port_devid->len;
 874         if (port->target_devid)
 875                 i += port->target_devid->len;
 876         if (port->init_devid)
 877                 i += port->init_devid->len;
 878         msg = malloc(i, M_CTL, M_WAITOK);
 879         bzero(&msg->port, sizeof(msg->port));
 880         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 881         msg->hdr.nexus.targ_port = port->targ_port;
 882         msg->port.port_type = port->port_type;
 883         msg->port.physical_port = port->physical_port;
 884         msg->port.virtual_port = port->virtual_port;
 885         msg->port.status = port->status;
 886         i = 0;
 887         msg->port.name_len = sprintf(&msg->port.data[i],
 888             "%d:%s", softc->ha_id, port->port_name) + 1;
 889         i += msg->port.name_len;
 890         if (port->lun_map) {
 891                 msg->port.lun_map_len = port->lun_map_size * sizeof(uint32_t);
 892                 memcpy(&msg->port.data[i], port->lun_map,
 893                     msg->port.lun_map_len);
 894                 i += msg->port.lun_map_len;
 895         }
 896         if (port->port_devid) {
 897                 msg->port.port_devid_len = port->port_devid->len;
 898                 memcpy(&msg->port.data[i], port->port_devid->data,
 899                     msg->port.port_devid_len);
 900                 i += msg->port.port_devid_len;
 901         }
 902         if (port->target_devid) {
 903                 msg->port.target_devid_len = port->target_devid->len;
 904                 memcpy(&msg->port.data[i], port->target_devid->data,
 905                     msg->port.target_devid_len);
 906                 i += msg->port.target_devid_len;
 907         }
 908         if (port->init_devid) {
 909                 msg->port.init_devid_len = port->init_devid->len;
 910                 memcpy(&msg->port.data[i], port->init_devid->data,
 911                     msg->port.init_devid_len);
 912                 i += msg->port.init_devid_len;
 913         }
 914         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 915             M_WAITOK);
 916         free(msg, M_CTL);
 917 }
 918
 919 void
 920 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 921 {
 922         struct ctl_softc *softc = port->ctl_softc;
 923         union ctl_ha_msg *msg;
 924         int i, l;
 925
 926         if (port->targ_port < softc->port_min ||
 927             port->targ_port >= softc->port_max ||
 928             softc->ha_link != CTL_HA_LINK_ONLINE)
 929                 return;
 930         mtx_lock(&softc->ctl_lock);
 931         i = sizeof(msg->iid);
 932         l = 0;
 933         if (port->wwpn_iid[iid].name)
 934                 l = strlen(port->wwpn_iid[iid].name) + 1;
 935         i += l;
 936         msg = malloc(i, M_CTL, M_NOWAIT);
 937         if (msg == NULL) {
 938                 mtx_unlock(&softc->ctl_lock);
 939                 return;
 940         }
 941         bzero(&msg->iid, sizeof(msg->iid));
 942         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 943         msg->hdr.nexus.targ_port = port->targ_port;
 944         msg->hdr.nexus.initid = iid;
 945         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 946         msg->iid.name_len = l;
 947         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 948         if (port->wwpn_iid[iid].name)
 949                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 950         mtx_unlock(&softc->ctl_lock);
 951         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 952         free(msg, M_CTL);
 953 }
 954
 955 void
 956 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
 957     uint8_t page, uint8_t subpage)
 958 {
 959         struct ctl_softc *softc = lun->ctl_softc;
 960         union ctl_ha_msg msg;
 961         u_int i;
 962
 963         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 964                 return;
 965         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 966                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
 967                     page && lun->mode_pages.index[i].subpage == subpage)
 968                         break;
 969         }
 970         if (i == CTL_NUM_MODE_PAGES)
 971                 return;
 972
 973         /* Don't try to replicate pages not present on this device. */
 974         if (lun->mode_pages.index[i].page_data == NULL)
 975                 return;
 976
 977         bzero(&msg.mode, sizeof(msg.mode));
 978         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
 979         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
 980         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
 981         msg.hdr.nexus.targ_lun = lun->lun;
 982         msg.hdr.nexus.targ_mapped_lun = lun->lun;
 983         msg.mode.page_code = page;
 984         msg.mode.subpage = subpage;
 985         msg.mode.page_len = lun->mode_pages.index[i].page_len;
 986         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
 987             msg.mode.page_len);
 988         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
 989             M_WAITOK);
 990 }
 991
 992 static void
 993 ctl_isc_ha_link_up(struct ctl_softc *softc)
 994 {
 995         struct ctl_port *port;
 996         struct ctl_lun *lun;
 997         union ctl_ha_msg msg;
 998         int i;
 999
1000         /* Announce this node parameters to peer for validation. */
1001         msg.login.msg_type = CTL_MSG_LOGIN;
1002         msg.login.version = CTL_HA_VERSION;
1003         msg.login.ha_mode = softc->ha_mode;
1004         msg.login.ha_id = softc->ha_id;
1005         msg.login.max_luns = CTL_MAX_LUNS;
1006         msg.login.max_ports = CTL_MAX_PORTS;
1007         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
1008         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
1009             M_WAITOK);
1010
1011         STAILQ_FOREACH(port, &softc->port_list, links) {
1012                 ctl_isc_announce_port(port);
1013                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1014                         if (port->wwpn_iid[i].in_use)
1015                                 ctl_isc_announce_iid(port, i);
1016                 }
1017         }
1018         STAILQ_FOREACH(lun, &softc->lun_list, links)
1019                 ctl_isc_announce_lun(lun);
1020 }
1021
1022 static void
1023 ctl_isc_ha_link_down(struct ctl_softc *softc)
1024 {
1025         struct ctl_port *port;
1026         struct ctl_lun *lun;
1027         union ctl_io *io;
1028         int i;
1029
1030         mtx_lock(&softc->ctl_lock);
1031         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1032                 mtx_lock(&lun->lun_lock);
1033                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
1034                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1035                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1036                 }
1037                 mtx_unlock(&lun->lun_lock);
1038
1039                 mtx_unlock(&softc->ctl_lock);
1040                 io = ctl_alloc_io(softc->othersc_pool);
1041                 mtx_lock(&softc->ctl_lock);
1042                 ctl_zero_io(io);
1043                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1044                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1045                 ctl_enqueue_isc(io);
1046         }
1047
1048         STAILQ_FOREACH(port, &softc->port_list, links) {
1049                 if (port->targ_port >= softc->port_min &&
1050                     port->targ_port < softc->port_max)
1051                         continue;
1052                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1053                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1054                         port->wwpn_iid[i].in_use = 0;
1055                         free(port->wwpn_iid[i].name, M_CTL);
1056                         port->wwpn_iid[i].name = NULL;
1057                 }
1058         }
1059         mtx_unlock(&softc->ctl_lock);
1060 }
1061
1062 static void
1063 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1064 {
1065         struct ctl_lun *lun;
1066         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
1067
1068         mtx_lock(&softc->ctl_lock);
1069         if (msg->hdr.nexus.targ_mapped_lun >= CTL_MAX_LUNS ||
1070             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) == NULL) {
1071                 mtx_unlock(&softc->ctl_lock);
1072                 return;
1073         }
1074         mtx_lock(&lun->lun_lock);
1075         mtx_unlock(&softc->ctl_lock);
1076         if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES && msg->ua.ua_set)
1077                 memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1078         if (msg->ua.ua_all) {
1079                 if (msg->ua.ua_set)
1080                         ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1081                 else
1082                         ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1083         } else {
1084                 if (msg->ua.ua_set)
1085                         ctl_est_ua(lun, iid, msg->ua.ua_type);
1086                 else
1087                         ctl_clr_ua(lun, iid, msg->ua.ua_type);
1088         }
1089         mtx_unlock(&lun->lun_lock);
1090 }
1091
1092 static void
1093 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1094 {
1095         struct ctl_lun *lun;
1096         struct ctl_ha_msg_lun_pr_key pr_key;
1097         int i, k;
1098         ctl_lun_flags oflags;
1099         uint32_t targ_lun;
1100
1101         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1102         mtx_lock(&softc->ctl_lock);
1103         if (targ_lun >= CTL_MAX_LUNS ||
1104             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1105                 mtx_unlock(&softc->ctl_lock);
1106                 return;
1107         }
1108         mtx_lock(&lun->lun_lock);
1109         mtx_unlock(&softc->ctl_lock);
1110         if (lun->flags & CTL_LUN_DISABLED) {
1111                 mtx_unlock(&lun->lun_lock);
1112                 return;
1113         }
1114         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1115         if (msg->lun.lun_devid_len != i || (i > 0 &&
1116             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1117                 mtx_unlock(&lun->lun_lock);
1118                 printf("%s: Received conflicting HA LUN %d\n",
1119                     __func__, targ_lun);
1120                 return;
1121         } else {
1122                 /* Record whether peer is primary. */
1123                 oflags = lun->flags;
1124                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1125                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1126                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1127                 else
1128                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1129                 if (oflags != lun->flags)
1130                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1131
1132                 /* If peer is primary and we are not -- use data */
1133                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1134                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1135                         lun->pr_generation = msg->lun.pr_generation;
1136                         lun->pr_res_idx = msg->lun.pr_res_idx;
1137                         lun->pr_res_type = msg->lun.pr_res_type;
1138                         lun->pr_key_count = msg->lun.pr_key_count;
1139                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1140                                 ctl_clr_prkey(lun, k);
1141                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1142                                 memcpy(&pr_key, &msg->lun.data[i],
1143                                     sizeof(pr_key));
1144                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1145                                 ctl_set_prkey(lun, pr_key.pr_iid,
1146                                     pr_key.pr_key);
1147                                 i += sizeof(pr_key);
1148                         }
1149                 }
1150
1151                 mtx_unlock(&lun->lun_lock);
1152                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1153                     __func__, targ_lun,
1154                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1155                     "primary" : "secondary"));
1156
1157                 /* If we are primary but peer doesn't know -- notify */
1158                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1159                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1160                         ctl_isc_announce_lun(lun);
1161         }
1162 }
1163
1164 static void
1165 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1166 {
1167         struct ctl_port *port;
1168         struct ctl_lun *lun;
1169         int i, new;
1170
1171         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1172         if (port == NULL) {
1173                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1174                     msg->hdr.nexus.targ_port));
1175                 new = 1;
1176                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1177                 port->frontend = &ha_frontend;
1178                 port->targ_port = msg->hdr.nexus.targ_port;
1179                 port->fe_datamove = ctl_ha_datamove;
1180                 port->fe_done = ctl_ha_done;
1181         } else if (port->frontend == &ha_frontend) {
1182                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1183                     msg->hdr.nexus.targ_port));
1184                 new = 0;
1185         } else {
1186                 printf("%s: Received conflicting HA port %d\n",
1187                     __func__, msg->hdr.nexus.targ_port);
1188                 return;
1189         }
1190         port->port_type = msg->port.port_type;
1191         port->physical_port = msg->port.physical_port;
1192         port->virtual_port = msg->port.virtual_port;
1193         port->status = msg->port.status;
1194         i = 0;
1195         free(port->port_name, M_CTL);
1196         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1197             M_CTL);
1198         i += msg->port.name_len;
1199         if (msg->port.lun_map_len != 0) {
1200                 if (port->lun_map == NULL ||
1201                     port->lun_map_size * sizeof(uint32_t) <
1202                     msg->port.lun_map_len) {
1203                         port->lun_map_size = 0;
1204                         free(port->lun_map, M_CTL);
1205                         port->lun_map = malloc(msg->port.lun_map_len,
1206                             M_CTL, M_WAITOK);
1207                 }
1208                 memcpy(port->lun_map, &msg->port.data[i], msg->port.lun_map_len);
1209                 port->lun_map_size = msg->port.lun_map_len / sizeof(uint32_t);
1210                 i += msg->port.lun_map_len;
1211         } else {
1212                 port->lun_map_size = 0;
1213                 free(port->lun_map, M_CTL);
1214                 port->lun_map = NULL;
1215         }
1216         if (msg->port.port_devid_len != 0) {
1217                 if (port->port_devid == NULL ||
1218                     port->port_devid->len < msg->port.port_devid_len) {
1219                         free(port->port_devid, M_CTL);
1220                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1221                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1222                 }
1223                 memcpy(port->port_devid->data, &msg->port.data[i],
1224                     msg->port.port_devid_len);
1225                 port->port_devid->len = msg->port.port_devid_len;
1226                 i += msg->port.port_devid_len;
1227         } else {
1228                 free(port->port_devid, M_CTL);
1229                 port->port_devid = NULL;
1230         }
1231         if (msg->port.target_devid_len != 0) {
1232                 if (port->target_devid == NULL ||
1233                     port->target_devid->len < msg->port.target_devid_len) {
1234                         free(port->target_devid, M_CTL);
1235                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1236                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1237                 }
1238                 memcpy(port->target_devid->data, &msg->port.data[i],
1239                     msg->port.target_devid_len);
1240                 port->target_devid->len = msg->port.target_devid_len;
1241                 i += msg->port.target_devid_len;
1242         } else {
1243                 free(port->target_devid, M_CTL);
1244                 port->target_devid = NULL;
1245         }
1246         if (msg->port.init_devid_len != 0) {
1247                 if (port->init_devid == NULL ||
1248                     port->init_devid->len < msg->port.init_devid_len) {
1249                         free(port->init_devid, M_CTL);
1250                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1251                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1252                 }
1253                 memcpy(port->init_devid->data, &msg->port.data[i],
1254                     msg->port.init_devid_len);
1255                 port->init_devid->len = msg->port.init_devid_len;
1256                 i += msg->port.init_devid_len;
1257         } else {
1258                 free(port->init_devid, M_CTL);
1259                 port->init_devid = NULL;
1260         }
1261         if (new) {
1262                 if (ctl_port_register(port) != 0) {
1263                         printf("%s: ctl_port_register() failed with error\n",
1264                             __func__);
1265                 }
1266         }
1267         mtx_lock(&softc->ctl_lock);
1268         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1269                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
1270                         continue;
1271                 mtx_lock(&lun->lun_lock);
1272                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1273                 mtx_unlock(&lun->lun_lock);
1274         }
1275         mtx_unlock(&softc->ctl_lock);
1276 }
1277
1278 static void
1279 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1280 {
1281         struct ctl_port *port;
1282         int iid;
1283
1284         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1285         if (port == NULL) {
1286                 printf("%s: Received IID for unknown port %d\n",
1287                     __func__, msg->hdr.nexus.targ_port);
1288                 return;
1289         }
1290         iid = msg->hdr.nexus.initid;
1291         if (port->wwpn_iid[iid].in_use != 0 &&
1292             msg->iid.in_use == 0)
1293                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
1294         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1295         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1296         free(port->wwpn_iid[iid].name, M_CTL);
1297         if (msg->iid.name_len) {
1298                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1299                     msg->iid.name_len, M_CTL);
1300         } else
1301                 port->wwpn_iid[iid].name = NULL;
1302 }
1303
1304 static void
1305 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1306 {
1307
1308         if (msg->login.version != CTL_HA_VERSION) {
1309                 printf("CTL HA peers have different versions %d != %d\n",
1310                     msg->login.version, CTL_HA_VERSION);
1311                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1312                 return;
1313         }
1314         if (msg->login.ha_mode != softc->ha_mode) {
1315                 printf("CTL HA peers have different ha_mode %d != %d\n",
1316                     msg->login.ha_mode, softc->ha_mode);
1317                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1318                 return;
1319         }
1320         if (msg->login.ha_id == softc->ha_id) {
1321                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1322                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1323                 return;
1324         }
1325         if (msg->login.max_luns != CTL_MAX_LUNS ||
1326             msg->login.max_ports != CTL_MAX_PORTS ||
1327             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1328                 printf("CTL HA peers have different limits\n");
1329                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1330                 return;
1331         }
1332 }
1333
1334 static void
1335 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1336 {
1337         struct ctl_lun *lun;
1338         u_int i;
1339         uint32_t initidx, targ_lun;
1340
1341         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1342         mtx_lock(&softc->ctl_lock);
1343         if (targ_lun >= CTL_MAX_LUNS ||
1344             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1345                 mtx_unlock(&softc->ctl_lock);
1346                 return;
1347         }
1348         mtx_lock(&lun->lun_lock);
1349         mtx_unlock(&softc->ctl_lock);
1350         if (lun->flags & CTL_LUN_DISABLED) {
1351                 mtx_unlock(&lun->lun_lock);
1352                 return;
1353         }
1354         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1355                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1356                     msg->mode.page_code &&
1357                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1358                         break;
1359         }
1360         if (i == CTL_NUM_MODE_PAGES) {
1361                 mtx_unlock(&lun->lun_lock);
1362                 return;
1363         }
1364         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1365             lun->mode_pages.index[i].page_len);
1366         initidx = ctl_get_initindex(&msg->hdr.nexus);
1367         if (initidx != -1)
1368                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1369         mtx_unlock(&lun->lun_lock);
1370 }
1371
1372 /*
1373  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1374  * subsystem come in here.
1375  */
1376 static void
1377 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1378 {
1379         struct ctl_softc *softc = control_softc;
1380         union ctl_io *io;
1381         struct ctl_prio *presio;
1382         ctl_ha_status isc_status;
1383
1384         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1385         if (event == CTL_HA_EVT_MSG_RECV) {
1386                 union ctl_ha_msg *msg, msgbuf;
1387
1388                 if (param > sizeof(msgbuf))
1389                         msg = malloc(param, M_CTL, M_WAITOK);
1390                 else
1391                         msg = &msgbuf;
1392                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1393                     M_WAITOK);
1394                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1395                         printf("%s: Error receiving message: %d\n",
1396                             __func__, isc_status);
1397                         if (msg != &msgbuf)
1398                                 free(msg, M_CTL);
1399                         return;
1400                 }
1401
1402                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1403                 switch (msg->hdr.msg_type) {
1404                 case CTL_MSG_SERIALIZE:
1405                         io = ctl_alloc_io(softc->othersc_pool);
1406                         ctl_zero_io(io);
1407                         // populate ctsio from msg
1408                         io->io_hdr.io_type = CTL_IO_SCSI;
1409                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1410                         io->io_hdr.original_sc = msg->hdr.original_sc;
1411                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1412                                             CTL_FLAG_IO_ACTIVE;
1413                         /*
1414                          * If we're in serialization-only mode, we don't
1415                          * want to go through full done processing.  Thus
1416                          * the COPY flag.
1417                          *
1418                          * XXX KDM add another flag that is more specific.
1419                          */
1420                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1421                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1422                         io->io_hdr.nexus = msg->hdr.nexus;
1423 #if 0
1424                         printf("port %u, iid %u, lun %u\n",
1425                                io->io_hdr.nexus.targ_port,
1426                                io->io_hdr.nexus.initid,
1427                                io->io_hdr.nexus.targ_lun);
1428 #endif
1429                         io->scsiio.tag_num = msg->scsi.tag_num;
1430                         io->scsiio.tag_type = msg->scsi.tag_type;
1431 #ifdef CTL_TIME_IO
1432                         io->io_hdr.start_time = time_uptime;
1433                         getbinuptime(&io->io_hdr.start_bt);
1434 #endif /* CTL_TIME_IO */
1435                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1436                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1437                                CTL_MAX_CDBLEN);
1438                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1439                                 const struct ctl_cmd_entry *entry;
1440
1441                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1442                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1443                                 io->io_hdr.flags |=
1444                                         entry->flags & CTL_FLAG_DATA_MASK;
1445                         }
1446                         ctl_enqueue_isc(io);
1447                         break;
1448
1449                 /* Performed on the Originating SC, XFER mode only */
1450                 case CTL_MSG_DATAMOVE: {
1451                         struct ctl_sg_entry *sgl;
1452                         int i, j;
1453
1454                         io = msg->hdr.original_sc;
1455                         if (io == NULL) {
1456                                 printf("%s: original_sc == NULL!\n", __func__);
1457                                 /* XXX KDM do something here */
1458                                 break;
1459                         }
1460                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1461                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1462                         /*
1463                          * Keep track of this, we need to send it back over
1464                          * when the datamove is complete.
1465                          */
1466                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1467                         if (msg->hdr.status == CTL_SUCCESS)
1468                                 io->io_hdr.status = msg->hdr.status;
1469
1470                         if (msg->dt.sg_sequence == 0) {
1471 #ifdef CTL_TIME_IO
1472                                 getbinuptime(&io->io_hdr.dma_start_bt);
1473 #endif
1474                                 i = msg->dt.kern_sg_entries +
1475                                     msg->dt.kern_data_len /
1476                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1477                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1478                                     M_WAITOK | M_ZERO);
1479                                 io->io_hdr.remote_sglist = sgl;
1480                                 io->io_hdr.local_sglist =
1481                                     &sgl[msg->dt.kern_sg_entries];
1482
1483                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1484
1485                                 io->scsiio.kern_sg_entries =
1486                                         msg->dt.kern_sg_entries;
1487                                 io->scsiio.rem_sg_entries =
1488                                         msg->dt.kern_sg_entries;
1489                                 io->scsiio.kern_data_len =
1490                                         msg->dt.kern_data_len;
1491                                 io->scsiio.kern_total_len =
1492                                         msg->dt.kern_total_len;
1493                                 io->scsiio.kern_data_resid =
1494                                         msg->dt.kern_data_resid;
1495                                 io->scsiio.kern_rel_offset =
1496                                         msg->dt.kern_rel_offset;
1497                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1498                                 io->io_hdr.flags |= msg->dt.flags &
1499                                     CTL_FLAG_BUS_ADDR;
1500                         } else
1501                                 sgl = (struct ctl_sg_entry *)
1502                                         io->scsiio.kern_data_ptr;
1503
1504                         for (i = msg->dt.sent_sg_entries, j = 0;
1505                              i < (msg->dt.sent_sg_entries +
1506                              msg->dt.cur_sg_entries); i++, j++) {
1507                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1508                                 sgl[i].len = msg->dt.sg_list[j].len;
1509
1510 #if 0
1511                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1512                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1513 #endif
1514                         }
1515
1516                         /*
1517                          * If this is the last piece of the I/O, we've got
1518                          * the full S/G list.  Queue processing in the thread.
1519                          * Otherwise wait for the next piece.
1520                          */
1521                         if (msg->dt.sg_last != 0)
1522                                 ctl_enqueue_isc(io);
1523                         break;
1524                 }
1525                 /* Performed on the Serializing (primary) SC, XFER mode only */
1526                 case CTL_MSG_DATAMOVE_DONE: {
1527                         if (msg->hdr.serializing_sc == NULL) {
1528                                 printf("%s: serializing_sc == NULL!\n",
1529                                        __func__);
1530                                 /* XXX KDM now what? */
1531                                 break;
1532                         }
1533                         /*
1534                          * We grab the sense information here in case
1535                          * there was a failure, so we can return status
1536                          * back to the initiator.
1537                          */
1538                         io = msg->hdr.serializing_sc;
1539                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1540                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1541                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1542                         io->io_hdr.port_status = msg->scsi.port_status;
1543                         io->scsiio.kern_data_resid = msg->scsi.kern_data_resid;
1544                         if (msg->hdr.status != CTL_STATUS_NONE) {
1545                                 io->io_hdr.status = msg->hdr.status;
1546                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1547                                 io->scsiio.sense_len = msg->scsi.sense_len;
1548                                 memcpy(&io->scsiio.sense_data,
1549                                     &msg->scsi.sense_data,
1550                                     msg->scsi.sense_len);
1551                                 if (msg->hdr.status == CTL_SUCCESS)
1552                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1553                         }
1554                         ctl_enqueue_isc(io);
1555                         break;
1556                 }
1557
1558                 /* Preformed on Originating SC, SER_ONLY mode */
1559                 case CTL_MSG_R2R:
1560                         io = msg->hdr.original_sc;
1561                         if (io == NULL) {
1562                                 printf("%s: original_sc == NULL!\n",
1563                                     __func__);
1564                                 break;
1565                         }
1566                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1567                         io->io_hdr.msg_type = CTL_MSG_R2R;
1568                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1569                         ctl_enqueue_isc(io);
1570                         break;
1571
1572                 /*
1573                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1574                  * mode.
1575                  * Performed on the Originating (i.e. secondary) SC in XFER
1576                  * mode
1577                  */
1578                 case CTL_MSG_FINISH_IO:
1579                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1580                                 ctl_isc_handler_finish_xfer(softc, msg);
1581                         else
1582                                 ctl_isc_handler_finish_ser_only(softc, msg);
1583                         break;
1584
1585                 /* Preformed on Originating SC */
1586                 case CTL_MSG_BAD_JUJU:
1587                         io = msg->hdr.original_sc;
1588                         if (io == NULL) {
1589                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1590                                        __func__);
1591                                 break;
1592                         }
1593                         ctl_copy_sense_data(msg, io);
1594                         /*
1595                          * IO should have already been cleaned up on other
1596                          * SC so clear this flag so we won't send a message
1597                          * back to finish the IO there.
1598                          */
1599                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1600                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1601
1602                         /* io = msg->hdr.serializing_sc; */
1603                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1604                         ctl_enqueue_isc(io);
1605                         break;
1606
1607                 /* Handle resets sent from the other side */
1608                 case CTL_MSG_MANAGE_TASKS: {
1609                         struct ctl_taskio *taskio;
1610                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1611                             softc->othersc_pool);
1612                         ctl_zero_io((union ctl_io *)taskio);
1613                         taskio->io_hdr.io_type = CTL_IO_TASK;
1614                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1615                         taskio->io_hdr.nexus = msg->hdr.nexus;
1616                         taskio->task_action = msg->task.task_action;
1617                         taskio->tag_num = msg->task.tag_num;
1618                         taskio->tag_type = msg->task.tag_type;
1619 #ifdef CTL_TIME_IO
1620                         taskio->io_hdr.start_time = time_uptime;
1621                         getbinuptime(&taskio->io_hdr.start_bt);
1622 #endif /* CTL_TIME_IO */
1623                         ctl_run_task((union ctl_io *)taskio);
1624                         break;
1625                 }
1626                 /* Persistent Reserve action which needs attention */
1627                 case CTL_MSG_PERS_ACTION:
1628                         presio = (struct ctl_prio *)ctl_alloc_io(
1629                             softc->othersc_pool);
1630                         ctl_zero_io((union ctl_io *)presio);
1631                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1632                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1633                         presio->io_hdr.nexus = msg->hdr.nexus;
1634                         presio->pr_msg = msg->pr;
1635                         ctl_enqueue_isc((union ctl_io *)presio);
1636                         break;
1637                 case CTL_MSG_UA:
1638                         ctl_isc_ua(softc, msg, param);
1639                         break;
1640                 case CTL_MSG_PORT_SYNC:
1641                         ctl_isc_port_sync(softc, msg, param);
1642                         break;
1643                 case CTL_MSG_LUN_SYNC:
1644                         ctl_isc_lun_sync(softc, msg, param);
1645                         break;
1646                 case CTL_MSG_IID_SYNC:
1647                         ctl_isc_iid_sync(softc, msg, param);
1648                         break;
1649                 case CTL_MSG_LOGIN:
1650                         ctl_isc_login(softc, msg, param);
1651                         break;
1652                 case CTL_MSG_MODE_SYNC:
1653                         ctl_isc_mode_sync(softc, msg, param);
1654                         break;
1655                 default:
1656                         printf("Received HA message of unknown type %d\n",
1657                             msg->hdr.msg_type);
1658                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1659                         break;
1660                 }
1661                 if (msg != &msgbuf)
1662                         free(msg, M_CTL);
1663         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1664                 printf("CTL: HA link status changed from %d to %d\n",
1665                     softc->ha_link, param);
1666                 if (param == softc->ha_link)
1667                         return;
1668                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1669                         softc->ha_link = param;
1670                         ctl_isc_ha_link_down(softc);
1671                 } else {
1672                         softc->ha_link = param;
1673                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1674                                 ctl_isc_ha_link_up(softc);
1675                 }
1676                 return;
1677         } else {
1678                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1679                 return;
1680         }
1681 }
1682
1683 static void
1684 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1685 {
1686
1687         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1688             src->scsi.sense_len);
1689         dest->scsiio.scsi_status = src->scsi.scsi_status;
1690         dest->scsiio.sense_len = src->scsi.sense_len;
1691         dest->io_hdr.status = src->hdr.status;
1692 }
1693
1694 static void
1695 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1696 {
1697
1698         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1699             src->scsiio.sense_len);
1700         dest->scsi.scsi_status = src->scsiio.scsi_status;
1701         dest->scsi.sense_len = src->scsiio.sense_len;
1702         dest->hdr.status = src->io_hdr.status;
1703 }
1704
1705 void
1706 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1707 {
1708         struct ctl_softc *softc = lun->ctl_softc;
1709         ctl_ua_type *pu;
1710
1711         if (initidx < softc->init_min || initidx >= softc->init_max)
1712                 return;
1713         mtx_assert(&lun->lun_lock, MA_OWNED);
1714         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1715         if (pu == NULL)
1716                 return;
1717         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1718 }
1719
1720 void
1721 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1722 {
1723         int i;
1724
1725         mtx_assert(&lun->lun_lock, MA_OWNED);
1726         if (lun->pending_ua[port] == NULL)
1727                 return;
1728         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1729                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1730                         continue;
1731                 lun->pending_ua[port][i] |= ua;
1732         }
1733 }
1734
1735 void
1736 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1737 {
1738         struct ctl_softc *softc = lun->ctl_softc;
1739         int i;
1740
1741         mtx_assert(&lun->lun_lock, MA_OWNED);
1742         for (i = softc->port_min; i < softc->port_max; i++)
1743                 ctl_est_ua_port(lun, i, except, ua);
1744 }
1745
1746 void
1747 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1748 {
1749         struct ctl_softc *softc = lun->ctl_softc;
1750         ctl_ua_type *pu;
1751
1752         if (initidx < softc->init_min || initidx >= softc->init_max)
1753                 return;
1754         mtx_assert(&lun->lun_lock, MA_OWNED);
1755         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1756         if (pu == NULL)
1757                 return;
1758         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1759 }
1760
1761 void
1762 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1763 {
1764         struct ctl_softc *softc = lun->ctl_softc;
1765         int i, j;
1766
1767         mtx_assert(&lun->lun_lock, MA_OWNED);
1768         for (i = softc->port_min; i < softc->port_max; i++) {
1769                 if (lun->pending_ua[i] == NULL)
1770                         continue;
1771                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1772                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1773                                 continue;
1774                         lun->pending_ua[i][j] &= ~ua;
1775                 }
1776         }
1777 }
1778
1779 void
1780 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1781     ctl_ua_type ua_type)
1782 {
1783         struct ctl_lun *lun;
1784
1785         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1786         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1787                 mtx_lock(&lun->lun_lock);
1788                 ctl_clr_ua(lun, initidx, ua_type);
1789                 mtx_unlock(&lun->lun_lock);
1790         }
1791 }
1792
1793 static int
1794 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1795 {
1796         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1797         struct ctl_lun *lun;
1798         struct ctl_lun_req ireq;
1799         int error, value;
1800
1801         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1802         error = sysctl_handle_int(oidp, &value, 0, req);
1803         if ((error != 0) || (req->newptr == NULL))
1804                 return (error);
1805
1806         mtx_lock(&softc->ctl_lock);
1807         if (value == 0)
1808                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1809         else
1810                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1811         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1812                 mtx_unlock(&softc->ctl_lock);
1813                 bzero(&ireq, sizeof(ireq));
1814                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1815                 ireq.reqdata.modify.lun_id = lun->lun;
1816                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1817                     curthread);
1818                 if (ireq.status != CTL_LUN_OK) {
1819                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1820                             __func__, ireq.status, ireq.error_str);
1821                 }
1822                 mtx_lock(&softc->ctl_lock);
1823         }
1824         mtx_unlock(&softc->ctl_lock);
1825         return (0);
1826 }
1827
1828 static int
1829 ctl_init(void)
1830 {
1831         struct make_dev_args args;
1832         struct ctl_softc *softc;
1833         int i, error;
1834
1835         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1836                                M_WAITOK | M_ZERO);
1837
1838         make_dev_args_init(&args);
1839         args.mda_devsw = &ctl_cdevsw;
1840         args.mda_uid = UID_ROOT;
1841         args.mda_gid = GID_OPERATOR;
1842         args.mda_mode = 0600;
1843         args.mda_si_drv1 = softc;
1844         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1845         if (error != 0) {
1846                 free(softc, M_DEVBUF);
1847                 control_softc = NULL;
1848                 return (error);
1849         }
1850
1851         sysctl_ctx_init(&softc->sysctl_ctx);
1852         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1853                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1854                 CTLFLAG_RD, 0, "CAM Target Layer");
1855
1856         if (softc->sysctl_tree == NULL) {
1857                 printf("%s: unable to allocate sysctl tree\n", __func__);
1858                 destroy_dev(softc->dev);
1859                 free(softc, M_DEVBUF);
1860                 control_softc = NULL;
1861                 return (ENOMEM);
1862         }
1863
1864         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1865         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1866             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1867         softc->flags = 0;
1868
1869         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1870             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1871             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1872
1873         /*
1874          * In Copan's HA scheme, the "master" and "slave" roles are
1875          * figured out through the slot the controller is in.  Although it
1876          * is an active/active system, someone has to be in charge.
1877          */
1878         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1879             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1880             "HA head ID (0 - no HA)");
1881         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1882                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1883                 softc->is_single = 1;
1884                 softc->port_cnt = CTL_MAX_PORTS;
1885                 softc->port_min = 0;
1886         } else {
1887                 softc->port_cnt = CTL_MAX_PORTS / NUM_HA_SHELVES;
1888                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1889         }
1890         softc->port_max = softc->port_min + softc->port_cnt;
1891         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1892         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1893
1894         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1895             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1896             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1897
1898         STAILQ_INIT(&softc->lun_list);
1899         STAILQ_INIT(&softc->pending_lun_queue);
1900         STAILQ_INIT(&softc->fe_list);
1901         STAILQ_INIT(&softc->port_list);
1902         STAILQ_INIT(&softc->be_list);
1903         ctl_tpc_init(softc);
1904
1905         if (worker_threads <= 0)
1906                 worker_threads = max(1, mp_ncpus / 4);
1907         if (worker_threads > CTL_MAX_THREADS)
1908                 worker_threads = CTL_MAX_THREADS;
1909
1910         for (i = 0; i < worker_threads; i++) {
1911                 struct ctl_thread *thr = &softc->threads[i];
1912
1913                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1914                 thr->ctl_softc = softc;
1915                 STAILQ_INIT(&thr->incoming_queue);
1916                 STAILQ_INIT(&thr->rtr_queue);
1917                 STAILQ_INIT(&thr->done_queue);
1918                 STAILQ_INIT(&thr->isc_queue);
1919
1920                 error = kproc_kthread_add(ctl_work_thread, thr,
1921                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1922                 if (error != 0) {
1923                         printf("error creating CTL work thread!\n");
1924                         return (error);
1925                 }
1926         }
1927         error = kproc_kthread_add(ctl_lun_thread, softc,
1928             &softc->ctl_proc, &softc->lun_thread, 0, 0, "ctl", "lun");
1929         if (error != 0) {
1930                 printf("error creating CTL lun thread!\n");
1931                 return (error);
1932         }
1933         error = kproc_kthread_add(ctl_thresh_thread, softc,
1934             &softc->ctl_proc, &softc->thresh_thread, 0, 0, "ctl", "thresh");
1935         if (error != 0) {
1936                 printf("error creating CTL threshold thread!\n");
1937                 return (error);
1938         }
1939
1940         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1941             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1942             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1943
1944         if (softc->is_single == 0) {
1945                 if (ctl_frontend_register(&ha_frontend) != 0)
1946                         softc->is_single = 1;
1947         }
1948         return (0);
1949 }
1950
1951 static int
1952 ctl_shutdown(void)
1953 {
1954         struct ctl_softc *softc = control_softc;
1955         int i;
1956
1957         if (softc->is_single == 0)
1958                 ctl_frontend_deregister(&ha_frontend);
1959
1960         destroy_dev(softc->dev);
1961
1962         /* Shutdown CTL threads. */
1963         softc->shutdown = 1;
1964         for (i = 0; i < worker_threads; i++) {
1965                 struct ctl_thread *thr = &softc->threads[i];
1966                 while (thr->thread != NULL) {
1967                         wakeup(thr);
1968                         if (thr->thread != NULL)
1969                                 pause("CTL thr shutdown", 1);
1970                 }
1971                 mtx_destroy(&thr->queue_lock);
1972         }
1973         while (softc->lun_thread != NULL) {
1974                 wakeup(&softc->pending_lun_queue);
1975                 if (softc->lun_thread != NULL)
1976                         pause("CTL thr shutdown", 1);
1977         }
1978         while (softc->thresh_thread != NULL) {
1979                 wakeup(softc->thresh_thread);
1980                 if (softc->thresh_thread != NULL)
1981                         pause("CTL thr shutdown", 1);
1982         }
1983
1984         ctl_tpc_shutdown(softc);
1985         uma_zdestroy(softc->io_zone);
1986         mtx_destroy(&softc->ctl_lock);
1987
1988         sysctl_ctx_free(&softc->sysctl_ctx);
1989
1990         free(softc, M_DEVBUF);
1991         control_softc = NULL;
1992         return (0);
1993 }
1994
1995 static int
1996 ctl_module_event_handler(module_t mod, int what, void *arg)
1997 {
1998
1999         switch (what) {
2000         case MOD_LOAD:
2001                 return (ctl_init());
2002         case MOD_UNLOAD:
2003                 return (ctl_shutdown());
2004         default:
2005                 return (EOPNOTSUPP);
2006         }
2007 }
2008
2009 /*
2010  * XXX KDM should we do some access checks here?  Bump a reference count to
2011  * prevent a CTL module from being unloaded while someone has it open?
2012  */
2013 static int
2014 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
2015 {
2016         return (0);
2017 }
2018
2019 static int
2020 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2021 {
2022         return (0);
2023 }
2024
2025 /*
2026  * Remove an initiator by port number and initiator ID.
2027  * Returns 0 for success, -1 for failure.
2028  */
2029 int
2030 ctl_remove_initiator(struct ctl_port *port, int iid)
2031 {
2032         struct ctl_softc *softc = port->ctl_softc;
2033         int last;
2034
2035         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2036
2037         if (iid > CTL_MAX_INIT_PER_PORT) {
2038                 printf("%s: initiator ID %u > maximun %u!\n",
2039                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2040                 return (-1);
2041         }
2042
2043         mtx_lock(&softc->ctl_lock);
2044         last = (--port->wwpn_iid[iid].in_use == 0);
2045         port->wwpn_iid[iid].last_use = time_uptime;
2046         mtx_unlock(&softc->ctl_lock);
2047         if (last)
2048                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
2049         ctl_isc_announce_iid(port, iid);
2050
2051         return (0);
2052 }
2053
2054 /*
2055  * Add an initiator to the initiator map.
2056  * Returns iid for success, < 0 for failure.
2057  */
2058 int
2059 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2060 {
2061         struct ctl_softc *softc = port->ctl_softc;
2062         time_t best_time;
2063         int i, best;
2064
2065         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2066
2067         if (iid >= CTL_MAX_INIT_PER_PORT) {
2068                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2069                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2070                 free(name, M_CTL);
2071                 return (-1);
2072         }
2073
2074         mtx_lock(&softc->ctl_lock);
2075
2076         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2077                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2078                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2079                                 iid = i;
2080                                 break;
2081                         }
2082                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2083                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2084                                 iid = i;
2085                                 break;
2086                         }
2087                 }
2088         }
2089
2090         if (iid < 0) {
2091                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2092                         if (port->wwpn_iid[i].in_use == 0 &&
2093                             port->wwpn_iid[i].wwpn == 0 &&
2094                             port->wwpn_iid[i].name == NULL) {
2095                                 iid = i;
2096                                 break;
2097                         }
2098                 }
2099         }
2100
2101         if (iid < 0) {
2102                 best = -1;
2103                 best_time = INT32_MAX;
2104                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2105                         if (port->wwpn_iid[i].in_use == 0) {
2106                                 if (port->wwpn_iid[i].last_use < best_time) {
2107                                         best = i;
2108                                         best_time = port->wwpn_iid[i].last_use;
2109                                 }
2110                         }
2111                 }
2112                 iid = best;
2113         }
2114
2115         if (iid < 0) {
2116                 mtx_unlock(&softc->ctl_lock);
2117                 free(name, M_CTL);
2118                 return (-2);
2119         }
2120
2121         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2122                 /*
2123                  * This is not an error yet.
2124                  */
2125                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2126 #if 0
2127                         printf("%s: port %d iid %u WWPN %#jx arrived"
2128                             " again\n", __func__, port->targ_port,
2129                             iid, (uintmax_t)wwpn);
2130 #endif
2131                         goto take;
2132                 }
2133                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2134                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2135 #if 0
2136                         printf("%s: port %d iid %u name '%s' arrived"
2137                             " again\n", __func__, port->targ_port,
2138                             iid, name);
2139 #endif
2140                         goto take;
2141                 }
2142
2143                 /*
2144                  * This is an error, but what do we do about it?  The
2145                  * driver is telling us we have a new WWPN for this
2146                  * initiator ID, so we pretty much need to use it.
2147                  */
2148                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2149                     " but WWPN %#jx '%s' is still at that address\n",
2150                     __func__, port->targ_port, iid, wwpn, name,
2151                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2152                     port->wwpn_iid[iid].name);
2153         }
2154 take:
2155         free(port->wwpn_iid[iid].name, M_CTL);
2156         port->wwpn_iid[iid].name = name;
2157         port->wwpn_iid[iid].wwpn = wwpn;
2158         port->wwpn_iid[iid].in_use++;
2159         mtx_unlock(&softc->ctl_lock);
2160         ctl_isc_announce_iid(port, iid);
2161
2162         return (iid);
2163 }
2164
2165 static int
2166 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2167 {
2168         int len;
2169
2170         switch (port->port_type) {
2171         case CTL_PORT_FC:
2172         {
2173                 struct scsi_transportid_fcp *id =
2174                     (struct scsi_transportid_fcp *)buf;
2175                 if (port->wwpn_iid[iid].wwpn == 0)
2176                         return (0);
2177                 memset(id, 0, sizeof(*id));
2178                 id->format_protocol = SCSI_PROTO_FC;
2179                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2180                 return (sizeof(*id));
2181         }
2182         case CTL_PORT_ISCSI:
2183         {
2184                 struct scsi_transportid_iscsi_port *id =
2185                     (struct scsi_transportid_iscsi_port *)buf;
2186                 if (port->wwpn_iid[iid].name == NULL)
2187                         return (0);
2188                 memset(id, 0, 256);
2189                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2190                     SCSI_PROTO_ISCSI;
2191                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2192                 len = roundup2(min(len, 252), 4);
2193                 scsi_ulto2b(len, id->additional_length);
2194                 return (sizeof(*id) + len);
2195         }
2196         case CTL_PORT_SAS:
2197         {
2198                 struct scsi_transportid_sas *id =
2199                     (struct scsi_transportid_sas *)buf;
2200                 if (port->wwpn_iid[iid].wwpn == 0)
2201                         return (0);
2202                 memset(id, 0, sizeof(*id));
2203                 id->format_protocol = SCSI_PROTO_SAS;
2204                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2205                 return (sizeof(*id));
2206         }
2207         default:
2208         {
2209                 struct scsi_transportid_spi *id =
2210                     (struct scsi_transportid_spi *)buf;
2211                 memset(id, 0, sizeof(*id));
2212                 id->format_protocol = SCSI_PROTO_SPI;
2213                 scsi_ulto2b(iid, id->scsi_addr);
2214                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2215                 return (sizeof(*id));
2216         }
2217         }
2218 }
2219
2220 /*
2221  * Serialize a command that went down the "wrong" side, and so was sent to
2222  * this controller for execution.  The logic is a little different than the
2223  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2224  * sent back to the other side, but in the success case, we execute the
2225  * command on this side (XFER mode) or tell the other side to execute it
2226  * (SER_ONLY mode).
2227  */
2228 static void
2229 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2230 {
2231         struct ctl_softc *softc = CTL_SOFTC(ctsio);
2232         struct ctl_port *port = CTL_PORT(ctsio);
2233         union ctl_ha_msg msg_info;
2234         struct ctl_lun *lun;
2235         const struct ctl_cmd_entry *entry;
2236         uint32_t targ_lun;
2237
2238         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2239
2240         /* Make sure that we know about this port. */
2241         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2242                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2243                                          /*retry_count*/ 1);
2244                 goto badjuju;
2245         }
2246
2247         /* Make sure that we know about this LUN. */
2248         mtx_lock(&softc->ctl_lock);
2249         if (targ_lun >= CTL_MAX_LUNS ||
2250             (lun = softc->ctl_luns[targ_lun]) == NULL) {
2251                 mtx_unlock(&softc->ctl_lock);
2252
2253                 /*
2254                  * The other node would not send this request to us unless
2255                  * received announce that we are primary node for this LUN.
2256                  * If this LUN does not exist now, it is probably result of
2257                  * a race, so respond to initiator in the most opaque way.
2258                  */
2259                 ctl_set_busy(ctsio);
2260                 goto badjuju;
2261         }
2262         mtx_lock(&lun->lun_lock);
2263         mtx_unlock(&softc->ctl_lock);
2264
2265         /*
2266          * If the LUN is invalid, pretend that it doesn't exist.
2267          * It will go away as soon as all pending I/Os completed.
2268          */
2269         if (lun->flags & CTL_LUN_DISABLED) {
2270                 mtx_unlock(&lun->lun_lock);
2271                 ctl_set_busy(ctsio);
2272                 goto badjuju;
2273         }
2274
2275         entry = ctl_get_cmd_entry(ctsio, NULL);
2276         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2277                 mtx_unlock(&lun->lun_lock);
2278                 goto badjuju;
2279         }
2280
2281         CTL_LUN(ctsio) = lun;
2282         CTL_BACKEND_LUN(ctsio) = lun->be_lun;
2283
2284         /*
2285          * Every I/O goes into the OOA queue for a
2286          * particular LUN, and stays there until completion.
2287          */
2288 #ifdef CTL_TIME_IO
2289         if (TAILQ_EMPTY(&lun->ooa_queue))
2290                 lun->idle_time += getsbinuptime() - lun->last_busy;
2291 #endif
2292         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2293
2294         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2295                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2296                  ooa_links))) {
2297         case CTL_ACTION_BLOCK:
2298                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2299                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2300                                   blocked_links);
2301                 mtx_unlock(&lun->lun_lock);
2302                 break;
2303         case CTL_ACTION_PASS:
2304         case CTL_ACTION_SKIP:
2305                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2306                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2307                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2308                         mtx_unlock(&lun->lun_lock);
2309                 } else {
2310                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2311                         mtx_unlock(&lun->lun_lock);
2312
2313                         /* send msg back to other side */
2314                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2315                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2316                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2317                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2318                             sizeof(msg_info.hdr), M_WAITOK);
2319                 }
2320                 break;
2321         case CTL_ACTION_OVERLAP:
2322                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2323                 mtx_unlock(&lun->lun_lock);
2324                 ctl_set_overlapped_cmd(ctsio);
2325                 goto badjuju;
2326         case CTL_ACTION_OVERLAP_TAG:
2327                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2328                 mtx_unlock(&lun->lun_lock);
2329                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2330                 goto badjuju;
2331         case CTL_ACTION_ERROR:
2332         default:
2333                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2334                 mtx_unlock(&lun->lun_lock);
2335
2336                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2337                                          /*retry_count*/ 0);
2338 badjuju:
2339                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2340                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2341                 msg_info.hdr.serializing_sc = NULL;
2342                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2343                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2344                     sizeof(msg_info.scsi), M_WAITOK);
2345                 ctl_free_io((union ctl_io *)ctsio);
2346                 break;
2347         }
2348 }
2349
2350 /*
2351  * Returns 0 for success, errno for failure.
2352  */
2353 static void
2354 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2355                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2356 {
2357         union ctl_io *io;
2358
2359         mtx_lock(&lun->lun_lock);
2360         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2361              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2362              ooa_links)) {
2363                 struct ctl_ooa_entry *entry;
2364
2365                 /*
2366                  * If we've got more than we can fit, just count the
2367                  * remaining entries.
2368                  */
2369                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2370                         continue;
2371
2372                 entry = &kern_entries[*cur_fill_num];
2373
2374                 entry->tag_num = io->scsiio.tag_num;
2375                 entry->lun_num = lun->lun;
2376 #ifdef CTL_TIME_IO
2377                 entry->start_bt = io->io_hdr.start_bt;
2378 #endif
2379                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2380                 entry->cdb_len = io->scsiio.cdb_len;
2381                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2382                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2383
2384                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2385                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2386
2387                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2388                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2389
2390                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2391                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2392
2393                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2394                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2395         }
2396         mtx_unlock(&lun->lun_lock);
2397 }
2398
2399 static void *
2400 ctl_copyin_alloc(void *user_addr, unsigned int len, char *error_str,
2401                  size_t error_str_len)
2402 {
2403         void *kptr;
2404
2405         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2406
2407         if (copyin(user_addr, kptr, len) != 0) {
2408                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2409                          "from user address %p to kernel address %p", len,
2410                          user_addr, kptr);
2411                 free(kptr, M_CTL);
2412                 return (NULL);
2413         }
2414
2415         return (kptr);
2416 }
2417
2418 static void
2419 ctl_free_args(int num_args, struct ctl_be_arg *args)
2420 {
2421         int i;
2422
2423         if (args == NULL)
2424                 return;
2425
2426         for (i = 0; i < num_args; i++) {
2427                 free(args[i].kname, M_CTL);
2428                 free(args[i].kvalue, M_CTL);
2429         }
2430
2431         free(args, M_CTL);
2432 }
2433
2434 static struct ctl_be_arg *
2435 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2436                 char *error_str, size_t error_str_len)
2437 {
2438         struct ctl_be_arg *args;
2439         int i;
2440
2441         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2442                                 error_str, error_str_len);
2443
2444         if (args == NULL)
2445                 goto bailout;
2446
2447         for (i = 0; i < num_args; i++) {
2448                 args[i].kname = NULL;
2449                 args[i].kvalue = NULL;
2450         }
2451
2452         for (i = 0; i < num_args; i++) {
2453                 uint8_t *tmpptr;
2454
2455                 if (args[i].namelen == 0) {
2456                         snprintf(error_str, error_str_len, "Argument %d "
2457                                  "name length is zero", i);
2458                         goto bailout;
2459                 }
2460
2461                 args[i].kname = ctl_copyin_alloc(args[i].name,
2462                         args[i].namelen, error_str, error_str_len);
2463                 if (args[i].kname == NULL)
2464                         goto bailout;
2465
2466                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2467                         snprintf(error_str, error_str_len, "Argument %d "
2468                                  "name is not NUL-terminated", i);
2469                         goto bailout;
2470                 }
2471
2472                 if (args[i].flags & CTL_BEARG_RD) {
2473                         if (args[i].vallen == 0) {
2474                                 snprintf(error_str, error_str_len, "Argument %d "
2475                                          "value length is zero", i);
2476                                 goto bailout;
2477                         }
2478
2479                         tmpptr = ctl_copyin_alloc(args[i].value,
2480                                 args[i].vallen, error_str, error_str_len);
2481                         if (tmpptr == NULL)
2482                                 goto bailout;
2483
2484                         if ((args[i].flags & CTL_BEARG_ASCII)
2485                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2486                                 snprintf(error_str, error_str_len, "Argument "
2487                                     "%d value is not NUL-terminated", i);
2488                                 free(tmpptr, M_CTL);
2489                                 goto bailout;
2490                         }
2491                         args[i].kvalue = tmpptr;
2492                 } else {
2493                         args[i].kvalue = malloc(args[i].vallen,
2494                             M_CTL, M_WAITOK | M_ZERO);
2495                 }
2496         }
2497
2498         return (args);
2499 bailout:
2500
2501         ctl_free_args(num_args, args);
2502
2503         return (NULL);
2504 }
2505
2506 static void
2507 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2508 {
2509         int i;
2510
2511         for (i = 0; i < num_args; i++) {
2512                 if (args[i].flags & CTL_BEARG_WR)
2513                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2514         }
2515 }
2516
2517 /*
2518  * Escape characters that are illegal or not recommended in XML.
2519  */
2520 int
2521 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2522 {
2523         char *end = str + size;
2524         int retval;
2525
2526         retval = 0;
2527
2528         for (; *str && str < end; str++) {
2529                 switch (*str) {
2530                 case '&':
2531                         retval = sbuf_printf(sb, "&amp;");
2532                         break;
2533                 case '>':
2534                         retval = sbuf_printf(sb, "&gt;");
2535                         break;
2536                 case '<':
2537                         retval = sbuf_printf(sb, "&lt;");
2538                         break;
2539                 default:
2540                         retval = sbuf_putc(sb, *str);
2541                         break;
2542                 }
2543
2544                 if (retval != 0)
2545                         break;
2546
2547         }
2548
2549         return (retval);
2550 }
2551
2552 static void
2553 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2554 {
2555         struct scsi_vpd_id_descriptor *desc;
2556         int i;
2557
2558         if (id == NULL || id->len < 4)
2559                 return;
2560         desc = (struct scsi_vpd_id_descriptor *)id->data;
2561         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2562         case SVPD_ID_TYPE_T10:
2563                 sbuf_printf(sb, "t10.");
2564                 break;
2565         case SVPD_ID_TYPE_EUI64:
2566                 sbuf_printf(sb, "eui.");
2567                 break;
2568         case SVPD_ID_TYPE_NAA:
2569                 sbuf_printf(sb, "naa.");
2570                 break;
2571         case SVPD_ID_TYPE_SCSI_NAME:
2572                 break;
2573         }
2574         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2575         case SVPD_ID_CODESET_BINARY:
2576                 for (i = 0; i < desc->length; i++)
2577                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2578                 break;
2579         case SVPD_ID_CODESET_ASCII:
2580                 sbuf_printf(sb, "%.*s", (int)desc->length,
2581                     (char *)desc->identifier);
2582                 break;
2583         case SVPD_ID_CODESET_UTF8:
2584                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2585                 break;
2586         }
2587 }
2588
2589 static int
2590 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2591           struct thread *td)
2592 {
2593         struct ctl_softc *softc = dev->si_drv1;
2594         struct ctl_port *port;
2595         struct ctl_lun *lun;
2596         int retval;
2597
2598         retval = 0;
2599
2600         switch (cmd) {
2601         case CTL_IO:
2602                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2603                 break;
2604         case CTL_ENABLE_PORT:
2605         case CTL_DISABLE_PORT:
2606         case CTL_SET_PORT_WWNS: {
2607                 struct ctl_port *port;
2608                 struct ctl_port_entry *entry;
2609
2610                 entry = (struct ctl_port_entry *)addr;
2611
2612                 mtx_lock(&softc->ctl_lock);
2613                 STAILQ_FOREACH(port, &softc->port_list, links) {
2614                         int action, done;
2615
2616                         if (port->targ_port < softc->port_min ||
2617                             port->targ_port >= softc->port_max)
2618                                 continue;
2619
2620                         action = 0;
2621                         done = 0;
2622                         if ((entry->port_type == CTL_PORT_NONE)
2623                          && (entry->targ_port == port->targ_port)) {
2624                                 /*
2625                                  * If the user only wants to enable or
2626                                  * disable or set WWNs on a specific port,
2627                                  * do the operation and we're done.
2628                                  */
2629                                 action = 1;
2630                                 done = 1;
2631                         } else if (entry->port_type & port->port_type) {
2632                                 /*
2633                                  * Compare the user's type mask with the
2634                                  * particular frontend type to see if we
2635                                  * have a match.
2636                                  */
2637                                 action = 1;
2638                                 done = 0;
2639
2640                                 /*
2641                                  * Make sure the user isn't trying to set
2642                                  * WWNs on multiple ports at the same time.
2643                                  */
2644                                 if (cmd == CTL_SET_PORT_WWNS) {
2645                                         printf("%s: Can't set WWNs on "
2646                                                "multiple ports\n", __func__);
2647                                         retval = EINVAL;
2648                                         break;
2649                                 }
2650                         }
2651                         if (action == 0)
2652                                 continue;
2653
2654                         /*
2655                          * XXX KDM we have to drop the lock here, because
2656                          * the online/offline operations can potentially
2657                          * block.  We need to reference count the frontends
2658                          * so they can't go away,
2659                          */
2660                         if (cmd == CTL_ENABLE_PORT) {
2661                                 mtx_unlock(&softc->ctl_lock);
2662                                 ctl_port_online(port);
2663                                 mtx_lock(&softc->ctl_lock);
2664                         } else if (cmd == CTL_DISABLE_PORT) {
2665                                 mtx_unlock(&softc->ctl_lock);
2666                                 ctl_port_offline(port);
2667                                 mtx_lock(&softc->ctl_lock);
2668                         } else if (cmd == CTL_SET_PORT_WWNS) {
2669                                 ctl_port_set_wwns(port,
2670                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2671                                     1 : 0, entry->wwnn,
2672                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2673                                     1 : 0, entry->wwpn);
2674                         }
2675                         if (done != 0)
2676                                 break;
2677                 }
2678                 mtx_unlock(&softc->ctl_lock);
2679                 break;
2680         }
2681         case CTL_GET_OOA: {
2682                 struct ctl_ooa *ooa_hdr;
2683                 struct ctl_ooa_entry *entries;
2684                 uint32_t cur_fill_num;
2685
2686                 ooa_hdr = (struct ctl_ooa *)addr;
2687
2688                 if ((ooa_hdr->alloc_len == 0)
2689                  || (ooa_hdr->alloc_num == 0)) {
2690                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2691                                "must be non-zero\n", __func__,
2692                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2693                         retval = EINVAL;
2694                         break;
2695                 }
2696
2697                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2698                     sizeof(struct ctl_ooa_entry))) {
2699                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2700                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2701                                __func__, ooa_hdr->alloc_len,
2702                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2703                         retval = EINVAL;
2704                         break;
2705                 }
2706
2707                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2708                 if (entries == NULL) {
2709                         printf("%s: could not allocate %d bytes for OOA "
2710                                "dump\n", __func__, ooa_hdr->alloc_len);
2711                         retval = ENOMEM;
2712                         break;
2713                 }
2714
2715                 mtx_lock(&softc->ctl_lock);
2716                 if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 &&
2717                     (ooa_hdr->lun_num >= CTL_MAX_LUNS ||
2718                      softc->ctl_luns[ooa_hdr->lun_num] == NULL)) {
2719                         mtx_unlock(&softc->ctl_lock);
2720                         free(entries, M_CTL);
2721                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2722                                __func__, (uintmax_t)ooa_hdr->lun_num);
2723                         retval = EINVAL;
2724                         break;
2725                 }
2726
2727                 cur_fill_num = 0;
2728
2729                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2730                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2731                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2732                                     ooa_hdr, entries);
2733                         }
2734                 } else {
2735                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2736                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2737                             entries);
2738                 }
2739                 mtx_unlock(&softc->ctl_lock);
2740
2741                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2742                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2743                         sizeof(struct ctl_ooa_entry);
2744                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2745                 if (retval != 0) {
2746                         printf("%s: error copying out %d bytes for OOA dump\n",
2747                                __func__, ooa_hdr->fill_len);
2748                 }
2749
2750                 getbinuptime(&ooa_hdr->cur_bt);
2751
2752                 if (cur_fill_num > ooa_hdr->alloc_num) {
2753                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2754                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2755                 } else {
2756                         ooa_hdr->dropped_num = 0;
2757                         ooa_hdr->status = CTL_OOA_OK;
2758                 }
2759
2760                 free(entries, M_CTL);
2761                 break;
2762         }
2763         case CTL_DELAY_IO: {
2764                 struct ctl_io_delay_info *delay_info;
2765
2766                 delay_info = (struct ctl_io_delay_info *)addr;
2767
2768 #ifdef CTL_IO_DELAY
2769                 mtx_lock(&softc->ctl_lock);
2770                 if (delay_info->lun_id >= CTL_MAX_LUNS ||
2771                     (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) {
2772                         mtx_unlock(&softc->ctl_lock);
2773                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2774                         break;
2775                 }
2776                 mtx_lock(&lun->lun_lock);
2777                 mtx_unlock(&softc->ctl_lock);
2778                 delay_info->status = CTL_DELAY_STATUS_OK;
2779                 switch (delay_info->delay_type) {
2780                 case CTL_DELAY_TYPE_CONT:
2781                 case CTL_DELAY_TYPE_ONESHOT:
2782                         break;
2783                 default:
2784                         delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE;
2785                         break;
2786                 }
2787                 switch (delay_info->delay_loc) {
2788                 case CTL_DELAY_LOC_DATAMOVE:
2789                         lun->delay_info.datamove_type = delay_info->delay_type;
2790                         lun->delay_info.datamove_delay = delay_info->delay_secs;
2791                         break;
2792                 case CTL_DELAY_LOC_DONE:
2793                         lun->delay_info.done_type = delay_info->delay_type;
2794                         lun->delay_info.done_delay = delay_info->delay_secs;
2795                         break;
2796                 default:
2797                         delay_info->status = CTL_DELAY_STATUS_INVALID_LOC;
2798                         break;
2799                 }
2800                 mtx_unlock(&lun->lun_lock);
2801 #else
2802                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2803 #endif /* CTL_IO_DELAY */
2804                 break;
2805         }
2806 #ifdef CTL_LEGACY_STATS
2807         case CTL_GETSTATS: {
2808                 struct ctl_stats *stats = (struct ctl_stats *)addr;
2809                 int i;
2810
2811                 /*
2812                  * XXX KDM no locking here.  If the LUN list changes,
2813                  * things can blow up.
2814                  */
2815                 i = 0;
2816                 stats->status = CTL_SS_OK;
2817                 stats->fill_len = 0;
2818                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2819                         if (stats->fill_len + sizeof(lun->legacy_stats) >
2820                             stats->alloc_len) {
2821                                 stats->status = CTL_SS_NEED_MORE_SPACE;
2822                                 break;
2823                         }
2824                         retval = copyout(&lun->legacy_stats, &stats->lun_stats[i++],
2825                                          sizeof(lun->legacy_stats));
2826                         if (retval != 0)
2827                                 break;
2828                         stats->fill_len += sizeof(lun->legacy_stats);
2829                 }
2830                 stats->num_luns = softc->num_luns;
2831                 stats->flags = CTL_STATS_FLAG_NONE;
2832 #ifdef CTL_TIME_IO
2833                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
2834 #endif
2835                 getnanouptime(&stats->timestamp);
2836                 break;
2837         }
2838 #endif /* CTL_LEGACY_STATS */
2839         case CTL_ERROR_INJECT: {
2840                 struct ctl_error_desc *err_desc, *new_err_desc;
2841
2842                 err_desc = (struct ctl_error_desc *)addr;
2843
2844                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2845                                       M_WAITOK | M_ZERO);
2846                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2847
2848                 mtx_lock(&softc->ctl_lock);
2849                 if (err_desc->lun_id >= CTL_MAX_LUNS ||
2850                     (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) {
2851                         mtx_unlock(&softc->ctl_lock);
2852                         free(new_err_desc, M_CTL);
2853                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2854                                __func__, (uintmax_t)err_desc->lun_id);
2855                         retval = EINVAL;
2856                         break;
2857                 }
2858                 mtx_lock(&lun->lun_lock);
2859                 mtx_unlock(&softc->ctl_lock);
2860
2861                 /*
2862                  * We could do some checking here to verify the validity
2863                  * of the request, but given the complexity of error
2864                  * injection requests, the checking logic would be fairly
2865                  * complex.
2866                  *
2867                  * For now, if the request is invalid, it just won't get
2868                  * executed and might get deleted.
2869                  */
2870                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2871
2872                 /*
2873                  * XXX KDM check to make sure the serial number is unique,
2874                  * in case we somehow manage to wrap.  That shouldn't
2875                  * happen for a very long time, but it's the right thing to
2876                  * do.
2877                  */
2878                 new_err_desc->serial = lun->error_serial;
2879                 err_desc->serial = lun->error_serial;
2880                 lun->error_serial++;
2881
2882                 mtx_unlock(&lun->lun_lock);
2883                 break;
2884         }
2885         case CTL_ERROR_INJECT_DELETE: {
2886                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2887                 int delete_done;
2888
2889                 delete_desc = (struct ctl_error_desc *)addr;
2890                 delete_done = 0;
2891
2892                 mtx_lock(&softc->ctl_lock);
2893                 if (delete_desc->lun_id >= CTL_MAX_LUNS ||
2894                     (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) {
2895                         mtx_unlock(&softc->ctl_lock);
2896                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2897                                __func__, (uintmax_t)delete_desc->lun_id);
2898                         retval = EINVAL;
2899                         break;
2900                 }
2901                 mtx_lock(&lun->lun_lock);
2902                 mtx_unlock(&softc->ctl_lock);
2903                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2904                         if (desc->serial != delete_desc->serial)
2905                                 continue;
2906
2907                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2908                                       links);
2909                         free(desc, M_CTL);
2910                         delete_done = 1;
2911                 }
2912                 mtx_unlock(&lun->lun_lock);
2913                 if (delete_done == 0) {
2914                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2915                                "error serial %ju on LUN %u\n", __func__,
2916                                delete_desc->serial, delete_desc->lun_id);
2917                         retval = EINVAL;
2918                         break;
2919                 }
2920                 break;
2921         }
2922         case CTL_DUMP_STRUCTS: {
2923                 int j, k;
2924                 struct ctl_port *port;
2925                 struct ctl_frontend *fe;
2926
2927                 mtx_lock(&softc->ctl_lock);
2928                 printf("CTL Persistent Reservation information start:\n");
2929                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2930                         mtx_lock(&lun->lun_lock);
2931                         if ((lun->flags & CTL_LUN_DISABLED) != 0) {
2932                                 mtx_unlock(&lun->lun_lock);
2933                                 continue;
2934                         }
2935
2936                         for (j = 0; j < CTL_MAX_PORTS; j++) {
2937                                 if (lun->pr_keys[j] == NULL)
2938                                         continue;
2939                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2940                                         if (lun->pr_keys[j][k] == 0)
2941                                                 continue;
2942                                         printf("  LUN %ju port %d iid %d key "
2943                                                "%#jx\n", lun->lun, j, k,
2944                                                (uintmax_t)lun->pr_keys[j][k]);
2945                                 }
2946                         }
2947                         mtx_unlock(&lun->lun_lock);
2948                 }
2949                 printf("CTL Persistent Reservation information end\n");
2950                 printf("CTL Ports:\n");
2951                 STAILQ_FOREACH(port, &softc->port_list, links) {
2952                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2953                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2954                                port->frontend->name, port->port_type,
2955                                port->physical_port, port->virtual_port,
2956                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2957                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2958                                 if (port->wwpn_iid[j].in_use == 0 &&
2959                                     port->wwpn_iid[j].wwpn == 0 &&
2960                                     port->wwpn_iid[j].name == NULL)
2961                                         continue;
2962
2963                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2964                                     j, port->wwpn_iid[j].in_use,
2965                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2966                                     port->wwpn_iid[j].name);
2967                         }
2968                 }
2969                 printf("CTL Port information end\n");
2970                 mtx_unlock(&softc->ctl_lock);
2971                 /*
2972                  * XXX KDM calling this without a lock.  We'd likely want
2973                  * to drop the lock before calling the frontend's dump
2974                  * routine anyway.
2975                  */
2976                 printf("CTL Frontends:\n");
2977                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2978                         printf("  Frontend '%s'\n", fe->name);
2979                         if (fe->fe_dump != NULL)
2980                                 fe->fe_dump();
2981                 }
2982                 printf("CTL Frontend information end\n");
2983                 break;
2984         }
2985         case CTL_LUN_REQ: {
2986                 struct ctl_lun_req *lun_req;
2987                 struct ctl_backend_driver *backend;
2988
2989                 lun_req = (struct ctl_lun_req *)addr;
2990
2991                 backend = ctl_backend_find(lun_req->backend);
2992                 if (backend == NULL) {
2993                         lun_req->status = CTL_LUN_ERROR;
2994                         snprintf(lun_req->error_str,
2995                                  sizeof(lun_req->error_str),
2996                                  "Backend \"%s\" not found.",
2997                                  lun_req->backend);
2998                         break;
2999                 }
3000                 if (lun_req->num_be_args > 0) {
3001                         lun_req->kern_be_args = ctl_copyin_args(
3002                                 lun_req->num_be_args,
3003                                 lun_req->be_args,
3004                                 lun_req->error_str,
3005                                 sizeof(lun_req->error_str));
3006                         if (lun_req->kern_be_args == NULL) {
3007                                 lun_req->status = CTL_LUN_ERROR;
3008                                 break;
3009                         }
3010                 }
3011
3012                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3013
3014                 if (lun_req->num_be_args > 0) {
3015                         ctl_copyout_args(lun_req->num_be_args,
3016                                       lun_req->kern_be_args);
3017                         ctl_free_args(lun_req->num_be_args,
3018                                       lun_req->kern_be_args);
3019                 }
3020                 break;
3021         }
3022         case CTL_LUN_LIST: {
3023                 struct sbuf *sb;
3024                 struct ctl_lun_list *list;
3025                 struct ctl_option *opt;
3026
3027                 list = (struct ctl_lun_list *)addr;
3028
3029                 /*
3030                  * Allocate a fixed length sbuf here, based on the length
3031                  * of the user's buffer.  We could allocate an auto-extending
3032                  * buffer, and then tell the user how much larger our
3033                  * amount of data is than his buffer, but that presents
3034                  * some problems:
3035                  *
3036                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3037                  *     we can't hold a lock while calling them with an
3038                  *     auto-extending buffer.
3039                  *
3040                  * 2.  There is not currently a LUN reference counting
3041                  *     mechanism, outside of outstanding transactions on
3042                  *     the LUN's OOA queue.  So a LUN could go away on us
3043                  *     while we're getting the LUN number, backend-specific
3044                  *     information, etc.  Thus, given the way things
3045                  *     currently work, we need to hold the CTL lock while
3046                  *     grabbing LUN information.
3047                  *
3048                  * So, from the user's standpoint, the best thing to do is
3049                  * allocate what he thinks is a reasonable buffer length,
3050                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3051                  * double the buffer length and try again.  (And repeat
3052                  * that until he succeeds.)
3053                  */
3054                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3055                 if (sb == NULL) {
3056                         list->status = CTL_LUN_LIST_ERROR;
3057                         snprintf(list->error_str, sizeof(list->error_str),
3058                                  "Unable to allocate %d bytes for LUN list",
3059                                  list->alloc_len);
3060                         break;
3061                 }
3062
3063                 sbuf_printf(sb, "<ctllunlist>\n");
3064
3065                 mtx_lock(&softc->ctl_lock);
3066                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3067                         mtx_lock(&lun->lun_lock);
3068                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3069                                              (uintmax_t)lun->lun);
3070
3071                         /*
3072                          * Bail out as soon as we see that we've overfilled
3073                          * the buffer.
3074                          */
3075                         if (retval != 0)
3076                                 break;
3077
3078                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3079                                              "</backend_type>\n",
3080                                              (lun->backend == NULL) ?  "none" :
3081                                              lun->backend->name);
3082
3083                         if (retval != 0)
3084                                 break;
3085
3086                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3087                                              lun->be_lun->lun_type);
3088
3089                         if (retval != 0)
3090                                 break;
3091
3092                         if (lun->backend == NULL) {
3093                                 retval = sbuf_printf(sb, "</lun>\n");
3094                                 if (retval != 0)
3095                                         break;
3096                                 continue;
3097                         }
3098
3099                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3100                                              (lun->be_lun->maxlba > 0) ?
3101                                              lun->be_lun->maxlba + 1 : 0);
3102
3103                         if (retval != 0)
3104                                 break;
3105
3106                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3107                                              lun->be_lun->blocksize);
3108
3109                         if (retval != 0)
3110                                 break;
3111
3112                         retval = sbuf_printf(sb, "\t<serial_number>");
3113
3114                         if (retval != 0)
3115                                 break;
3116
3117                         retval = ctl_sbuf_printf_esc(sb,
3118                             lun->be_lun->serial_num,
3119                             sizeof(lun->be_lun->serial_num));
3120
3121                         if (retval != 0)
3122                                 break;
3123
3124                         retval = sbuf_printf(sb, "</serial_number>\n");
3125
3126                         if (retval != 0)
3127                                 break;
3128
3129                         retval = sbuf_printf(sb, "\t<device_id>");
3130
3131                         if (retval != 0)
3132                                 break;
3133
3134                         retval = ctl_sbuf_printf_esc(sb,
3135                             lun->be_lun->device_id,
3136                             sizeof(lun->be_lun->device_id));
3137
3138                         if (retval != 0)
3139                                 break;
3140
3141                         retval = sbuf_printf(sb, "</device_id>\n");
3142
3143                         if (retval != 0)
3144                                 break;
3145
3146                         if (lun->backend->lun_info != NULL) {
3147                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3148                                 if (retval != 0)
3149                                         break;
3150                         }
3151                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3152                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3153                                     opt->name, opt->value, opt->name);
3154                                 if (retval != 0)
3155                                         break;
3156                         }
3157
3158                         retval = sbuf_printf(sb, "</lun>\n");
3159
3160                         if (retval != 0)
3161                                 break;
3162                         mtx_unlock(&lun->lun_lock);
3163                 }
3164                 if (lun != NULL)
3165                         mtx_unlock(&lun->lun_lock);
3166                 mtx_unlock(&softc->ctl_lock);
3167
3168                 if ((retval != 0)
3169                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3170                         retval = 0;
3171                         sbuf_delete(sb);
3172                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3173                         snprintf(list->error_str, sizeof(list->error_str),
3174                                  "Out of space, %d bytes is too small",
3175                                  list->alloc_len);
3176                         break;
3177                 }
3178
3179                 sbuf_finish(sb);
3180
3181                 retval = copyout(sbuf_data(sb), list->lun_xml,
3182                                  sbuf_len(sb) + 1);
3183
3184                 list->fill_len = sbuf_len(sb) + 1;
3185                 list->status = CTL_LUN_LIST_OK;
3186                 sbuf_delete(sb);
3187                 break;
3188         }
3189         case CTL_ISCSI: {
3190                 struct ctl_iscsi *ci;
3191                 struct ctl_frontend *fe;
3192
3193                 ci = (struct ctl_iscsi *)addr;
3194
3195                 fe = ctl_frontend_find("iscsi");
3196                 if (fe == NULL) {
3197                         ci->status = CTL_ISCSI_ERROR;
3198                         snprintf(ci->error_str, sizeof(ci->error_str),
3199                             "Frontend \"iscsi\" not found.");
3200                         break;
3201                 }
3202
3203                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3204                 break;
3205         }
3206         case CTL_PORT_REQ: {
3207                 struct ctl_req *req;
3208                 struct ctl_frontend *fe;
3209
3210                 req = (struct ctl_req *)addr;
3211
3212                 fe = ctl_frontend_find(req->driver);
3213                 if (fe == NULL) {
3214                         req->status = CTL_LUN_ERROR;
3215                         snprintf(req->error_str, sizeof(req->error_str),
3216                             "Frontend \"%s\" not found.", req->driver);
3217                         break;
3218                 }
3219                 if (req->num_args > 0) {
3220                         req->kern_args = ctl_copyin_args(req->num_args,
3221                             req->args, req->error_str, sizeof(req->error_str));
3222                         if (req->kern_args == NULL) {
3223                                 req->status = CTL_LUN_ERROR;
3224                                 break;
3225                         }
3226                 }
3227
3228                 if (fe->ioctl)
3229                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3230                 else
3231                         retval = ENODEV;
3232
3233                 if (req->num_args > 0) {
3234                         ctl_copyout_args(req->num_args, req->kern_args);
3235                         ctl_free_args(req->num_args, req->kern_args);
3236                 }
3237                 break;
3238         }
3239         case CTL_PORT_LIST: {
3240                 struct sbuf *sb;
3241                 struct ctl_port *port;
3242                 struct ctl_lun_list *list;
3243                 struct ctl_option *opt;
3244                 int j;
3245                 uint32_t plun;
3246
3247                 list = (struct ctl_lun_list *)addr;
3248
3249                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3250                 if (sb == NULL) {
3251                         list->status = CTL_LUN_LIST_ERROR;
3252                         snprintf(list->error_str, sizeof(list->error_str),
3253                                  "Unable to allocate %d bytes for LUN list",
3254                                  list->alloc_len);
3255                         break;
3256                 }
3257
3258                 sbuf_printf(sb, "<ctlportlist>\n");
3259
3260                 mtx_lock(&softc->ctl_lock);
3261                 STAILQ_FOREACH(port, &softc->port_list, links) {
3262                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3263                                              (uintmax_t)port->targ_port);
3264
3265                         /*
3266                          * Bail out as soon as we see that we've overfilled
3267                          * the buffer.
3268                          */
3269                         if (retval != 0)
3270                                 break;
3271
3272                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3273                             "</frontend_type>\n", port->frontend->name);
3274                         if (retval != 0)
3275                                 break;
3276
3277                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3278                                              port->port_type);
3279                         if (retval != 0)
3280                                 break;
3281
3282                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3283                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3284                         if (retval != 0)
3285                                 break;
3286
3287                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3288                             port->port_name);
3289                         if (retval != 0)
3290                                 break;
3291
3292                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3293                             port->physical_port);
3294                         if (retval != 0)
3295                                 break;
3296
3297                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3298                             port->virtual_port);
3299                         if (retval != 0)
3300                                 break;
3301
3302                         if (port->target_devid != NULL) {
3303                                 sbuf_printf(sb, "\t<target>");
3304                                 ctl_id_sbuf(port->target_devid, sb);
3305                                 sbuf_printf(sb, "</target>\n");
3306                         }
3307
3308                         if (port->port_devid != NULL) {
3309                                 sbuf_printf(sb, "\t<port>");
3310                                 ctl_id_sbuf(port->port_devid, sb);
3311                                 sbuf_printf(sb, "</port>\n");
3312                         }
3313
3314                         if (port->port_info != NULL) {
3315                                 retval = port->port_info(port->onoff_arg, sb);
3316                                 if (retval != 0)
3317                                         break;
3318                         }
3319                         STAILQ_FOREACH(opt, &port->options, links) {
3320                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3321                                     opt->name, opt->value, opt->name);
3322                                 if (retval != 0)
3323                                         break;
3324                         }
3325
3326                         if (port->lun_map != NULL) {
3327                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3328                                 for (j = 0; j < port->lun_map_size; j++) {
3329                                         plun = ctl_lun_map_from_port(port, j);
3330                                         if (plun == UINT32_MAX)
3331                                                 continue;
3332                                         sbuf_printf(sb,
3333                                             "\t<lun id=\"%u\">%u</lun>\n",
3334                                             j, plun);
3335                                 }
3336                         }
3337
3338                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3339                                 if (port->wwpn_iid[j].in_use == 0 ||
3340                                     (port->wwpn_iid[j].wwpn == 0 &&
3341                                      port->wwpn_iid[j].name == NULL))
3342                                         continue;
3343
3344                                 if (port->wwpn_iid[j].name != NULL)
3345                                         retval = sbuf_printf(sb,
3346                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3347                                             j, port->wwpn_iid[j].name);
3348                                 else
3349                                         retval = sbuf_printf(sb,
3350                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3351                                             j, port->wwpn_iid[j].wwpn);
3352                                 if (retval != 0)
3353                                         break;
3354                         }
3355                         if (retval != 0)
3356                                 break;
3357
3358                         retval = sbuf_printf(sb, "</targ_port>\n");
3359                         if (retval != 0)
3360                                 break;
3361                 }
3362                 mtx_unlock(&softc->ctl_lock);
3363
3364                 if ((retval != 0)
3365                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3366                         retval = 0;
3367                         sbuf_delete(sb);
3368                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3369                         snprintf(list->error_str, sizeof(list->error_str),
3370                                  "Out of space, %d bytes is too small",
3371                                  list->alloc_len);
3372                         break;
3373                 }
3374
3375                 sbuf_finish(sb);
3376
3377                 retval = copyout(sbuf_data(sb), list->lun_xml,
3378                                  sbuf_len(sb) + 1);
3379
3380                 list->fill_len = sbuf_len(sb) + 1;
3381                 list->status = CTL_LUN_LIST_OK;
3382                 sbuf_delete(sb);
3383                 break;
3384         }
3385         case CTL_LUN_MAP: {
3386                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3387                 struct ctl_port *port;
3388
3389                 mtx_lock(&softc->ctl_lock);
3390                 if (lm->port < softc->port_min ||
3391                     lm->port >= softc->port_max ||
3392                     (port = softc->ctl_ports[lm->port]) == NULL) {
3393                         mtx_unlock(&softc->ctl_lock);
3394                         return (ENXIO);
3395                 }
3396                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3397                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3398                                 if (ctl_lun_map_to_port(port, lun->lun) ==
3399                                     UINT32_MAX)
3400                                         continue;
3401                                 mtx_lock(&lun->lun_lock);
3402                                 ctl_est_ua_port(lun, lm->port, -1,
3403                                     CTL_UA_LUN_CHANGE);
3404                                 mtx_unlock(&lun->lun_lock);
3405                         }
3406                 }
3407                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3408                 if (lm->plun != UINT32_MAX) {
3409                         if (lm->lun == UINT32_MAX)
3410                                 retval = ctl_lun_map_unset(port, lm->plun);
3411                         else if (lm->lun < CTL_MAX_LUNS &&
3412                             softc->ctl_luns[lm->lun] != NULL)
3413                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3414                         else
3415                                 return (ENXIO);
3416                 } else {
3417                         if (lm->lun == UINT32_MAX)
3418                                 retval = ctl_lun_map_deinit(port);
3419                         else
3420                                 retval = ctl_lun_map_init(port);
3421                 }
3422                 if (port->status & CTL_PORT_STATUS_ONLINE)
3423                         ctl_isc_announce_port(port);
3424                 break;
3425         }
3426         case CTL_GET_LUN_STATS: {
3427                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3428                 int i;
3429
3430                 /*
3431                  * XXX KDM no locking here.  If the LUN list changes,
3432                  * things can blow up.
3433                  */
3434                 i = 0;
3435                 stats->status = CTL_SS_OK;
3436                 stats->fill_len = 0;
3437                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3438                         if (lun->lun < stats->first_item)
3439                                 continue;
3440                         if (stats->fill_len + sizeof(lun->stats) >
3441                             stats->alloc_len) {
3442                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3443                                 break;
3444                         }
3445                         retval = copyout(&lun->stats, &stats->stats[i++],
3446                                          sizeof(lun->stats));
3447                         if (retval != 0)
3448                                 break;
3449                         stats->fill_len += sizeof(lun->stats);
3450                 }
3451                 stats->num_items = softc->num_luns;
3452                 stats->flags = CTL_STATS_FLAG_NONE;
3453 #ifdef CTL_TIME_IO
3454                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3455 #endif
3456                 getnanouptime(&stats->timestamp);
3457                 break;
3458         }
3459         case CTL_GET_PORT_STATS: {
3460                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3461                 int i;
3462
3463                 /*
3464                  * XXX KDM no locking here.  If the LUN list changes,
3465                  * things can blow up.
3466                  */
3467                 i = 0;
3468                 stats->status = CTL_SS_OK;
3469                 stats->fill_len = 0;
3470                 STAILQ_FOREACH(port, &softc->port_list, links) {
3471                         if (port->targ_port < stats->first_item)
3472                                 continue;
3473                         if (stats->fill_len + sizeof(port->stats) >
3474                             stats->alloc_len) {
3475                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3476                                 break;
3477                         }
3478                         retval = copyout(&port->stats, &stats->stats[i++],
3479                                          sizeof(port->stats));
3480                         if (retval != 0)
3481                                 break;
3482                         stats->fill_len += sizeof(port->stats);
3483                 }
3484                 stats->num_items = softc->num_ports;
3485                 stats->flags = CTL_STATS_FLAG_NONE;
3486 #ifdef CTL_TIME_IO
3487                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3488 #endif
3489                 getnanouptime(&stats->timestamp);
3490                 break;
3491         }
3492         default: {
3493                 /* XXX KDM should we fix this? */
3494 #if 0
3495                 struct ctl_backend_driver *backend;
3496                 unsigned int type;
3497                 int found;
3498
3499                 found = 0;
3500
3501                 /*
3502                  * We encode the backend type as the ioctl type for backend
3503                  * ioctls.  So parse it out here, and then search for a
3504                  * backend of this type.
3505                  */
3506                 type = _IOC_TYPE(cmd);
3507
3508                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3509                         if (backend->type == type) {
3510                                 found = 1;
3511                                 break;
3512                         }
3513                 }
3514                 if (found == 0) {
3515                         printf("ctl: unknown ioctl command %#lx or backend "
3516                                "%d\n", cmd, type);
3517                         retval = EINVAL;
3518                         break;
3519                 }
3520                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3521 #endif
3522                 retval = ENOTTY;
3523                 break;
3524         }
3525         }
3526         return (retval);
3527 }
3528
3529 uint32_t
3530 ctl_get_initindex(struct ctl_nexus *nexus)
3531 {
3532         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3533 }
3534
3535 int
3536 ctl_lun_map_init(struct ctl_port *port)
3537 {
3538         struct ctl_softc *softc = port->ctl_softc;
3539         struct ctl_lun *lun;
3540         int size = ctl_lun_map_size;
3541         uint32_t i;
3542
3543         if (port->lun_map == NULL || port->lun_map_size < size) {
3544                 port->lun_map_size = 0;
3545                 free(port->lun_map, M_CTL);
3546                 port->lun_map = malloc(size * sizeof(uint32_t),
3547                     M_CTL, M_NOWAIT);
3548         }
3549         if (port->lun_map == NULL)
3550                 return (ENOMEM);
3551         for (i = 0; i < size; i++)
3552                 port->lun_map[i] = UINT32_MAX;
3553         port->lun_map_size = size;
3554         if (port->status & CTL_PORT_STATUS_ONLINE) {
3555                 if (port->lun_disable != NULL) {
3556                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3557                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3558                 }
3559                 ctl_isc_announce_port(port);
3560         }
3561         return (0);
3562 }
3563
3564 int
3565 ctl_lun_map_deinit(struct ctl_port *port)
3566 {
3567         struct ctl_softc *softc = port->ctl_softc;
3568         struct ctl_lun *lun;
3569
3570         if (port->lun_map == NULL)
3571                 return (0);
3572         port->lun_map_size = 0;
3573         free(port->lun_map, M_CTL);
3574         port->lun_map = NULL;
3575         if (port->status & CTL_PORT_STATUS_ONLINE) {
3576                 if (port->lun_enable != NULL) {
3577                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3578                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3579                 }
3580                 ctl_isc_announce_port(port);
3581         }
3582         return (0);
3583 }
3584
3585 int
3586 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3587 {
3588         int status;
3589         uint32_t old;
3590
3591         if (port->lun_map == NULL) {
3592                 status = ctl_lun_map_init(port);
3593                 if (status != 0)
3594                         return (status);
3595         }
3596         if (plun >= port->lun_map_size)
3597                 return (EINVAL);
3598         old = port->lun_map[plun];
3599         port->lun_map[plun] = glun;
3600         if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) {
3601                 if (port->lun_enable != NULL)
3602                         port->lun_enable(port->targ_lun_arg, plun);
3603                 ctl_isc_announce_port(port);
3604         }
3605         return (0);
3606 }
3607
3608 int
3609 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3610 {
3611         uint32_t old;
3612
3613         if (port->lun_map == NULL || plun >= port->lun_map_size)
3614                 return (0);
3615         old = port->lun_map[plun];
3616         port->lun_map[plun] = UINT32_MAX;
3617         if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) {
3618                 if (port->lun_disable != NULL)
3619                         port->lun_disable(port->targ_lun_arg, plun);
3620                 ctl_isc_announce_port(port);
3621         }
3622         return (0);
3623 }
3624
3625 uint32_t
3626 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3627 {
3628
3629         if (port == NULL)
3630                 return (UINT32_MAX);
3631         if (port->lun_map == NULL)
3632                 return (lun_id);
3633         if (lun_id > port->lun_map_size)
3634                 return (UINT32_MAX);
3635         return (port->lun_map[lun_id]);
3636 }
3637
3638 uint32_t
3639 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3640 {
3641         uint32_t i;
3642
3643         if (port == NULL)
3644                 return (UINT32_MAX);
3645         if (port->lun_map == NULL)
3646                 return (lun_id);
3647         for (i = 0; i < port->lun_map_size; i++) {
3648                 if (port->lun_map[i] == lun_id)
3649                         return (i);
3650         }
3651         return (UINT32_MAX);
3652 }
3653
3654 uint32_t
3655 ctl_decode_lun(uint64_t encoded)
3656 {
3657         uint8_t lun[8];
3658         uint32_t result = 0xffffffff;
3659
3660         be64enc(lun, encoded);
3661         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3662         case RPL_LUNDATA_ATYP_PERIPH:
3663                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3664                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3665                         result = lun[1];
3666                 break;
3667         case RPL_LUNDATA_ATYP_FLAT:
3668                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3669                     lun[6] == 0 && lun[7] == 0)
3670                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3671                 break;
3672         case RPL_LUNDATA_ATYP_EXTLUN:
3673                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3674                 case 0x02:
3675                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3676                         case 0x00:
3677                                 result = lun[1];
3678                                 break;
3679                         case 0x10:
3680                                 result = (lun[1] << 16) + (lun[2] << 8) +
3681                                     lun[3];
3682                                 break;
3683                         case 0x20:
3684                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3685                                         result = (lun[2] << 24) +
3686                                             (lun[3] << 16) + (lun[4] << 8) +
3687                                             lun[5];
3688                                 break;
3689                         }
3690                         break;
3691                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3692                         result = 0xffffffff;
3693                         break;
3694                 }
3695                 break;
3696         }
3697         return (result);
3698 }
3699
3700 uint64_t
3701 ctl_encode_lun(uint32_t decoded)
3702 {
3703         uint64_t l = decoded;
3704
3705         if (l <= 0xff)
3706                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3707         if (l <= 0x3fff)
3708                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3709         if (l <= 0xffffff)
3710                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3711                     (l << 32));
3712         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3713 }
3714
3715 int
3716 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3717 {
3718         int i;
3719
3720         for (i = first; i < last; i++) {
3721                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3722                         return (i);
3723         }
3724         return (-1);
3725 }
3726
3727 int
3728 ctl_set_mask(uint32_t *mask, uint32_t bit)
3729 {
3730         uint32_t chunk, piece;
3731
3732         chunk = bit >> 5;
3733         piece = bit % (sizeof(uint32_t) * 8);
3734
3735         if ((mask[chunk] & (1 << piece)) != 0)
3736                 return (-1);
3737         else
3738                 mask[chunk] |= (1 << piece);
3739
3740         return (0);
3741 }
3742
3743 int
3744 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3745 {
3746         uint32_t chunk, piece;
3747
3748         chunk = bit >> 5;
3749         piece = bit % (sizeof(uint32_t) * 8);
3750
3751         if ((mask[chunk] & (1 << piece)) == 0)
3752                 return (-1);
3753         else
3754                 mask[chunk] &= ~(1 << piece);
3755
3756         return (0);
3757 }
3758
3759 int
3760 ctl_is_set(uint32_t *mask, uint32_t bit)
3761 {
3762         uint32_t chunk, piece;
3763
3764         chunk = bit >> 5;
3765         piece = bit % (sizeof(uint32_t) * 8);
3766
3767         if ((mask[chunk] & (1 << piece)) == 0)
3768                 return (0);
3769         else
3770                 return (1);
3771 }
3772
3773 static uint64_t
3774 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3775 {
3776         uint64_t *t;
3777
3778         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3779         if (t == NULL)
3780                 return (0);
3781         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3782 }
3783
3784 static void
3785 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3786 {
3787         uint64_t *t;
3788
3789         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3790         if (t == NULL)
3791                 return;
3792         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3793 }
3794
3795 static void
3796 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3797 {
3798         uint64_t *p;
3799         u_int i;
3800
3801         i = residx/CTL_MAX_INIT_PER_PORT;
3802         if (lun->pr_keys[i] != NULL)
3803                 return;
3804         mtx_unlock(&lun->lun_lock);
3805         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3806             M_WAITOK | M_ZERO);
3807         mtx_lock(&lun->lun_lock);
3808         if (lun->pr_keys[i] == NULL)
3809                 lun->pr_keys[i] = p;
3810         else
3811                 free(p, M_CTL);
3812 }
3813
3814 static void
3815 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3816 {
3817         uint64_t *t;
3818
3819         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3820         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3821         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3822 }
3823
3824 /*
3825  * ctl_softc, pool_name, total_ctl_io are passed in.
3826  * npool is passed out.
3827  */
3828 int
3829 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3830                 uint32_t total_ctl_io, void **npool)
3831 {
3832         struct ctl_io_pool *pool;
3833
3834         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3835                                             M_NOWAIT | M_ZERO);
3836         if (pool == NULL)
3837                 return (ENOMEM);
3838
3839         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3840         pool->ctl_softc = ctl_softc;
3841 #ifdef IO_POOLS
3842         pool->zone = uma_zsecond_create(pool->name, NULL,
3843             NULL, NULL, NULL, ctl_softc->io_zone);
3844         /* uma_prealloc(pool->zone, total_ctl_io); */
3845 #else
3846         pool->zone = ctl_softc->io_zone;
3847 #endif
3848
3849         *npool = pool;
3850         return (0);
3851 }
3852
3853 void
3854 ctl_pool_free(struct ctl_io_pool *pool)
3855 {
3856
3857         if (pool == NULL)
3858                 return;
3859
3860 #ifdef IO_POOLS
3861         uma_zdestroy(pool->zone);
3862 #endif
3863         free(pool, M_CTL);
3864 }
3865
3866 union ctl_io *
3867 ctl_alloc_io(void *pool_ref)
3868 {
3869         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3870         union ctl_io *io;
3871
3872         io = uma_zalloc(pool->zone, M_WAITOK);
3873         if (io != NULL) {
3874                 io->io_hdr.pool = pool_ref;
3875                 CTL_SOFTC(io) = pool->ctl_softc;
3876         }
3877         return (io);
3878 }
3879
3880 union ctl_io *
3881 ctl_alloc_io_nowait(void *pool_ref)
3882 {
3883         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3884         union ctl_io *io;
3885
3886         io = uma_zalloc(pool->zone, M_NOWAIT);
3887         if (io != NULL) {
3888                 io->io_hdr.pool = pool_ref;
3889                 CTL_SOFTC(io) = pool->ctl_softc;
3890         }
3891         return (io);
3892 }
3893
3894 void
3895 ctl_free_io(union ctl_io *io)
3896 {
3897         struct ctl_io_pool *pool;
3898
3899         if (io == NULL)
3900                 return;
3901
3902         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3903         uma_zfree(pool->zone, io);
3904 }
3905
3906 void
3907 ctl_zero_io(union ctl_io *io)
3908 {
3909         struct ctl_io_pool *pool;
3910
3911         if (io == NULL)
3912                 return;
3913
3914         /*
3915          * May need to preserve linked list pointers at some point too.
3916          */
3917         pool = io->io_hdr.pool;
3918         memset(io, 0, sizeof(*io));
3919         io->io_hdr.pool = pool;
3920         CTL_SOFTC(io) = pool->ctl_softc;
3921 }
3922
3923 int
3924 ctl_expand_number(const char *buf, uint64_t *num)
3925 {
3926         char *endptr;
3927         uint64_t number;
3928         unsigned shift;
3929
3930         number = strtoq(buf, &endptr, 0);
3931
3932         switch (tolower((unsigned char)*endptr)) {
3933         case 'e':
3934                 shift = 60;
3935                 break;
3936         case 'p':
3937                 shift = 50;
3938                 break;
3939         case 't':
3940                 shift = 40;
3941                 break;
3942         case 'g':
3943                 shift = 30;
3944                 break;
3945         case 'm':
3946                 shift = 20;
3947                 break;
3948         case 'k':
3949                 shift = 10;
3950                 break;
3951         case 'b':
3952         case '\0': /* No unit. */
3953                 *num = number;
3954                 return (0);
3955         default:
3956                 /* Unrecognized unit. */
3957                 return (-1);
3958         }
3959
3960         if ((number << shift) >> shift != number) {
3961                 /* Overflow */
3962                 return (-1);
3963         }
3964         *num = number << shift;
3965         return (0);
3966 }
3967
3968
3969 /*
3970  * This routine could be used in the future to load default and/or saved
3971  * mode page parameters for a particuar lun.
3972  */
3973 static int
3974 ctl_init_page_index(struct ctl_lun *lun)
3975 {
3976         int i, page_code;
3977         struct ctl_page_index *page_index;
3978         const char *value;
3979         uint64_t ival;
3980
3981         memcpy(&lun->mode_pages.index, page_index_template,
3982                sizeof(page_index_template));
3983
3984         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3985
3986                 page_index = &lun->mode_pages.index[i];
3987                 if (lun->be_lun->lun_type == T_DIRECT &&
3988                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
3989                         continue;
3990                 if (lun->be_lun->lun_type == T_PROCESSOR &&
3991                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
3992                         continue;
3993                 if (lun->be_lun->lun_type == T_CDROM &&
3994                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
3995                         continue;
3996
3997                 page_code = page_index->page_code & SMPH_PC_MASK;
3998                 switch (page_code) {
3999                 case SMS_RW_ERROR_RECOVERY_PAGE: {
4000                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4001                             ("subpage %#x for page %#x is incorrect!",
4002                             page_index->subpage, page_code));
4003                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4004                                &rw_er_page_default,
4005                                sizeof(rw_er_page_default));
4006                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4007                                &rw_er_page_changeable,
4008                                sizeof(rw_er_page_changeable));
4009                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4010                                &rw_er_page_default,
4011                                sizeof(rw_er_page_default));
4012                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4013                                &rw_er_page_default,
4014                                sizeof(rw_er_page_default));
4015                         page_index->page_data =
4016                                 (uint8_t *)lun->mode_pages.rw_er_page;
4017                         break;
4018                 }
4019                 case SMS_FORMAT_DEVICE_PAGE: {
4020                         struct scsi_format_page *format_page;
4021
4022                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4023                             ("subpage %#x for page %#x is incorrect!",
4024                             page_index->subpage, page_code));
4025
4026                         /*
4027                          * Sectors per track are set above.  Bytes per
4028                          * sector need to be set here on a per-LUN basis.
4029                          */
4030                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4031                                &format_page_default,
4032                                sizeof(format_page_default));
4033                         memcpy(&lun->mode_pages.format_page[
4034                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4035                                sizeof(format_page_changeable));
4036                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4037                                &format_page_default,
4038                                sizeof(format_page_default));
4039                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4040                                &format_page_default,
4041                                sizeof(format_page_default));
4042
4043                         format_page = &lun->mode_pages.format_page[
4044                                 CTL_PAGE_CURRENT];
4045                         scsi_ulto2b(lun->be_lun->blocksize,
4046                                     format_page->bytes_per_sector);
4047
4048                         format_page = &lun->mode_pages.format_page[
4049                                 CTL_PAGE_DEFAULT];
4050                         scsi_ulto2b(lun->be_lun->blocksize,
4051                                     format_page->bytes_per_sector);
4052
4053                         format_page = &lun->mode_pages.format_page[
4054                                 CTL_PAGE_SAVED];
4055                         scsi_ulto2b(lun->be_lun->blocksize,
4056                                     format_page->bytes_per_sector);
4057
4058                         page_index->page_data =
4059                                 (uint8_t *)lun->mode_pages.format_page;
4060                         break;
4061                 }
4062                 case SMS_RIGID_DISK_PAGE: {
4063                         struct scsi_rigid_disk_page *rigid_disk_page;
4064                         uint32_t sectors_per_cylinder;
4065                         uint64_t cylinders;
4066 #ifndef __XSCALE__
4067                         int shift;
4068 #endif /* !__XSCALE__ */
4069
4070                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4071                             ("subpage %#x for page %#x is incorrect!",
4072                             page_index->subpage, page_code));
4073
4074                         /*
4075                          * Rotation rate and sectors per track are set
4076                          * above.  We calculate the cylinders here based on
4077                          * capacity.  Due to the number of heads and
4078                          * sectors per track we're using, smaller arrays
4079                          * may turn out to have 0 cylinders.  Linux and
4080                          * FreeBSD don't pay attention to these mode pages
4081                          * to figure out capacity, but Solaris does.  It
4082                          * seems to deal with 0 cylinders just fine, and
4083                          * works out a fake geometry based on the capacity.
4084                          */
4085                         memcpy(&lun->mode_pages.rigid_disk_page[
4086                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4087                                sizeof(rigid_disk_page_default));
4088                         memcpy(&lun->mode_pages.rigid_disk_page[
4089                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4090                                sizeof(rigid_disk_page_changeable));
4091
4092                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4093                                 CTL_DEFAULT_HEADS;
4094
4095                         /*
4096                          * The divide method here will be more accurate,
4097                          * probably, but results in floating point being
4098                          * used in the kernel on i386 (__udivdi3()).  On the
4099                          * XScale, though, __udivdi3() is implemented in
4100                          * software.
4101                          *
4102                          * The shift method for cylinder calculation is
4103                          * accurate if sectors_per_cylinder is a power of
4104                          * 2.  Otherwise it might be slightly off -- you
4105                          * might have a bit of a truncation problem.
4106                          */
4107 #ifdef  __XSCALE__
4108                         cylinders = (lun->be_lun->maxlba + 1) /
4109                                 sectors_per_cylinder;
4110 #else
4111                         for (shift = 31; shift > 0; shift--) {
4112                                 if (sectors_per_cylinder & (1 << shift))
4113                                         break;
4114                         }
4115                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4116 #endif
4117
4118                         /*
4119                          * We've basically got 3 bytes, or 24 bits for the
4120                          * cylinder size in the mode page.  If we're over,
4121                          * just round down to 2^24.
4122                          */
4123                         if (cylinders > 0xffffff)
4124                                 cylinders = 0xffffff;
4125
4126                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4127                                 CTL_PAGE_DEFAULT];
4128                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4129
4130                         if ((value = ctl_get_opt(&lun->be_lun->options,
4131                             "rpm")) != NULL) {
4132                                 scsi_ulto2b(strtol(value, NULL, 0),
4133                                      rigid_disk_page->rotation_rate);
4134                         }
4135
4136                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4137                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4138                                sizeof(rigid_disk_page_default));
4139                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4140                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4141                                sizeof(rigid_disk_page_default));
4142
4143                         page_index->page_data =
4144                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4145                         break;
4146                 }
4147                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4148                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4149                             ("subpage %#x for page %#x is incorrect!",
4150                             page_index->subpage, page_code));
4151                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4152                                &verify_er_page_default,
4153                                sizeof(verify_er_page_default));
4154                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4155                                &verify_er_page_changeable,
4156                                sizeof(verify_er_page_changeable));
4157                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4158                                &verify_er_page_default,
4159                                sizeof(verify_er_page_default));
4160                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4161                                &verify_er_page_default,
4162                                sizeof(verify_er_page_default));
4163                         page_index->page_data =
4164                                 (uint8_t *)lun->mode_pages.verify_er_page;
4165                         break;
4166                 }
4167                 case SMS_CACHING_PAGE: {
4168                         struct scsi_caching_page *caching_page;
4169
4170                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4171                             ("subpage %#x for page %#x is incorrect!",
4172                             page_index->subpage, page_code));
4173                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4174                                &caching_page_default,
4175                                sizeof(caching_page_default));
4176                         memcpy(&lun->mode_pages.caching_page[
4177                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4178                                sizeof(caching_page_changeable));
4179                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4180                                &caching_page_default,
4181                                sizeof(caching_page_default));
4182                         caching_page = &lun->mode_pages.caching_page[
4183                             CTL_PAGE_SAVED];
4184                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
4185                         if (value != NULL && strcmp(value, "off") == 0)
4186                                 caching_page->flags1 &= ~SCP_WCE;
4187                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4188                         if (value != NULL && strcmp(value, "off") == 0)
4189                                 caching_page->flags1 |= SCP_RCD;
4190                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4191                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4192                                sizeof(caching_page_default));
4193                         page_index->page_data =
4194                                 (uint8_t *)lun->mode_pages.caching_page;
4195                         break;
4196                 }
4197                 case SMS_CONTROL_MODE_PAGE: {
4198                         switch (page_index->subpage) {
4199                         case SMS_SUBPAGE_PAGE_0: {
4200                                 struct scsi_control_page *control_page;
4201
4202                                 memcpy(&lun->mode_pages.control_page[
4203                                     CTL_PAGE_DEFAULT],
4204                                        &control_page_default,
4205                                        sizeof(control_page_default));
4206                                 memcpy(&lun->mode_pages.control_page[
4207                                     CTL_PAGE_CHANGEABLE],
4208                                        &control_page_changeable,
4209                                        sizeof(control_page_changeable));
4210                                 memcpy(&lun->mode_pages.control_page[
4211                                     CTL_PAGE_SAVED],
4212                                        &control_page_default,
4213                                        sizeof(control_page_default));
4214                                 control_page = &lun->mode_pages.control_page[
4215                                     CTL_PAGE_SAVED];
4216                                 value = ctl_get_opt(&lun->be_lun->options,
4217                                     "reordering");
4218                                 if (value != NULL &&
4219                                     strcmp(value, "unrestricted") == 0) {
4220                                         control_page->queue_flags &=
4221                                             ~SCP_QUEUE_ALG_MASK;
4222                                         control_page->queue_flags |=
4223                                             SCP_QUEUE_ALG_UNRESTRICTED;
4224                                 }
4225                                 memcpy(&lun->mode_pages.control_page[
4226                                     CTL_PAGE_CURRENT],
4227                                        &lun->mode_pages.control_page[
4228                                     CTL_PAGE_SAVED],
4229                                        sizeof(control_page_default));
4230                                 page_index->page_data =
4231                                     (uint8_t *)lun->mode_pages.control_page;
4232                                 break;
4233                         }
4234                         case 0x01:
4235                                 memcpy(&lun->mode_pages.control_ext_page[
4236                                     CTL_PAGE_DEFAULT],
4237                                        &control_ext_page_default,
4238                                        sizeof(control_ext_page_default));
4239                                 memcpy(&lun->mode_pages.control_ext_page[
4240                                     CTL_PAGE_CHANGEABLE],
4241                                        &control_ext_page_changeable,
4242                                        sizeof(control_ext_page_changeable));
4243                                 memcpy(&lun->mode_pages.control_ext_page[
4244                                     CTL_PAGE_SAVED],
4245                                        &control_ext_page_default,
4246                                        sizeof(control_ext_page_default));
4247                                 memcpy(&lun->mode_pages.control_ext_page[
4248                                     CTL_PAGE_CURRENT],
4249                                        &lun->mode_pages.control_ext_page[
4250                                     CTL_PAGE_SAVED],
4251                                        sizeof(control_ext_page_default));
4252                                 page_index->page_data =
4253                                     (uint8_t *)lun->mode_pages.control_ext_page;
4254                                 break;
4255                         default:
4256                                 panic("subpage %#x for page %#x is incorrect!",
4257                                       page_index->subpage, page_code);
4258                         }
4259                         break;
4260                 }
4261                 case SMS_INFO_EXCEPTIONS_PAGE: {
4262                         switch (page_index->subpage) {
4263                         case SMS_SUBPAGE_PAGE_0:
4264                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4265                                        &ie_page_default,
4266                                        sizeof(ie_page_default));
4267                                 memcpy(&lun->mode_pages.ie_page[
4268                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4269                                        sizeof(ie_page_changeable));
4270                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4271                                        &ie_page_default,
4272                                        sizeof(ie_page_default));
4273                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4274                                        &ie_page_default,
4275                                        sizeof(ie_page_default));
4276                                 page_index->page_data =
4277                                         (uint8_t *)lun->mode_pages.ie_page;
4278                                 break;
4279                         case 0x02: {
4280                                 struct ctl_logical_block_provisioning_page *page;
4281
4282                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4283                                        &lbp_page_default,
4284                                        sizeof(lbp_page_default));
4285                                 memcpy(&lun->mode_pages.lbp_page[
4286                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4287                                        sizeof(lbp_page_changeable));
4288                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4289                                        &lbp_page_default,
4290                                        sizeof(lbp_page_default));
4291                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4292                                 value = ctl_get_opt(&lun->be_lun->options,
4293                                     "avail-threshold");
4294                                 if (value != NULL &&
4295                                     ctl_expand_number(value, &ival) == 0) {
4296                                         page->descr[0].flags |= SLBPPD_ENABLED |
4297                                             SLBPPD_ARMING_DEC;
4298                                         if (lun->be_lun->blocksize)
4299                                                 ival /= lun->be_lun->blocksize;
4300                                         else
4301                                                 ival /= 512;
4302                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4303                                             page->descr[0].count);
4304                                 }
4305                                 value = ctl_get_opt(&lun->be_lun->options,
4306                                     "used-threshold");
4307                                 if (value != NULL &&
4308                                     ctl_expand_number(value, &ival) == 0) {
4309                                         page->descr[1].flags |= SLBPPD_ENABLED |
4310                                             SLBPPD_ARMING_INC;
4311                                         if (lun->be_lun->blocksize)
4312                                                 ival /= lun->be_lun->blocksize;
4313                                         else
4314                                                 ival /= 512;
4315                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4316                                             page->descr[1].count);
4317                                 }
4318                                 value = ctl_get_opt(&lun->be_lun->options,
4319                                     "pool-avail-threshold");
4320                                 if (value != NULL &&
4321                                     ctl_expand_number(value, &ival) == 0) {
4322                                         page->descr[2].flags |= SLBPPD_ENABLED |
4323                                             SLBPPD_ARMING_DEC;
4324                                         if (lun->be_lun->blocksize)
4325                                                 ival /= lun->be_lun->blocksize;
4326                                         else
4327                                                 ival /= 512;
4328                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4329                                             page->descr[2].count);
4330                                 }
4331                                 value = ctl_get_opt(&lun->be_lun->options,
4332                                     "pool-used-threshold");
4333                                 if (value != NULL &&
4334                                     ctl_expand_number(value, &ival) == 0) {
4335                                         page->descr[3].flags |= SLBPPD_ENABLED |
4336                                             SLBPPD_ARMING_INC;
4337                                         if (lun->be_lun->blocksize)
4338                                                 ival /= lun->be_lun->blocksize;
4339                                         else
4340                                                 ival /= 512;
4341                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4342                                             page->descr[3].count);
4343                                 }
4344                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4345                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4346                                        sizeof(lbp_page_default));
4347                                 page_index->page_data =
4348                                         (uint8_t *)lun->mode_pages.lbp_page;
4349                                 break;
4350                         }
4351                         default:
4352                                 panic("subpage %#x for page %#x is incorrect!",
4353                                       page_index->subpage, page_code);
4354                         }
4355                         break;
4356                 }
4357                 case SMS_CDDVD_CAPS_PAGE:{
4358                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4359                             ("subpage %#x for page %#x is incorrect!",
4360                             page_index->subpage, page_code));
4361                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4362                                &cddvd_page_default,
4363                                sizeof(cddvd_page_default));
4364                         memcpy(&lun->mode_pages.cddvd_page[
4365                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4366                                sizeof(cddvd_page_changeable));
4367                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4368                                &cddvd_page_default,
4369                                sizeof(cddvd_page_default));
4370                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4371                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4372                                sizeof(cddvd_page_default));
4373                         page_index->page_data =
4374                                 (uint8_t *)lun->mode_pages.cddvd_page;
4375                         break;
4376                 }
4377                 default:
4378                         panic("invalid page code value %#x", page_code);
4379                 }
4380         }
4381
4382         return (CTL_RETVAL_COMPLETE);
4383 }
4384
4385 static int
4386 ctl_init_log_page_index(struct ctl_lun *lun)
4387 {
4388         struct ctl_page_index *page_index;
4389         int i, j, k, prev;
4390
4391         memcpy(&lun->log_pages.index, log_page_index_template,
4392                sizeof(log_page_index_template));
4393
4394         prev = -1;
4395         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4396
4397                 page_index = &lun->log_pages.index[i];
4398                 if (lun->be_lun->lun_type == T_DIRECT &&
4399                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4400                         continue;
4401                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4402                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4403                         continue;
4404                 if (lun->be_lun->lun_type == T_CDROM &&
4405                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4406                         continue;
4407
4408                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4409                     lun->backend->lun_attr == NULL)
4410                         continue;
4411
4412                 if (page_index->page_code != prev) {
4413                         lun->log_pages.pages_page[j] = page_index->page_code;
4414                         prev = page_index->page_code;
4415                         j++;
4416                 }
4417                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4418                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4419                 k++;
4420         }
4421         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4422         lun->log_pages.index[0].page_len = j;
4423         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4424         lun->log_pages.index[1].page_len = k * 2;
4425         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4426         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4427         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4428         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4429         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.ie_page;
4430         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.ie_page);
4431
4432         return (CTL_RETVAL_COMPLETE);
4433 }
4434
4435 static int
4436 hex2bin(const char *str, uint8_t *buf, int buf_size)
4437 {
4438         int i;
4439         u_char c;
4440
4441         memset(buf, 0, buf_size);
4442         while (isspace(str[0]))
4443                 str++;
4444         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4445                 str += 2;
4446         buf_size *= 2;
4447         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4448                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4449                         str++;
4450                 c = str[i];
4451                 if (isdigit(c))
4452                         c -= '0';
4453                 else if (isalpha(c))
4454                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4455                 else
4456                         break;
4457                 if (c >= 16)
4458                         break;
4459                 if ((i & 1) == 0)
4460                         buf[i / 2] |= (c << 4);
4461                 else
4462                         buf[i / 2] |= c;
4463         }
4464         return ((i + 1) / 2);
4465 }
4466
4467 /*
4468  * LUN allocation.
4469  *
4470  * Requirements:
4471  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4472  *   wants us to allocate the LUN and he can block.
4473  * - ctl_softc is always set
4474  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4475  *
4476  * Returns 0 for success, non-zero (errno) for failure.
4477  */
4478 static int
4479 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4480               struct ctl_be_lun *const be_lun)
4481 {
4482         struct ctl_lun *nlun, *lun;
4483         struct scsi_vpd_id_descriptor *desc;
4484         struct scsi_vpd_id_t10 *t10id;
4485         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4486         int lun_number, lun_malloced;
4487         int devidlen, idlen1, idlen2 = 0, len;
4488
4489         if (be_lun == NULL)
4490                 return (EINVAL);
4491
4492         /*
4493          * We currently only support Direct Access or Processor LUN types.
4494          */
4495         switch (be_lun->lun_type) {
4496         case T_DIRECT:
4497         case T_PROCESSOR:
4498         case T_CDROM:
4499                 break;
4500         case T_SEQUENTIAL:
4501         case T_CHANGER:
4502         default:
4503                 be_lun->lun_config_status(be_lun->be_lun,
4504                                           CTL_LUN_CONFIG_FAILURE);
4505                 break;
4506         }
4507         if (ctl_lun == NULL) {
4508                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4509                 lun_malloced = 1;
4510         } else {
4511                 lun_malloced = 0;
4512                 lun = ctl_lun;
4513         }
4514
4515         memset(lun, 0, sizeof(*lun));
4516         if (lun_malloced)
4517                 lun->flags = CTL_LUN_MALLOCED;
4518
4519         /* Generate LUN ID. */
4520         devidlen = max(CTL_DEVID_MIN_LEN,
4521             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4522         idlen1 = sizeof(*t10id) + devidlen;
4523         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4524         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4525         if (scsiname != NULL) {
4526                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4527                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4528         }
4529         eui = ctl_get_opt(&be_lun->options, "eui");
4530         if (eui != NULL) {
4531                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4532         }
4533         naa = ctl_get_opt(&be_lun->options, "naa");
4534         if (naa != NULL) {
4535                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4536         }
4537         uuid = ctl_get_opt(&be_lun->options, "uuid");
4538         if (uuid != NULL) {
4539                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4540         }
4541         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4542             M_CTL, M_WAITOK | M_ZERO);
4543         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4544         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4545         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4546         desc->length = idlen1;
4547         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4548         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4549         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4550                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4551         } else {
4552                 strncpy(t10id->vendor, vendor,
4553                     min(sizeof(t10id->vendor), strlen(vendor)));
4554         }
4555         strncpy((char *)t10id->vendor_spec_id,
4556             (char *)be_lun->device_id, devidlen);
4557         if (scsiname != NULL) {
4558                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4559                     desc->length);
4560                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4561                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4562                     SVPD_ID_TYPE_SCSI_NAME;
4563                 desc->length = idlen2;
4564                 strlcpy(desc->identifier, scsiname, idlen2);
4565         }
4566         if (eui != NULL) {
4567                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4568                     desc->length);
4569                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4570                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4571                     SVPD_ID_TYPE_EUI64;
4572                 desc->length = hex2bin(eui, desc->identifier, 16);
4573                 desc->length = desc->length > 12 ? 16 :
4574                     (desc->length > 8 ? 12 : 8);
4575                 len -= 16 - desc->length;
4576         }
4577         if (naa != NULL) {
4578                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4579                     desc->length);
4580                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4581                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4582                     SVPD_ID_TYPE_NAA;
4583                 desc->length = hex2bin(naa, desc->identifier, 16);
4584                 desc->length = desc->length > 8 ? 16 : 8;
4585                 len -= 16 - desc->length;
4586         }
4587         if (uuid != NULL) {
4588                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4589                     desc->length);
4590                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4591                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4592                     SVPD_ID_TYPE_UUID;
4593                 desc->identifier[0] = 0x10;
4594                 hex2bin(uuid, &desc->identifier[2], 16);
4595                 desc->length = 18;
4596         }
4597         lun->lun_devid->len = len;
4598
4599         mtx_lock(&ctl_softc->ctl_lock);
4600         /*
4601          * See if the caller requested a particular LUN number.  If so, see
4602          * if it is available.  Otherwise, allocate the first available LUN.
4603          */
4604         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4605                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4606                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4607                         mtx_unlock(&ctl_softc->ctl_lock);
4608                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4609                                 printf("ctl: requested LUN ID %d is higher "
4610                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4611                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4612                         } else {
4613                                 /*
4614                                  * XXX KDM return an error, or just assign
4615                                  * another LUN ID in this case??
4616                                  */
4617                                 printf("ctl: requested LUN ID %d is already "
4618                                        "in use\n", be_lun->req_lun_id);
4619                         }
4620 fail:
4621                         free(lun->lun_devid, M_CTL);
4622                         if (lun->flags & CTL_LUN_MALLOCED)
4623                                 free(lun, M_CTL);
4624                         be_lun->lun_config_status(be_lun->be_lun,
4625                                                   CTL_LUN_CONFIG_FAILURE);
4626                         return (ENOSPC);
4627                 }
4628                 lun_number = be_lun->req_lun_id;
4629         } else {
4630                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4631                 if (lun_number == -1) {
4632                         mtx_unlock(&ctl_softc->ctl_lock);
4633                         printf("ctl: can't allocate LUN, out of LUNs\n");
4634                         goto fail;
4635                 }
4636         }
4637         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4638         mtx_unlock(&ctl_softc->ctl_lock);
4639
4640         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4641         lun->lun = lun_number;
4642         lun->be_lun = be_lun;
4643         /*
4644          * The processor LUN is always enabled.  Disk LUNs come on line
4645          * disabled, and must be enabled by the backend.
4646          */
4647         lun->flags |= CTL_LUN_DISABLED;
4648         lun->backend = be_lun->be;
4649         be_lun->ctl_lun = lun;
4650         be_lun->lun_id = lun_number;
4651         atomic_add_int(&be_lun->be->num_luns, 1);
4652         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4653                 lun->flags |= CTL_LUN_EJECTED;
4654         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4655                 lun->flags |= CTL_LUN_NO_MEDIA;
4656         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4657                 lun->flags |= CTL_LUN_STOPPED;
4658
4659         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4660                 lun->flags |= CTL_LUN_PRIMARY_SC;
4661
4662         value = ctl_get_opt(&be_lun->options, "removable");
4663         if (value != NULL) {
4664                 if (strcmp(value, "on") == 0)
4665                         lun->flags |= CTL_LUN_REMOVABLE;
4666         } else if (be_lun->lun_type == T_CDROM)
4667                 lun->flags |= CTL_LUN_REMOVABLE;
4668
4669         lun->ctl_softc = ctl_softc;
4670 #ifdef CTL_TIME_IO
4671         lun->last_busy = getsbinuptime();
4672 #endif
4673         TAILQ_INIT(&lun->ooa_queue);
4674         TAILQ_INIT(&lun->blocked_queue);
4675         STAILQ_INIT(&lun->error_list);
4676         lun->ie_reported = 1;
4677         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4678         ctl_tpc_lun_init(lun);
4679         if (lun->flags & CTL_LUN_REMOVABLE) {
4680                 lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4,
4681                     M_CTL, M_WAITOK);
4682         }
4683
4684         /*
4685          * Initialize the mode and log page index.
4686          */
4687         ctl_init_page_index(lun);
4688         ctl_init_log_page_index(lun);
4689
4690         /* Setup statistics gathering */
4691 #ifdef CTL_LEGACY_STATS
4692         lun->legacy_stats.device_type = be_lun->lun_type;
4693         lun->legacy_stats.lun_number = lun_number;
4694         lun->legacy_stats.blocksize = be_lun->blocksize;
4695         if (be_lun->blocksize == 0)
4696                 lun->legacy_stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4697         for (len = 0; len < CTL_MAX_PORTS; len++)
4698                 lun->legacy_stats.ports[len].targ_port = len;
4699 #endif /* CTL_LEGACY_STATS */
4700         lun->stats.item = lun_number;
4701
4702         /*
4703          * Now, before we insert this lun on the lun list, set the lun
4704          * inventory changed UA for all other luns.
4705          */
4706         mtx_lock(&ctl_softc->ctl_lock);
4707         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4708                 mtx_lock(&nlun->lun_lock);
4709                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4710                 mtx_unlock(&nlun->lun_lock);
4711         }
4712         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4713         ctl_softc->ctl_luns[lun_number] = lun;
4714         ctl_softc->num_luns++;
4715         mtx_unlock(&ctl_softc->ctl_lock);
4716
4717         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4718         return (0);
4719 }
4720
4721 /*
4722  * Delete a LUN.
4723  * Assumptions:
4724  * - LUN has already been marked invalid and any pending I/O has been taken
4725  *   care of.
4726  */
4727 static int
4728 ctl_free_lun(struct ctl_lun *lun)
4729 {
4730         struct ctl_softc *softc = lun->ctl_softc;
4731         struct ctl_lun *nlun;
4732         int i;
4733
4734         mtx_assert(&softc->ctl_lock, MA_OWNED);
4735
4736         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4737
4738         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4739
4740         softc->ctl_luns[lun->lun] = NULL;
4741
4742         if (!TAILQ_EMPTY(&lun->ooa_queue))
4743                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4744
4745         softc->num_luns--;
4746
4747         /*
4748          * Tell the backend to free resources, if this LUN has a backend.
4749          */
4750         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4751         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4752
4753         lun->ie_reportcnt = UINT32_MAX;
4754         callout_drain(&lun->ie_callout);
4755
4756         ctl_tpc_lun_shutdown(lun);
4757         mtx_destroy(&lun->lun_lock);
4758         free(lun->lun_devid, M_CTL);
4759         for (i = 0; i < CTL_MAX_PORTS; i++)
4760                 free(lun->pending_ua[i], M_CTL);
4761         for (i = 0; i < CTL_MAX_PORTS; i++)
4762                 free(lun->pr_keys[i], M_CTL);
4763         free(lun->write_buffer, M_CTL);
4764         free(lun->prevent, M_CTL);
4765         if (lun->flags & CTL_LUN_MALLOCED)
4766                 free(lun, M_CTL);
4767
4768         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4769                 mtx_lock(&nlun->lun_lock);
4770                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4771                 mtx_unlock(&nlun->lun_lock);
4772         }
4773
4774         return (0);
4775 }
4776
4777 static void
4778 ctl_create_lun(struct ctl_be_lun *be_lun)
4779 {
4780
4781         /*
4782          * ctl_alloc_lun() should handle all potential failure cases.
4783          */
4784         ctl_alloc_lun(control_softc, NULL, be_lun);
4785 }
4786
4787 int
4788 ctl_add_lun(struct ctl_be_lun *be_lun)
4789 {
4790         struct ctl_softc *softc = control_softc;
4791
4792         mtx_lock(&softc->ctl_lock);
4793         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4794         mtx_unlock(&softc->ctl_lock);
4795         wakeup(&softc->pending_lun_queue);
4796
4797         return (0);
4798 }
4799
4800 int
4801 ctl_enable_lun(struct ctl_be_lun *be_lun)
4802 {
4803         struct ctl_softc *softc;
4804         struct ctl_port *port, *nport;
4805         struct ctl_lun *lun;
4806         int retval;
4807
4808         lun = (struct ctl_lun *)be_lun->ctl_lun;
4809         softc = lun->ctl_softc;
4810
4811         mtx_lock(&softc->ctl_lock);
4812         mtx_lock(&lun->lun_lock);
4813         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4814                 /*
4815                  * eh?  Why did we get called if the LUN is already
4816                  * enabled?
4817                  */
4818                 mtx_unlock(&lun->lun_lock);
4819                 mtx_unlock(&softc->ctl_lock);
4820                 return (0);
4821         }
4822         lun->flags &= ~CTL_LUN_DISABLED;
4823         mtx_unlock(&lun->lun_lock);
4824
4825         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4826                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4827                     port->lun_map != NULL || port->lun_enable == NULL)
4828                         continue;
4829
4830                 /*
4831                  * Drop the lock while we call the FETD's enable routine.
4832                  * This can lead to a callback into CTL (at least in the
4833                  * case of the internal initiator frontend.
4834                  */
4835                 mtx_unlock(&softc->ctl_lock);
4836                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4837                 mtx_lock(&softc->ctl_lock);
4838                 if (retval != 0) {
4839                         printf("%s: FETD %s port %d returned error "
4840                                "%d for lun_enable on lun %jd\n",
4841                                __func__, port->port_name, port->targ_port,
4842                                retval, (intmax_t)lun->lun);
4843                 }
4844         }
4845
4846         mtx_unlock(&softc->ctl_lock);
4847         ctl_isc_announce_lun(lun);
4848
4849         return (0);
4850 }
4851
4852 int
4853 ctl_disable_lun(struct ctl_be_lun *be_lun)
4854 {
4855         struct ctl_softc *softc;
4856         struct ctl_port *port;
4857         struct ctl_lun *lun;
4858         int retval;
4859
4860         lun = (struct ctl_lun *)be_lun->ctl_lun;
4861         softc = lun->ctl_softc;
4862
4863         mtx_lock(&softc->ctl_lock);
4864         mtx_lock(&lun->lun_lock);
4865         if (lun->flags & CTL_LUN_DISABLED) {
4866                 mtx_unlock(&lun->lun_lock);
4867                 mtx_unlock(&softc->ctl_lock);
4868                 return (0);
4869         }
4870         lun->flags |= CTL_LUN_DISABLED;
4871         mtx_unlock(&lun->lun_lock);
4872
4873         STAILQ_FOREACH(port, &softc->port_list, links) {
4874                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4875                     port->lun_map != NULL || port->lun_disable == NULL)
4876                         continue;
4877
4878                 /*
4879                  * Drop the lock before we call the frontend's disable
4880                  * routine, to avoid lock order reversals.
4881                  *
4882                  * XXX KDM what happens if the frontend list changes while
4883                  * we're traversing it?  It's unlikely, but should be handled.
4884                  */
4885                 mtx_unlock(&softc->ctl_lock);
4886                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4887                 mtx_lock(&softc->ctl_lock);
4888                 if (retval != 0) {
4889                         printf("%s: FETD %s port %d returned error "
4890                                "%d for lun_disable on lun %jd\n",
4891                                __func__, port->port_name, port->targ_port,
4892                                retval, (intmax_t)lun->lun);
4893                 }
4894         }
4895
4896         mtx_unlock(&softc->ctl_lock);
4897         ctl_isc_announce_lun(lun);
4898
4899         return (0);
4900 }
4901
4902 int
4903 ctl_start_lun(struct ctl_be_lun *be_lun)
4904 {
4905         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4906
4907         mtx_lock(&lun->lun_lock);
4908         lun->flags &= ~CTL_LUN_STOPPED;
4909         mtx_unlock(&lun->lun_lock);
4910         return (0);
4911 }
4912
4913 int
4914 ctl_stop_lun(struct ctl_be_lun *be_lun)
4915 {
4916         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4917
4918         mtx_lock(&lun->lun_lock);
4919         lun->flags |= CTL_LUN_STOPPED;
4920         mtx_unlock(&lun->lun_lock);
4921         return (0);
4922 }
4923
4924 int
4925 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4926 {
4927         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4928
4929         mtx_lock(&lun->lun_lock);
4930         lun->flags |= CTL_LUN_NO_MEDIA;
4931         mtx_unlock(&lun->lun_lock);
4932         return (0);
4933 }
4934
4935 int
4936 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4937 {
4938         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4939         union ctl_ha_msg msg;
4940
4941         mtx_lock(&lun->lun_lock);
4942         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4943         if (lun->flags & CTL_LUN_REMOVABLE)
4944                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4945         mtx_unlock(&lun->lun_lock);
4946         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4947             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4948                 bzero(&msg.ua, sizeof(msg.ua));
4949                 msg.hdr.msg_type = CTL_MSG_UA;
4950                 msg.hdr.nexus.initid = -1;
4951                 msg.hdr.nexus.targ_port = -1;
4952                 msg.hdr.nexus.targ_lun = lun->lun;
4953                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4954                 msg.ua.ua_all = 1;
4955                 msg.ua.ua_set = 1;
4956                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4957                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4958                     M_WAITOK);
4959         }
4960         return (0);
4961 }
4962
4963 int
4964 ctl_lun_ejected(struct ctl_be_lun *be_lun)
4965 {
4966         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4967
4968         mtx_lock(&lun->lun_lock);
4969         lun->flags |= CTL_LUN_EJECTED;
4970         mtx_unlock(&lun->lun_lock);
4971         return (0);
4972 }
4973
4974 int
4975 ctl_lun_primary(struct ctl_be_lun *be_lun)
4976 {
4977         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4978
4979         mtx_lock(&lun->lun_lock);
4980         lun->flags |= CTL_LUN_PRIMARY_SC;
4981         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4982         mtx_unlock(&lun->lun_lock);
4983         ctl_isc_announce_lun(lun);
4984         return (0);
4985 }
4986
4987 int
4988 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4989 {
4990         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4991
4992         mtx_lock(&lun->lun_lock);
4993         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4994         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4995         mtx_unlock(&lun->lun_lock);
4996         ctl_isc_announce_lun(lun);
4997         return (0);
4998 }
4999
5000 int
5001 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
5002 {
5003         struct ctl_softc *softc;
5004         struct ctl_lun *lun;
5005
5006         lun = (struct ctl_lun *)be_lun->ctl_lun;
5007         softc = lun->ctl_softc;
5008
5009         mtx_lock(&lun->lun_lock);
5010
5011         /*
5012          * The LUN needs to be disabled before it can be marked invalid.
5013          */
5014         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
5015                 mtx_unlock(&lun->lun_lock);
5016                 return (-1);
5017         }
5018         /*
5019          * Mark the LUN invalid.
5020          */
5021         lun->flags |= CTL_LUN_INVALID;
5022
5023         /*
5024          * If there is nothing in the OOA queue, go ahead and free the LUN.
5025          * If we have something in the OOA queue, we'll free it when the
5026          * last I/O completes.
5027          */
5028         if (TAILQ_EMPTY(&lun->ooa_queue)) {
5029                 mtx_unlock(&lun->lun_lock);
5030                 mtx_lock(&softc->ctl_lock);
5031                 ctl_free_lun(lun);
5032                 mtx_unlock(&softc->ctl_lock);
5033         } else
5034                 mtx_unlock(&lun->lun_lock);
5035
5036         return (0);
5037 }
5038
5039 void
5040 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5041 {
5042         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5043         union ctl_ha_msg msg;
5044
5045         mtx_lock(&lun->lun_lock);
5046         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
5047         mtx_unlock(&lun->lun_lock);
5048         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5049                 /* Send msg to other side. */
5050                 bzero(&msg.ua, sizeof(msg.ua));
5051                 msg.hdr.msg_type = CTL_MSG_UA;
5052                 msg.hdr.nexus.initid = -1;
5053                 msg.hdr.nexus.targ_port = -1;
5054                 msg.hdr.nexus.targ_lun = lun->lun;
5055                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5056                 msg.ua.ua_all = 1;
5057                 msg.ua.ua_set = 1;
5058                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
5059                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5060                     M_WAITOK);
5061         }
5062 }
5063
5064 /*
5065  * Backend "memory move is complete" callback for requests that never
5066  * make it down to say RAIDCore's configuration code.
5067  */
5068 int
5069 ctl_config_move_done(union ctl_io *io)
5070 {
5071         int retval;
5072
5073         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5074         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5075             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5076
5077         if ((io->io_hdr.port_status != 0) &&
5078             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5079              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5080                 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
5081                     /*retry_count*/ io->io_hdr.port_status);
5082         } else if (io->scsiio.kern_data_resid != 0 &&
5083             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
5084             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5085              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5086                 ctl_set_invalid_field_ciu(&io->scsiio);
5087         }
5088
5089         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5090                 ctl_data_print(io);
5091         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5092             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5093              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5094             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5095                 /*
5096                  * XXX KDM just assuming a single pointer here, and not a
5097                  * S/G list.  If we start using S/G lists for config data,
5098                  * we'll need to know how to clean them up here as well.
5099                  */
5100                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5101                         free(io->scsiio.kern_data_ptr, M_CTL);
5102                 ctl_done(io);
5103                 retval = CTL_RETVAL_COMPLETE;
5104         } else {
5105                 /*
5106                  * XXX KDM now we need to continue data movement.  Some
5107                  * options:
5108                  * - call ctl_scsiio() again?  We don't do this for data
5109                  *   writes, because for those at least we know ahead of
5110                  *   time where the write will go and how long it is.  For
5111                  *   config writes, though, that information is largely
5112                  *   contained within the write itself, thus we need to
5113                  *   parse out the data again.
5114                  *
5115                  * - Call some other function once the data is in?
5116                  */
5117
5118                 /*
5119                  * XXX KDM call ctl_scsiio() again for now, and check flag
5120                  * bits to see whether we're allocated or not.
5121                  */
5122                 retval = ctl_scsiio(&io->scsiio);
5123         }
5124         return (retval);
5125 }
5126
5127 /*
5128  * This gets called by a backend driver when it is done with a
5129  * data_submit method.
5130  */
5131 void
5132 ctl_data_submit_done(union ctl_io *io)
5133 {
5134         /*
5135          * If the IO_CONT flag is set, we need to call the supplied
5136          * function to continue processing the I/O, instead of completing
5137          * the I/O just yet.
5138          *
5139          * If there is an error, though, we don't want to keep processing.
5140          * Instead, just send status back to the initiator.
5141          */
5142         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5143             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5144             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5145              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5146                 io->scsiio.io_cont(io);
5147                 return;
5148         }
5149         ctl_done(io);
5150 }
5151
5152 /*
5153  * This gets called by a backend driver when it is done with a
5154  * configuration write.
5155  */
5156 void
5157 ctl_config_write_done(union ctl_io *io)
5158 {
5159         uint8_t *buf;
5160
5161         /*
5162          * If the IO_CONT flag is set, we need to call the supplied
5163          * function to continue processing the I/O, instead of completing
5164          * the I/O just yet.
5165          *
5166          * If there is an error, though, we don't want to keep processing.
5167          * Instead, just send status back to the initiator.
5168          */
5169         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5170             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5171             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5172              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5173                 io->scsiio.io_cont(io);
5174                 return;
5175         }
5176         /*
5177          * Since a configuration write can be done for commands that actually
5178          * have data allocated, like write buffer, and commands that have
5179          * no data, like start/stop unit, we need to check here.
5180          */
5181         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5182                 buf = io->scsiio.kern_data_ptr;
5183         else
5184                 buf = NULL;
5185         ctl_done(io);
5186         if (buf)
5187                 free(buf, M_CTL);
5188 }
5189
5190 void
5191 ctl_config_read_done(union ctl_io *io)
5192 {
5193         uint8_t *buf;
5194
5195         /*
5196          * If there is some error -- we are done, skip data transfer.
5197          */
5198         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5199             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5200              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5201                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5202                         buf = io->scsiio.kern_data_ptr;
5203                 else
5204                         buf = NULL;
5205                 ctl_done(io);
5206                 if (buf)
5207                         free(buf, M_CTL);
5208                 return;
5209         }
5210
5211         /*
5212          * If the IO_CONT flag is set, we need to call the supplied
5213          * function to continue processing the I/O, instead of completing
5214          * the I/O just yet.
5215          */
5216         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5217                 io->scsiio.io_cont(io);
5218                 return;
5219         }
5220
5221         ctl_datamove(io);
5222 }
5223
5224 /*
5225  * SCSI release command.
5226  */
5227 int
5228 ctl_scsi_release(struct ctl_scsiio *ctsio)
5229 {
5230         struct ctl_lun *lun = CTL_LUN(ctsio);
5231         uint32_t residx;
5232
5233         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5234
5235         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5236
5237         /*
5238          * XXX KDM right now, we only support LUN reservation.  We don't
5239          * support 3rd party reservations, or extent reservations, which
5240          * might actually need the parameter list.  If we've gotten this
5241          * far, we've got a LUN reservation.  Anything else got kicked out
5242          * above.  So, according to SPC, ignore the length.
5243          */
5244
5245         mtx_lock(&lun->lun_lock);
5246
5247         /*
5248          * According to SPC, it is not an error for an intiator to attempt
5249          * to release a reservation on a LUN that isn't reserved, or that
5250          * is reserved by another initiator.  The reservation can only be
5251          * released, though, by the initiator who made it or by one of
5252          * several reset type events.
5253          */
5254         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5255                         lun->flags &= ~CTL_LUN_RESERVED;
5256
5257         mtx_unlock(&lun->lun_lock);
5258
5259         ctl_set_success(ctsio);
5260         ctl_done((union ctl_io *)ctsio);
5261         return (CTL_RETVAL_COMPLETE);
5262 }
5263
5264 int
5265 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5266 {
5267         struct ctl_lun *lun = CTL_LUN(ctsio);
5268         uint32_t residx;
5269
5270         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5271
5272         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5273
5274         /*
5275          * XXX KDM right now, we only support LUN reservation.  We don't
5276          * support 3rd party reservations, or extent reservations, which
5277          * might actually need the parameter list.  If we've gotten this
5278          * far, we've got a LUN reservation.  Anything else got kicked out
5279          * above.  So, according to SPC, ignore the length.
5280          */
5281
5282         mtx_lock(&lun->lun_lock);
5283         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5284                 ctl_set_reservation_conflict(ctsio);
5285                 goto bailout;
5286         }
5287
5288         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5289         if (lun->flags & CTL_LUN_PR_RESERVED) {
5290                 ctl_set_success(ctsio);
5291                 goto bailout;
5292         }
5293
5294         lun->flags |= CTL_LUN_RESERVED;
5295         lun->res_idx = residx;
5296         ctl_set_success(ctsio);
5297
5298 bailout:
5299         mtx_unlock(&lun->lun_lock);
5300         ctl_done((union ctl_io *)ctsio);
5301         return (CTL_RETVAL_COMPLETE);
5302 }
5303
5304 int
5305 ctl_start_stop(struct ctl_scsiio *ctsio)
5306 {
5307         struct ctl_lun *lun = CTL_LUN(ctsio);
5308         struct scsi_start_stop_unit *cdb;
5309         int retval;
5310
5311         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5312
5313         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5314
5315         if ((cdb->how & SSS_PC_MASK) == 0) {
5316                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5317                     (cdb->how & SSS_START) == 0) {
5318                         uint32_t residx;
5319
5320                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5321                         if (ctl_get_prkey(lun, residx) == 0 ||
5322                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5323
5324                                 ctl_set_reservation_conflict(ctsio);
5325                                 ctl_done((union ctl_io *)ctsio);
5326                                 return (CTL_RETVAL_COMPLETE);
5327                         }
5328                 }
5329
5330                 if ((cdb->how & SSS_LOEJ) &&
5331                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5332                         ctl_set_invalid_field(ctsio,
5333                                               /*sks_valid*/ 1,
5334                                               /*command*/ 1,
5335                                               /*field*/ 4,
5336                                               /*bit_valid*/ 1,
5337                                               /*bit*/ 1);
5338                         ctl_done((union ctl_io *)ctsio);
5339                         return (CTL_RETVAL_COMPLETE);
5340                 }
5341
5342                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5343                     lun->prevent_count > 0) {
5344                         /* "Medium removal prevented" */
5345                         ctl_set_sense(ctsio, /*current_error*/ 1,
5346                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5347                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5348                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5349                         ctl_done((union ctl_io *)ctsio);
5350                         return (CTL_RETVAL_COMPLETE);
5351                 }
5352         }
5353
5354         retval = lun->backend->config_write((union ctl_io *)ctsio);
5355         return (retval);
5356 }
5357
5358 int
5359 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5360 {
5361         struct ctl_lun *lun = CTL_LUN(ctsio);
5362         struct scsi_prevent *cdb;
5363         int retval;
5364         uint32_t initidx;
5365
5366         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5367
5368         cdb = (struct scsi_prevent *)ctsio->cdb;
5369
5370         if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) {
5371                 ctl_set_invalid_opcode(ctsio);
5372                 ctl_done((union ctl_io *)ctsio);
5373                 return (CTL_RETVAL_COMPLETE);
5374         }
5375
5376         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5377         mtx_lock(&lun->lun_lock);
5378         if ((cdb->how & PR_PREVENT) &&
5379             ctl_is_set(lun->prevent, initidx) == 0) {
5380                 ctl_set_mask(lun->prevent, initidx);
5381                 lun->prevent_count++;
5382         } else if ((cdb->how & PR_PREVENT) == 0 &&
5383             ctl_is_set(lun->prevent, initidx)) {
5384                 ctl_clear_mask(lun->prevent, initidx);
5385                 lun->prevent_count--;
5386         }
5387         mtx_unlock(&lun->lun_lock);
5388         retval = lun->backend->config_write((union ctl_io *)ctsio);
5389         return (retval);
5390 }
5391
5392 /*
5393  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5394  * we don't really do anything with the LBA and length fields if the user
5395  * passes them in.  Instead we'll just flush out the cache for the entire
5396  * LUN.
5397  */
5398 int
5399 ctl_sync_cache(struct ctl_scsiio *ctsio)
5400 {
5401         struct ctl_lun *lun = CTL_LUN(ctsio);
5402         struct ctl_lba_len_flags *lbalen;
5403         uint64_t starting_lba;
5404         uint32_t block_count;
5405         int retval;
5406         uint8_t byte2;
5407
5408         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5409
5410         retval = 0;
5411
5412         switch (ctsio->cdb[0]) {
5413         case SYNCHRONIZE_CACHE: {
5414                 struct scsi_sync_cache *cdb;
5415                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5416
5417                 starting_lba = scsi_4btoul(cdb->begin_lba);
5418                 block_count = scsi_2btoul(cdb->lb_count);
5419                 byte2 = cdb->byte2;
5420                 break;
5421         }
5422         case SYNCHRONIZE_CACHE_16: {
5423                 struct scsi_sync_cache_16 *cdb;
5424                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5425
5426                 starting_lba = scsi_8btou64(cdb->begin_lba);
5427                 block_count = scsi_4btoul(cdb->lb_count);
5428                 byte2 = cdb->byte2;
5429                 break;
5430         }
5431         default:
5432                 ctl_set_invalid_opcode(ctsio);
5433                 ctl_done((union ctl_io *)ctsio);
5434                 goto bailout;
5435                 break; /* NOTREACHED */
5436         }
5437
5438         /*
5439          * We check the LBA and length, but don't do anything with them.
5440          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5441          * get flushed.  This check will just help satisfy anyone who wants
5442          * to see an error for an out of range LBA.
5443          */
5444         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5445                 ctl_set_lba_out_of_range(ctsio,
5446                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5447                 ctl_done((union ctl_io *)ctsio);
5448                 goto bailout;
5449         }
5450
5451         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5452         lbalen->lba = starting_lba;
5453         lbalen->len = block_count;
5454         lbalen->flags = byte2;
5455         retval = lun->backend->config_write((union ctl_io *)ctsio);
5456
5457 bailout:
5458         return (retval);
5459 }
5460
5461 int
5462 ctl_format(struct ctl_scsiio *ctsio)
5463 {
5464         struct scsi_format *cdb;
5465         int length, defect_list_len;
5466
5467         CTL_DEBUG_PRINT(("ctl_format\n"));
5468
5469         cdb = (struct scsi_format *)ctsio->cdb;
5470
5471         length = 0;
5472         if (cdb->byte2 & SF_FMTDATA) {
5473                 if (cdb->byte2 & SF_LONGLIST)
5474                         length = sizeof(struct scsi_format_header_long);
5475                 else
5476                         length = sizeof(struct scsi_format_header_short);
5477         }
5478
5479         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5480          && (length > 0)) {
5481                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5482                 ctsio->kern_data_len = length;
5483                 ctsio->kern_total_len = length;
5484                 ctsio->kern_rel_offset = 0;
5485                 ctsio->kern_sg_entries = 0;
5486                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5487                 ctsio->be_move_done = ctl_config_move_done;
5488                 ctl_datamove((union ctl_io *)ctsio);
5489
5490                 return (CTL_RETVAL_COMPLETE);
5491         }
5492
5493         defect_list_len = 0;
5494
5495         if (cdb->byte2 & SF_FMTDATA) {
5496                 if (cdb->byte2 & SF_LONGLIST) {
5497                         struct scsi_format_header_long *header;
5498
5499                         header = (struct scsi_format_header_long *)
5500                                 ctsio->kern_data_ptr;
5501
5502                         defect_list_len = scsi_4btoul(header->defect_list_len);
5503                         if (defect_list_len != 0) {
5504                                 ctl_set_invalid_field(ctsio,
5505                                                       /*sks_valid*/ 1,
5506                                                       /*command*/ 0,
5507                                                       /*field*/ 2,
5508                                                       /*bit_valid*/ 0,
5509                                                       /*bit*/ 0);
5510                                 goto bailout;
5511                         }
5512                 } else {
5513                         struct scsi_format_header_short *header;
5514
5515                         header = (struct scsi_format_header_short *)
5516                                 ctsio->kern_data_ptr;
5517
5518                         defect_list_len = scsi_2btoul(header->defect_list_len);
5519                         if (defect_list_len != 0) {
5520                                 ctl_set_invalid_field(ctsio,
5521                                                       /*sks_valid*/ 1,
5522                                                       /*command*/ 0,
5523                                                       /*field*/ 2,
5524                                                       /*bit_valid*/ 0,
5525                                                       /*bit*/ 0);
5526                                 goto bailout;
5527                         }
5528                 }
5529         }
5530
5531         ctl_set_success(ctsio);
5532 bailout:
5533
5534         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5535                 free(ctsio->kern_data_ptr, M_CTL);
5536                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5537         }
5538
5539         ctl_done((union ctl_io *)ctsio);
5540         return (CTL_RETVAL_COMPLETE);
5541 }
5542
5543 int
5544 ctl_read_buffer(struct ctl_scsiio *ctsio)
5545 {
5546         struct ctl_lun *lun = CTL_LUN(ctsio);
5547         uint64_t buffer_offset;
5548         uint32_t len;
5549         uint8_t byte2;
5550         static uint8_t descr[4];
5551         static uint8_t echo_descr[4] = { 0 };
5552
5553         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5554
5555         switch (ctsio->cdb[0]) {
5556         case READ_BUFFER: {
5557                 struct scsi_read_buffer *cdb;
5558
5559                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5560                 buffer_offset = scsi_3btoul(cdb->offset);
5561                 len = scsi_3btoul(cdb->length);
5562                 byte2 = cdb->byte2;
5563                 break;
5564         }
5565         case READ_BUFFER_16: {
5566                 struct scsi_read_buffer_16 *cdb;
5567
5568                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5569                 buffer_offset = scsi_8btou64(cdb->offset);
5570                 len = scsi_4btoul(cdb->length);
5571                 byte2 = cdb->byte2;
5572                 break;
5573         }
5574         default: /* This shouldn't happen. */
5575                 ctl_set_invalid_opcode(ctsio);
5576                 ctl_done((union ctl_io *)ctsio);
5577                 return (CTL_RETVAL_COMPLETE);
5578         }
5579
5580         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5581             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5582                 ctl_set_invalid_field(ctsio,
5583                                       /*sks_valid*/ 1,
5584                                       /*command*/ 1,
5585                                       /*field*/ 6,
5586                                       /*bit_valid*/ 0,
5587                                       /*bit*/ 0);
5588                 ctl_done((union ctl_io *)ctsio);
5589                 return (CTL_RETVAL_COMPLETE);
5590         }
5591
5592         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5593                 descr[0] = 0;
5594                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5595                 ctsio->kern_data_ptr = descr;
5596                 len = min(len, sizeof(descr));
5597         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5598                 ctsio->kern_data_ptr = echo_descr;
5599                 len = min(len, sizeof(echo_descr));
5600         } else {
5601                 if (lun->write_buffer == NULL) {
5602                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5603                             M_CTL, M_WAITOK);
5604                 }
5605                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5606         }
5607         ctsio->kern_data_len = len;
5608         ctsio->kern_total_len = len;
5609         ctsio->kern_rel_offset = 0;
5610         ctsio->kern_sg_entries = 0;
5611         ctl_set_success(ctsio);
5612         ctsio->be_move_done = ctl_config_move_done;
5613         ctl_datamove((union ctl_io *)ctsio);
5614         return (CTL_RETVAL_COMPLETE);
5615 }
5616
5617 int
5618 ctl_write_buffer(struct ctl_scsiio *ctsio)
5619 {
5620         struct ctl_lun *lun = CTL_LUN(ctsio);
5621         struct scsi_write_buffer *cdb;
5622         int buffer_offset, len;
5623
5624         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5625
5626         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5627
5628         len = scsi_3btoul(cdb->length);
5629         buffer_offset = scsi_3btoul(cdb->offset);
5630
5631         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5632                 ctl_set_invalid_field(ctsio,
5633                                       /*sks_valid*/ 1,
5634                                       /*command*/ 1,
5635                                       /*field*/ 6,
5636                                       /*bit_valid*/ 0,
5637                                       /*bit*/ 0);
5638                 ctl_done((union ctl_io *)ctsio);
5639                 return (CTL_RETVAL_COMPLETE);
5640         }
5641
5642         /*
5643          * If we've got a kernel request that hasn't been malloced yet,
5644          * malloc it and tell the caller the data buffer is here.
5645          */
5646         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5647                 if (lun->write_buffer == NULL) {
5648                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5649                             M_CTL, M_WAITOK);
5650                 }
5651                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5652                 ctsio->kern_data_len = len;
5653                 ctsio->kern_total_len = len;
5654                 ctsio->kern_rel_offset = 0;
5655                 ctsio->kern_sg_entries = 0;
5656                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5657                 ctsio->be_move_done = ctl_config_move_done;
5658                 ctl_datamove((union ctl_io *)ctsio);
5659
5660                 return (CTL_RETVAL_COMPLETE);
5661         }
5662
5663         ctl_set_success(ctsio);
5664         ctl_done((union ctl_io *)ctsio);
5665         return (CTL_RETVAL_COMPLETE);
5666 }
5667
5668 int
5669 ctl_write_same(struct ctl_scsiio *ctsio)
5670 {
5671         struct ctl_lun *lun = CTL_LUN(ctsio);
5672         struct ctl_lba_len_flags *lbalen;
5673         uint64_t lba;
5674         uint32_t num_blocks;
5675         int len, retval;
5676         uint8_t byte2;
5677
5678         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5679
5680         switch (ctsio->cdb[0]) {
5681         case WRITE_SAME_10: {
5682                 struct scsi_write_same_10 *cdb;
5683
5684                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5685
5686                 lba = scsi_4btoul(cdb->addr);
5687                 num_blocks = scsi_2btoul(cdb->length);
5688                 byte2 = cdb->byte2;
5689                 break;
5690         }
5691         case WRITE_SAME_16: {
5692                 struct scsi_write_same_16 *cdb;
5693
5694                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5695
5696                 lba = scsi_8btou64(cdb->addr);
5697                 num_blocks = scsi_4btoul(cdb->length);
5698                 byte2 = cdb->byte2;
5699                 break;
5700         }
5701         default:
5702                 /*
5703                  * We got a command we don't support.  This shouldn't
5704                  * happen, commands should be filtered out above us.
5705                  */
5706                 ctl_set_invalid_opcode(ctsio);
5707                 ctl_done((union ctl_io *)ctsio);
5708
5709                 return (CTL_RETVAL_COMPLETE);
5710                 break; /* NOTREACHED */
5711         }
5712
5713         /* ANCHOR flag can be used only together with UNMAP */
5714         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5715                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5716                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5717                 ctl_done((union ctl_io *)ctsio);
5718                 return (CTL_RETVAL_COMPLETE);
5719         }
5720
5721         /*
5722          * The first check is to make sure we're in bounds, the second
5723          * check is to catch wrap-around problems.  If the lba + num blocks
5724          * is less than the lba, then we've wrapped around and the block
5725          * range is invalid anyway.
5726          */
5727         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5728          || ((lba + num_blocks) < lba)) {
5729                 ctl_set_lba_out_of_range(ctsio,
5730                     MAX(lba, lun->be_lun->maxlba + 1));
5731                 ctl_done((union ctl_io *)ctsio);
5732                 return (CTL_RETVAL_COMPLETE);
5733         }
5734
5735         /* Zero number of blocks means "to the last logical block" */
5736         if (num_blocks == 0) {
5737                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5738                         ctl_set_invalid_field(ctsio,
5739                                               /*sks_valid*/ 0,
5740                                               /*command*/ 1,
5741                                               /*field*/ 0,
5742                                               /*bit_valid*/ 0,
5743                                               /*bit*/ 0);
5744                         ctl_done((union ctl_io *)ctsio);
5745                         return (CTL_RETVAL_COMPLETE);
5746                 }
5747                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5748         }
5749
5750         len = lun->be_lun->blocksize;
5751
5752         /*
5753          * If we've got a kernel request that hasn't been malloced yet,
5754          * malloc it and tell the caller the data buffer is here.
5755          */
5756         if ((byte2 & SWS_NDOB) == 0 &&
5757             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5758                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5759                 ctsio->kern_data_len = len;
5760                 ctsio->kern_total_len = len;
5761                 ctsio->kern_rel_offset = 0;
5762                 ctsio->kern_sg_entries = 0;
5763                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5764                 ctsio->be_move_done = ctl_config_move_done;
5765                 ctl_datamove((union ctl_io *)ctsio);
5766
5767                 return (CTL_RETVAL_COMPLETE);
5768         }
5769
5770         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5771         lbalen->lba = lba;
5772         lbalen->len = num_blocks;
5773         lbalen->flags = byte2;
5774         retval = lun->backend->config_write((union ctl_io *)ctsio);
5775
5776         return (retval);
5777 }
5778
5779 int
5780 ctl_unmap(struct ctl_scsiio *ctsio)
5781 {
5782         struct ctl_lun *lun = CTL_LUN(ctsio);
5783         struct scsi_unmap *cdb;
5784         struct ctl_ptr_len_flags *ptrlen;
5785         struct scsi_unmap_header *hdr;
5786         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5787         uint64_t lba;
5788         uint32_t num_blocks;
5789         int len, retval;
5790         uint8_t byte2;
5791
5792         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5793
5794         cdb = (struct scsi_unmap *)ctsio->cdb;
5795         len = scsi_2btoul(cdb->length);
5796         byte2 = cdb->byte2;
5797
5798         /*
5799          * If we've got a kernel request that hasn't been malloced yet,
5800          * malloc it and tell the caller the data buffer is here.
5801          */
5802         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5803                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5804                 ctsio->kern_data_len = len;
5805                 ctsio->kern_total_len = len;
5806                 ctsio->kern_rel_offset = 0;
5807                 ctsio->kern_sg_entries = 0;
5808                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5809                 ctsio->be_move_done = ctl_config_move_done;
5810                 ctl_datamove((union ctl_io *)ctsio);
5811
5812                 return (CTL_RETVAL_COMPLETE);
5813         }
5814
5815         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5816         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5817         if (len < sizeof (*hdr) ||
5818             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5819             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5820             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5821                 ctl_set_invalid_field(ctsio,
5822                                       /*sks_valid*/ 0,
5823                                       /*command*/ 0,
5824                                       /*field*/ 0,
5825                                       /*bit_valid*/ 0,
5826                                       /*bit*/ 0);
5827                 goto done;
5828         }
5829         len = scsi_2btoul(hdr->desc_length);
5830         buf = (struct scsi_unmap_desc *)(hdr + 1);
5831         end = buf + len / sizeof(*buf);
5832
5833         endnz = buf;
5834         for (range = buf; range < end; range++) {
5835                 lba = scsi_8btou64(range->lba);
5836                 num_blocks = scsi_4btoul(range->length);
5837                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5838                  || ((lba + num_blocks) < lba)) {
5839                         ctl_set_lba_out_of_range(ctsio,
5840                             MAX(lba, lun->be_lun->maxlba + 1));
5841                         ctl_done((union ctl_io *)ctsio);
5842                         return (CTL_RETVAL_COMPLETE);
5843                 }
5844                 if (num_blocks != 0)
5845                         endnz = range + 1;
5846         }
5847
5848         /*
5849          * Block backend can not handle zero last range.
5850          * Filter it out and return if there is nothing left.
5851          */
5852         len = (uint8_t *)endnz - (uint8_t *)buf;
5853         if (len == 0) {
5854                 ctl_set_success(ctsio);
5855                 goto done;
5856         }
5857
5858         mtx_lock(&lun->lun_lock);
5859         ptrlen = (struct ctl_ptr_len_flags *)
5860             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5861         ptrlen->ptr = (void *)buf;
5862         ptrlen->len = len;
5863         ptrlen->flags = byte2;
5864         ctl_check_blocked(lun);
5865         mtx_unlock(&lun->lun_lock);
5866
5867         retval = lun->backend->config_write((union ctl_io *)ctsio);
5868         return (retval);
5869
5870 done:
5871         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5872                 free(ctsio->kern_data_ptr, M_CTL);
5873                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5874         }
5875         ctl_done((union ctl_io *)ctsio);
5876         return (CTL_RETVAL_COMPLETE);
5877 }
5878
5879 int
5880 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5881                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5882 {
5883         struct ctl_lun *lun = CTL_LUN(ctsio);
5884         uint8_t *current_cp;
5885         int set_ua;
5886         uint32_t initidx;
5887
5888         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5889         set_ua = 0;
5890
5891         current_cp = (page_index->page_data + (page_index->page_len *
5892             CTL_PAGE_CURRENT));
5893
5894         mtx_lock(&lun->lun_lock);
5895         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5896                 memcpy(current_cp, page_ptr, page_index->page_len);
5897                 set_ua = 1;
5898         }
5899         if (set_ua != 0)
5900                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5901         mtx_unlock(&lun->lun_lock);
5902         if (set_ua) {
5903                 ctl_isc_announce_mode(lun,
5904                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5905                     page_index->page_code, page_index->subpage);
5906         }
5907         return (CTL_RETVAL_COMPLETE);
5908 }
5909
5910 static void
5911 ctl_ie_timer(void *arg)
5912 {
5913         struct ctl_lun *lun = arg;
5914         uint64_t t;
5915
5916         if (lun->ie_asc == 0)
5917                 return;
5918
5919         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5920                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5921         else
5922                 lun->ie_reported = 0;
5923
5924         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5925                 lun->ie_reportcnt++;
5926                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5927                 if (t == 0 || t == UINT32_MAX)
5928                         t = 3000;  /* 5 min */
5929                 callout_schedule(&lun->ie_callout, t * hz / 10);
5930         }
5931 }
5932
5933 int
5934 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5935                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5936 {
5937         struct ctl_lun *lun = CTL_LUN(ctsio);
5938         struct scsi_info_exceptions_page *pg;
5939         uint64_t t;
5940
5941         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5942
5943         pg = (struct scsi_info_exceptions_page *)page_ptr;
5944         mtx_lock(&lun->lun_lock);
5945         if (pg->info_flags & SIEP_FLAGS_TEST) {
5946                 lun->ie_asc = 0x5d;
5947                 lun->ie_ascq = 0xff;
5948                 if (pg->mrie == SIEP_MRIE_UA) {
5949                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
5950                         lun->ie_reported = 1;
5951                 } else {
5952                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5953                         lun->ie_reported = -1;
5954                 }
5955                 lun->ie_reportcnt = 1;
5956                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
5957                         lun->ie_reportcnt++;
5958                         t = scsi_4btoul(pg->interval_timer);
5959                         if (t == 0 || t == UINT32_MAX)
5960                                 t = 3000;  /* 5 min */
5961                         callout_reset(&lun->ie_callout, t * hz / 10,
5962                             ctl_ie_timer, lun);
5963                 }
5964         } else {
5965                 lun->ie_asc = 0;
5966                 lun->ie_ascq = 0;
5967                 lun->ie_reported = 1;
5968                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5969                 lun->ie_reportcnt = UINT32_MAX;
5970                 callout_stop(&lun->ie_callout);
5971         }
5972         mtx_unlock(&lun->lun_lock);
5973         return (CTL_RETVAL_COMPLETE);
5974 }
5975
5976 static int
5977 ctl_do_mode_select(union ctl_io *io)
5978 {
5979         struct ctl_lun *lun = CTL_LUN(io);
5980         struct scsi_mode_page_header *page_header;
5981         struct ctl_page_index *page_index;
5982         struct ctl_scsiio *ctsio;
5983         int page_len, page_len_offset, page_len_size;
5984         union ctl_modepage_info *modepage_info;
5985         uint16_t *len_left, *len_used;
5986         int retval, i;
5987
5988         ctsio = &io->scsiio;
5989         page_index = NULL;
5990         page_len = 0;
5991
5992         modepage_info = (union ctl_modepage_info *)
5993                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5994         len_left = &modepage_info->header.len_left;
5995         len_used = &modepage_info->header.len_used;
5996
5997 do_next_page:
5998
5999         page_header = (struct scsi_mode_page_header *)
6000                 (ctsio->kern_data_ptr + *len_used);
6001
6002         if (*len_left == 0) {
6003                 free(ctsio->kern_data_ptr, M_CTL);
6004                 ctl_set_success(ctsio);
6005                 ctl_done((union ctl_io *)ctsio);
6006                 return (CTL_RETVAL_COMPLETE);
6007         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6008
6009                 free(ctsio->kern_data_ptr, M_CTL);
6010                 ctl_set_param_len_error(ctsio);
6011                 ctl_done((union ctl_io *)ctsio);
6012                 return (CTL_RETVAL_COMPLETE);
6013
6014         } else if ((page_header->page_code & SMPH_SPF)
6015                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6016
6017                 free(ctsio->kern_data_ptr, M_CTL);
6018                 ctl_set_param_len_error(ctsio);
6019                 ctl_done((union ctl_io *)ctsio);
6020                 return (CTL_RETVAL_COMPLETE);
6021         }
6022
6023
6024         /*
6025          * XXX KDM should we do something with the block descriptor?
6026          */
6027         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6028                 page_index = &lun->mode_pages.index[i];
6029                 if (lun->be_lun->lun_type == T_DIRECT &&
6030                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6031                         continue;
6032                 if (lun->be_lun->lun_type == T_PROCESSOR &&
6033                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6034                         continue;
6035                 if (lun->be_lun->lun_type == T_CDROM &&
6036                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6037                         continue;
6038
6039                 if ((page_index->page_code & SMPH_PC_MASK) !=
6040                     (page_header->page_code & SMPH_PC_MASK))
6041                         continue;
6042
6043                 /*
6044                  * If neither page has a subpage code, then we've got a
6045                  * match.
6046                  */
6047                 if (((page_index->page_code & SMPH_SPF) == 0)
6048                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6049                         page_len = page_header->page_length;
6050                         break;
6051                 }
6052
6053                 /*
6054                  * If both pages have subpages, then the subpage numbers
6055                  * have to match.
6056                  */
6057                 if ((page_index->page_code & SMPH_SPF)
6058                   && (page_header->page_code & SMPH_SPF)) {
6059                         struct scsi_mode_page_header_sp *sph;
6060
6061                         sph = (struct scsi_mode_page_header_sp *)page_header;
6062                         if (page_index->subpage == sph->subpage) {
6063                                 page_len = scsi_2btoul(sph->page_length);
6064                                 break;
6065                         }
6066                 }
6067         }
6068
6069         /*
6070          * If we couldn't find the page, or if we don't have a mode select
6071          * handler for it, send back an error to the user.
6072          */
6073         if ((i >= CTL_NUM_MODE_PAGES)
6074          || (page_index->select_handler == NULL)) {
6075                 ctl_set_invalid_field(ctsio,
6076                                       /*sks_valid*/ 1,
6077                                       /*command*/ 0,
6078                                       /*field*/ *len_used,
6079                                       /*bit_valid*/ 0,
6080                                       /*bit*/ 0);
6081                 free(ctsio->kern_data_ptr, M_CTL);
6082                 ctl_done((union ctl_io *)ctsio);
6083                 return (CTL_RETVAL_COMPLETE);
6084         }
6085
6086         if (page_index->page_code & SMPH_SPF) {
6087                 page_len_offset = 2;
6088                 page_len_size = 2;
6089         } else {
6090                 page_len_size = 1;
6091                 page_len_offset = 1;
6092         }
6093
6094         /*
6095          * If the length the initiator gives us isn't the one we specify in
6096          * the mode page header, or if they didn't specify enough data in
6097          * the CDB to avoid truncating this page, kick out the request.
6098          */
6099         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6100                 ctl_set_invalid_field(ctsio,
6101                                       /*sks_valid*/ 1,
6102                                       /*command*/ 0,
6103                                       /*field*/ *len_used + page_len_offset,
6104                                       /*bit_valid*/ 0,
6105                                       /*bit*/ 0);
6106                 free(ctsio->kern_data_ptr, M_CTL);
6107                 ctl_done((union ctl_io *)ctsio);
6108                 return (CTL_RETVAL_COMPLETE);
6109         }
6110         if (*len_left < page_index->page_len) {
6111                 free(ctsio->kern_data_ptr, M_CTL);
6112                 ctl_set_param_len_error(ctsio);
6113                 ctl_done((union ctl_io *)ctsio);
6114                 return (CTL_RETVAL_COMPLETE);
6115         }
6116
6117         /*
6118          * Run through the mode page, checking to make sure that the bits
6119          * the user changed are actually legal for him to change.
6120          */
6121         for (i = 0; i < page_index->page_len; i++) {
6122                 uint8_t *user_byte, *change_mask, *current_byte;
6123                 int bad_bit;
6124                 int j;
6125
6126                 user_byte = (uint8_t *)page_header + i;
6127                 change_mask = page_index->page_data +
6128                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6129                 current_byte = page_index->page_data +
6130                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6131
6132                 /*
6133                  * Check to see whether the user set any bits in this byte
6134                  * that he is not allowed to set.
6135                  */
6136                 if ((*user_byte & ~(*change_mask)) ==
6137                     (*current_byte & ~(*change_mask)))
6138                         continue;
6139
6140                 /*
6141                  * Go through bit by bit to determine which one is illegal.
6142                  */
6143                 bad_bit = 0;
6144                 for (j = 7; j >= 0; j--) {
6145                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6146                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6147                                 bad_bit = i;
6148                                 break;
6149                         }
6150                 }
6151                 ctl_set_invalid_field(ctsio,
6152                                       /*sks_valid*/ 1,
6153                                       /*command*/ 0,
6154                                       /*field*/ *len_used + i,
6155                                       /*bit_valid*/ 1,
6156                                       /*bit*/ bad_bit);
6157                 free(ctsio->kern_data_ptr, M_CTL);
6158                 ctl_done((union ctl_io *)ctsio);
6159                 return (CTL_RETVAL_COMPLETE);
6160         }
6161
6162         /*
6163          * Decrement these before we call the page handler, since we may
6164          * end up getting called back one way or another before the handler
6165          * returns to this context.
6166          */
6167         *len_left -= page_index->page_len;
6168         *len_used += page_index->page_len;
6169
6170         retval = page_index->select_handler(ctsio, page_index,
6171                                             (uint8_t *)page_header);
6172
6173         /*
6174          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6175          * wait until this queued command completes to finish processing
6176          * the mode page.  If it returns anything other than
6177          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6178          * already set the sense information, freed the data pointer, and
6179          * completed the io for us.
6180          */
6181         if (retval != CTL_RETVAL_COMPLETE)
6182                 goto bailout_no_done;
6183
6184         /*
6185          * If the initiator sent us more than one page, parse the next one.
6186          */
6187         if (*len_left > 0)
6188                 goto do_next_page;
6189
6190         ctl_set_success(ctsio);
6191         free(ctsio->kern_data_ptr, M_CTL);
6192         ctl_done((union ctl_io *)ctsio);
6193
6194 bailout_no_done:
6195
6196         return (CTL_RETVAL_COMPLETE);
6197
6198 }
6199
6200 int
6201 ctl_mode_select(struct ctl_scsiio *ctsio)
6202 {
6203         struct ctl_lun *lun = CTL_LUN(ctsio);
6204         union ctl_modepage_info *modepage_info;
6205         int bd_len, i, header_size, param_len, pf, rtd, sp;
6206         uint32_t initidx;
6207
6208         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6209         switch (ctsio->cdb[0]) {
6210         case MODE_SELECT_6: {
6211                 struct scsi_mode_select_6 *cdb;
6212
6213                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6214
6215                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6216                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6217                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6218                 param_len = cdb->length;
6219                 header_size = sizeof(struct scsi_mode_header_6);
6220                 break;
6221         }
6222         case MODE_SELECT_10: {
6223                 struct scsi_mode_select_10 *cdb;
6224
6225                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6226
6227                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6228                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6229                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6230                 param_len = scsi_2btoul(cdb->length);
6231                 header_size = sizeof(struct scsi_mode_header_10);
6232                 break;
6233         }
6234         default:
6235                 ctl_set_invalid_opcode(ctsio);
6236                 ctl_done((union ctl_io *)ctsio);
6237                 return (CTL_RETVAL_COMPLETE);
6238         }
6239
6240         if (rtd) {
6241                 if (param_len != 0) {
6242                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 0,
6243                             /*command*/ 1, /*field*/ 0,
6244                             /*bit_valid*/ 0, /*bit*/ 0);
6245                         ctl_done((union ctl_io *)ctsio);
6246                         return (CTL_RETVAL_COMPLETE);
6247                 }
6248
6249                 /* Revert to defaults. */
6250                 ctl_init_page_index(lun);
6251                 mtx_lock(&lun->lun_lock);
6252                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6253                 mtx_unlock(&lun->lun_lock);
6254                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6255                         ctl_isc_announce_mode(lun, -1,
6256                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
6257                             lun->mode_pages.index[i].subpage);
6258                 }
6259                 ctl_set_success(ctsio);
6260                 ctl_done((union ctl_io *)ctsio);
6261                 return (CTL_RETVAL_COMPLETE);
6262         }
6263
6264         /*
6265          * From SPC-3:
6266          * "A parameter list length of zero indicates that the Data-Out Buffer
6267          * shall be empty. This condition shall not be considered as an error."
6268          */
6269         if (param_len == 0) {
6270                 ctl_set_success(ctsio);
6271                 ctl_done((union ctl_io *)ctsio);
6272                 return (CTL_RETVAL_COMPLETE);
6273         }
6274
6275         /*
6276          * Since we'll hit this the first time through, prior to
6277          * allocation, we don't need to free a data buffer here.
6278          */
6279         if (param_len < header_size) {
6280                 ctl_set_param_len_error(ctsio);
6281                 ctl_done((union ctl_io *)ctsio);
6282                 return (CTL_RETVAL_COMPLETE);
6283         }
6284
6285         /*
6286          * Allocate the data buffer and grab the user's data.  In theory,
6287          * we shouldn't have to sanity check the parameter list length here
6288          * because the maximum size is 64K.  We should be able to malloc
6289          * that much without too many problems.
6290          */
6291         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6292                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6293                 ctsio->kern_data_len = param_len;
6294                 ctsio->kern_total_len = param_len;
6295                 ctsio->kern_rel_offset = 0;
6296                 ctsio->kern_sg_entries = 0;
6297                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6298                 ctsio->be_move_done = ctl_config_move_done;
6299                 ctl_datamove((union ctl_io *)ctsio);
6300
6301                 return (CTL_RETVAL_COMPLETE);
6302         }
6303
6304         switch (ctsio->cdb[0]) {
6305         case MODE_SELECT_6: {
6306                 struct scsi_mode_header_6 *mh6;
6307
6308                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6309                 bd_len = mh6->blk_desc_len;
6310                 break;
6311         }
6312         case MODE_SELECT_10: {
6313                 struct scsi_mode_header_10 *mh10;
6314
6315                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6316                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6317                 break;
6318         }
6319         default:
6320                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6321         }
6322
6323         if (param_len < (header_size + bd_len)) {
6324                 free(ctsio->kern_data_ptr, M_CTL);
6325                 ctl_set_param_len_error(ctsio);
6326                 ctl_done((union ctl_io *)ctsio);
6327                 return (CTL_RETVAL_COMPLETE);
6328         }
6329
6330         /*
6331          * Set the IO_CONT flag, so that if this I/O gets passed to
6332          * ctl_config_write_done(), it'll get passed back to
6333          * ctl_do_mode_select() for further processing, or completion if
6334          * we're all done.
6335          */
6336         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6337         ctsio->io_cont = ctl_do_mode_select;
6338
6339         modepage_info = (union ctl_modepage_info *)
6340                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6341         memset(modepage_info, 0, sizeof(*modepage_info));
6342         modepage_info->header.len_left = param_len - header_size - bd_len;
6343         modepage_info->header.len_used = header_size + bd_len;
6344
6345         return (ctl_do_mode_select((union ctl_io *)ctsio));
6346 }
6347
6348 int
6349 ctl_mode_sense(struct ctl_scsiio *ctsio)
6350 {
6351         struct ctl_lun *lun = CTL_LUN(ctsio);
6352         int pc, page_code, dbd, llba, subpage;
6353         int alloc_len, page_len, header_len, total_len;
6354         struct scsi_mode_block_descr *block_desc;
6355         struct ctl_page_index *page_index;
6356
6357         dbd = 0;
6358         llba = 0;
6359         block_desc = NULL;
6360
6361         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6362
6363         switch (ctsio->cdb[0]) {
6364         case MODE_SENSE_6: {
6365                 struct scsi_mode_sense_6 *cdb;
6366
6367                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6368
6369                 header_len = sizeof(struct scsi_mode_hdr_6);
6370                 if (cdb->byte2 & SMS_DBD)
6371                         dbd = 1;
6372                 else
6373                         header_len += sizeof(struct scsi_mode_block_descr);
6374
6375                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6376                 page_code = cdb->page & SMS_PAGE_CODE;
6377                 subpage = cdb->subpage;
6378                 alloc_len = cdb->length;
6379                 break;
6380         }
6381         case MODE_SENSE_10: {
6382                 struct scsi_mode_sense_10 *cdb;
6383
6384                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6385
6386                 header_len = sizeof(struct scsi_mode_hdr_10);
6387
6388                 if (cdb->byte2 & SMS_DBD)
6389                         dbd = 1;
6390                 else
6391                         header_len += sizeof(struct scsi_mode_block_descr);
6392                 if (cdb->byte2 & SMS10_LLBAA)
6393                         llba = 1;
6394                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6395                 page_code = cdb->page & SMS_PAGE_CODE;
6396                 subpage = cdb->subpage;
6397                 alloc_len = scsi_2btoul(cdb->length);
6398                 break;
6399         }
6400         default:
6401                 ctl_set_invalid_opcode(ctsio);
6402                 ctl_done((union ctl_io *)ctsio);
6403                 return (CTL_RETVAL_COMPLETE);
6404                 break; /* NOTREACHED */
6405         }
6406
6407         /*
6408          * We have to make a first pass through to calculate the size of
6409          * the pages that match the user's query.  Then we allocate enough
6410          * memory to hold it, and actually copy the data into the buffer.
6411          */
6412         switch (page_code) {
6413         case SMS_ALL_PAGES_PAGE: {
6414                 u_int i;
6415
6416                 page_len = 0;
6417
6418                 /*
6419                  * At the moment, values other than 0 and 0xff here are
6420                  * reserved according to SPC-3.
6421                  */
6422                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6423                  && (subpage != SMS_SUBPAGE_ALL)) {
6424                         ctl_set_invalid_field(ctsio,
6425                                               /*sks_valid*/ 1,
6426                                               /*command*/ 1,
6427                                               /*field*/ 3,
6428                                               /*bit_valid*/ 0,
6429                                               /*bit*/ 0);
6430                         ctl_done((union ctl_io *)ctsio);
6431                         return (CTL_RETVAL_COMPLETE);
6432                 }
6433
6434                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6435                         page_index = &lun->mode_pages.index[i];
6436
6437                         /* Make sure the page is supported for this dev type */
6438                         if (lun->be_lun->lun_type == T_DIRECT &&
6439                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6440                                 continue;
6441                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6442                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6443                                 continue;
6444                         if (lun->be_lun->lun_type == T_CDROM &&
6445                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6446                                 continue;
6447
6448                         /*
6449                          * We don't use this subpage if the user didn't
6450                          * request all subpages.
6451                          */
6452                         if ((page_index->subpage != 0)
6453                          && (subpage == SMS_SUBPAGE_PAGE_0))
6454                                 continue;
6455
6456 #if 0
6457                         printf("found page %#x len %d\n",
6458                                page_index->page_code & SMPH_PC_MASK,
6459                                page_index->page_len);
6460 #endif
6461                         page_len += page_index->page_len;
6462                 }
6463                 break;
6464         }
6465         default: {
6466                 u_int i;
6467
6468                 page_len = 0;
6469
6470                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6471                         page_index = &lun->mode_pages.index[i];
6472
6473                         /* Make sure the page is supported for this dev type */
6474                         if (lun->be_lun->lun_type == T_DIRECT &&
6475                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6476                                 continue;
6477                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6478                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6479                                 continue;
6480                         if (lun->be_lun->lun_type == T_CDROM &&
6481                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6482                                 continue;
6483
6484                         /* Look for the right page code */
6485                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6486                                 continue;
6487
6488                         /* Look for the right subpage or the subpage wildcard*/
6489                         if ((page_index->subpage != subpage)
6490                          && (subpage != SMS_SUBPAGE_ALL))
6491                                 continue;
6492
6493 #if 0
6494                         printf("found page %#x len %d\n",
6495                                page_index->page_code & SMPH_PC_MASK,
6496                                page_index->page_len);
6497 #endif
6498
6499                         page_len += page_index->page_len;
6500                 }
6501
6502                 if (page_len == 0) {
6503                         ctl_set_invalid_field(ctsio,
6504                                               /*sks_valid*/ 1,
6505                                               /*command*/ 1,
6506                                               /*field*/ 2,
6507                                               /*bit_valid*/ 1,
6508                                               /*bit*/ 5);
6509                         ctl_done((union ctl_io *)ctsio);
6510                         return (CTL_RETVAL_COMPLETE);
6511                 }
6512                 break;
6513         }
6514         }
6515
6516         total_len = header_len + page_len;
6517 #if 0
6518         printf("header_len = %d, page_len = %d, total_len = %d\n",
6519                header_len, page_len, total_len);
6520 #endif
6521
6522         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6523         ctsio->kern_sg_entries = 0;
6524         ctsio->kern_rel_offset = 0;
6525         ctsio->kern_data_len = min(total_len, alloc_len);
6526         ctsio->kern_total_len = ctsio->kern_data_len;
6527
6528         switch (ctsio->cdb[0]) {
6529         case MODE_SENSE_6: {
6530                 struct scsi_mode_hdr_6 *header;
6531
6532                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6533
6534                 header->datalen = MIN(total_len - 1, 254);
6535                 if (lun->be_lun->lun_type == T_DIRECT) {
6536                         header->dev_specific = 0x10; /* DPOFUA */
6537                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6538                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6539                                 header->dev_specific |= 0x80; /* WP */
6540                 }
6541                 if (dbd)
6542                         header->block_descr_len = 0;
6543                 else
6544                         header->block_descr_len =
6545                                 sizeof(struct scsi_mode_block_descr);
6546                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6547                 break;
6548         }
6549         case MODE_SENSE_10: {
6550                 struct scsi_mode_hdr_10 *header;
6551                 int datalen;
6552
6553                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6554
6555                 datalen = MIN(total_len - 2, 65533);
6556                 scsi_ulto2b(datalen, header->datalen);
6557                 if (lun->be_lun->lun_type == T_DIRECT) {
6558                         header->dev_specific = 0x10; /* DPOFUA */
6559                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6560                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6561                                 header->dev_specific |= 0x80; /* WP */
6562                 }
6563                 if (dbd)
6564                         scsi_ulto2b(0, header->block_descr_len);
6565                 else
6566                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6567                                     header->block_descr_len);
6568                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6569                 break;
6570         }
6571         default:
6572                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6573         }
6574
6575         /*
6576          * If we've got a disk, use its blocksize in the block
6577          * descriptor.  Otherwise, just set it to 0.
6578          */
6579         if (dbd == 0) {
6580                 if (lun->be_lun->lun_type == T_DIRECT)
6581                         scsi_ulto3b(lun->be_lun->blocksize,
6582                                     block_desc->block_len);
6583                 else
6584                         scsi_ulto3b(0, block_desc->block_len);
6585         }
6586
6587         switch (page_code) {
6588         case SMS_ALL_PAGES_PAGE: {
6589                 int i, data_used;
6590
6591                 data_used = header_len;
6592                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6593                         struct ctl_page_index *page_index;
6594
6595                         page_index = &lun->mode_pages.index[i];
6596                         if (lun->be_lun->lun_type == T_DIRECT &&
6597                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6598                                 continue;
6599                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6600                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6601                                 continue;
6602                         if (lun->be_lun->lun_type == T_CDROM &&
6603                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6604                                 continue;
6605
6606                         /*
6607                          * We don't use this subpage if the user didn't
6608                          * request all subpages.  We already checked (above)
6609                          * to make sure the user only specified a subpage
6610                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6611                          */
6612                         if ((page_index->subpage != 0)
6613                          && (subpage == SMS_SUBPAGE_PAGE_0))
6614                                 continue;
6615
6616                         /*
6617                          * Call the handler, if it exists, to update the
6618                          * page to the latest values.
6619                          */
6620                         if (page_index->sense_handler != NULL)
6621                                 page_index->sense_handler(ctsio, page_index,pc);
6622
6623                         memcpy(ctsio->kern_data_ptr + data_used,
6624                                page_index->page_data +
6625                                (page_index->page_len * pc),
6626                                page_index->page_len);
6627                         data_used += page_index->page_len;
6628                 }
6629                 break;
6630         }
6631         default: {
6632                 int i, data_used;
6633
6634                 data_used = header_len;
6635
6636                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6637                         struct ctl_page_index *page_index;
6638
6639                         page_index = &lun->mode_pages.index[i];
6640
6641                         /* Look for the right page code */
6642                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6643                                 continue;
6644
6645                         /* Look for the right subpage or the subpage wildcard*/
6646                         if ((page_index->subpage != subpage)
6647                          && (subpage != SMS_SUBPAGE_ALL))
6648                                 continue;
6649
6650                         /* Make sure the page is supported for this dev type */
6651                         if (lun->be_lun->lun_type == T_DIRECT &&
6652                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6653                                 continue;
6654                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6655                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6656                                 continue;
6657                         if (lun->be_lun->lun_type == T_CDROM &&
6658                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6659                                 continue;
6660
6661                         /*
6662                          * Call the handler, if it exists, to update the
6663                          * page to the latest values.
6664                          */
6665                         if (page_index->sense_handler != NULL)
6666                                 page_index->sense_handler(ctsio, page_index,pc);
6667
6668                         memcpy(ctsio->kern_data_ptr + data_used,
6669                                page_index->page_data +
6670                                (page_index->page_len * pc),
6671                                page_index->page_len);
6672                         data_used += page_index->page_len;
6673                 }
6674                 break;
6675         }
6676         }
6677
6678         ctl_set_success(ctsio);
6679         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6680         ctsio->be_move_done = ctl_config_move_done;
6681         ctl_datamove((union ctl_io *)ctsio);
6682         return (CTL_RETVAL_COMPLETE);
6683 }
6684
6685 int
6686 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6687                                struct ctl_page_index *page_index,
6688                                int pc)
6689 {
6690         struct ctl_lun *lun = CTL_LUN(ctsio);
6691         struct scsi_log_param_header *phdr;
6692         uint8_t *data;
6693         uint64_t val;
6694
6695         data = page_index->page_data;
6696
6697         if (lun->backend->lun_attr != NULL &&
6698             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6699              != UINT64_MAX) {
6700                 phdr = (struct scsi_log_param_header *)data;
6701                 scsi_ulto2b(0x0001, phdr->param_code);
6702                 phdr->param_control = SLP_LBIN | SLP_LP;
6703                 phdr->param_len = 8;
6704                 data = (uint8_t *)(phdr + 1);
6705                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6706                 data[4] = 0x02; /* per-pool */
6707                 data += phdr->param_len;
6708         }
6709
6710         if (lun->backend->lun_attr != NULL &&
6711             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6712              != UINT64_MAX) {
6713                 phdr = (struct scsi_log_param_header *)data;
6714                 scsi_ulto2b(0x0002, phdr->param_code);
6715                 phdr->param_control = SLP_LBIN | SLP_LP;
6716                 phdr->param_len = 8;
6717                 data = (uint8_t *)(phdr + 1);
6718                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6719                 data[4] = 0x01; /* per-LUN */
6720                 data += phdr->param_len;
6721         }
6722
6723         if (lun->backend->lun_attr != NULL &&
6724             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6725              != UINT64_MAX) {
6726                 phdr = (struct scsi_log_param_header *)data;
6727                 scsi_ulto2b(0x00f1, phdr->param_code);
6728                 phdr->param_control = SLP_LBIN | SLP_LP;
6729                 phdr->param_len = 8;
6730                 data = (uint8_t *)(phdr + 1);
6731                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6732                 data[4] = 0x02; /* per-pool */
6733                 data += phdr->param_len;
6734         }
6735
6736         if (lun->backend->lun_attr != NULL &&
6737             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6738              != UINT64_MAX) {
6739                 phdr = (struct scsi_log_param_header *)data;
6740                 scsi_ulto2b(0x00f2, phdr->param_code);
6741                 phdr->param_control = SLP_LBIN | SLP_LP;
6742                 phdr->param_len = 8;
6743                 data = (uint8_t *)(phdr + 1);
6744                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6745                 data[4] = 0x02; /* per-pool */
6746                 data += phdr->param_len;
6747         }
6748
6749         page_index->page_len = data - page_index->page_data;
6750         return (0);
6751 }
6752
6753 int
6754 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6755                                struct ctl_page_index *page_index,
6756                                int pc)
6757 {
6758         struct ctl_lun *lun = CTL_LUN(ctsio);
6759         struct stat_page *data;
6760         struct bintime *t;
6761
6762         data = (struct stat_page *)page_index->page_data;
6763
6764         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6765         data->sap.hdr.param_control = SLP_LBIN;
6766         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6767             sizeof(struct scsi_log_param_header);
6768         scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
6769             data->sap.read_num);
6770         scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
6771             data->sap.write_num);
6772         if (lun->be_lun->blocksize > 0) {
6773                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
6774                     lun->be_lun->blocksize, data->sap.recvieved_lba);
6775                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
6776                     lun->be_lun->blocksize, data->sap.transmitted_lba);
6777         }
6778         t = &lun->stats.time[CTL_STATS_READ];
6779         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6780             data->sap.read_int);
6781         t = &lun->stats.time[CTL_STATS_WRITE];
6782         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6783             data->sap.write_int);
6784         scsi_u64to8b(0, data->sap.weighted_num);
6785         scsi_u64to8b(0, data->sap.weighted_int);
6786         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6787         data->it.hdr.param_control = SLP_LBIN;
6788         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6789             sizeof(struct scsi_log_param_header);
6790 #ifdef CTL_TIME_IO
6791         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6792 #endif
6793         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6794         data->it.hdr.param_control = SLP_LBIN;
6795         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6796             sizeof(struct scsi_log_param_header);
6797         scsi_ulto4b(3, data->ti.exponent);
6798         scsi_ulto4b(1, data->ti.integer);
6799         return (0);
6800 }
6801
6802 int
6803 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6804                                struct ctl_page_index *page_index,
6805                                int pc)
6806 {
6807         struct ctl_lun *lun = CTL_LUN(ctsio);
6808         struct scsi_log_informational_exceptions *data;
6809
6810         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6811
6812         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6813         data->hdr.param_control = SLP_LBIN;
6814         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6815             sizeof(struct scsi_log_param_header);
6816         data->ie_asc = lun->ie_asc;
6817         data->ie_ascq = lun->ie_ascq;
6818         data->temperature = 0xff;
6819         return (0);
6820 }
6821
6822 int
6823 ctl_log_sense(struct ctl_scsiio *ctsio)
6824 {
6825         struct ctl_lun *lun = CTL_LUN(ctsio);
6826         int i, pc, page_code, subpage;
6827         int alloc_len, total_len;
6828         struct ctl_page_index *page_index;
6829         struct scsi_log_sense *cdb;
6830         struct scsi_log_header *header;
6831
6832         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6833
6834         cdb = (struct scsi_log_sense *)ctsio->cdb;
6835         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6836         page_code = cdb->page & SLS_PAGE_CODE;
6837         subpage = cdb->subpage;
6838         alloc_len = scsi_2btoul(cdb->length);
6839
6840         page_index = NULL;
6841         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6842                 page_index = &lun->log_pages.index[i];
6843
6844                 /* Look for the right page code */
6845                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6846                         continue;
6847
6848                 /* Look for the right subpage or the subpage wildcard*/
6849                 if (page_index->subpage != subpage)
6850                         continue;
6851
6852                 break;
6853         }
6854         if (i >= CTL_NUM_LOG_PAGES) {
6855                 ctl_set_invalid_field(ctsio,
6856                                       /*sks_valid*/ 1,
6857                                       /*command*/ 1,
6858                                       /*field*/ 2,
6859                                       /*bit_valid*/ 0,
6860                                       /*bit*/ 0);
6861                 ctl_done((union ctl_io *)ctsio);
6862                 return (CTL_RETVAL_COMPLETE);
6863         }
6864
6865         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6866
6867         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6868         ctsio->kern_sg_entries = 0;
6869         ctsio->kern_rel_offset = 0;
6870         ctsio->kern_data_len = min(total_len, alloc_len);
6871         ctsio->kern_total_len = ctsio->kern_data_len;
6872
6873         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6874         header->page = page_index->page_code;
6875         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6876                 header->page |= SL_DS;
6877         if (page_index->subpage) {
6878                 header->page |= SL_SPF;
6879                 header->subpage = page_index->subpage;
6880         }
6881         scsi_ulto2b(page_index->page_len, header->datalen);
6882
6883         /*
6884          * Call the handler, if it exists, to update the
6885          * page to the latest values.
6886          */
6887         if (page_index->sense_handler != NULL)
6888                 page_index->sense_handler(ctsio, page_index, pc);
6889
6890         memcpy(header + 1, page_index->page_data, page_index->page_len);
6891
6892         ctl_set_success(ctsio);
6893         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6894         ctsio->be_move_done = ctl_config_move_done;
6895         ctl_datamove((union ctl_io *)ctsio);
6896         return (CTL_RETVAL_COMPLETE);
6897 }
6898
6899 int
6900 ctl_read_capacity(struct ctl_scsiio *ctsio)
6901 {
6902         struct ctl_lun *lun = CTL_LUN(ctsio);
6903         struct scsi_read_capacity *cdb;
6904         struct scsi_read_capacity_data *data;
6905         uint32_t lba;
6906
6907         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6908
6909         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6910
6911         lba = scsi_4btoul(cdb->addr);
6912         if (((cdb->pmi & SRC_PMI) == 0)
6913          && (lba != 0)) {
6914                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6915                                       /*sks_valid*/ 1,
6916                                       /*command*/ 1,
6917                                       /*field*/ 2,
6918                                       /*bit_valid*/ 0,
6919                                       /*bit*/ 0);
6920                 ctl_done((union ctl_io *)ctsio);
6921                 return (CTL_RETVAL_COMPLETE);
6922         }
6923
6924         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6925         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6926         ctsio->kern_data_len = sizeof(*data);
6927         ctsio->kern_total_len = sizeof(*data);
6928         ctsio->kern_rel_offset = 0;
6929         ctsio->kern_sg_entries = 0;
6930
6931         /*
6932          * If the maximum LBA is greater than 0xfffffffe, the user must
6933          * issue a SERVICE ACTION IN (16) command, with the read capacity
6934          * serivce action set.
6935          */
6936         if (lun->be_lun->maxlba > 0xfffffffe)
6937                 scsi_ulto4b(0xffffffff, data->addr);
6938         else
6939                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6940
6941         /*
6942          * XXX KDM this may not be 512 bytes...
6943          */
6944         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6945
6946         ctl_set_success(ctsio);
6947         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6948         ctsio->be_move_done = ctl_config_move_done;
6949         ctl_datamove((union ctl_io *)ctsio);
6950         return (CTL_RETVAL_COMPLETE);
6951 }
6952
6953 int
6954 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6955 {
6956         struct ctl_lun *lun = CTL_LUN(ctsio);
6957         struct scsi_read_capacity_16 *cdb;
6958         struct scsi_read_capacity_data_long *data;
6959         uint64_t lba;
6960         uint32_t alloc_len;
6961
6962         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6963
6964         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6965
6966         alloc_len = scsi_4btoul(cdb->alloc_len);
6967         lba = scsi_8btou64(cdb->addr);
6968
6969         if ((cdb->reladr & SRC16_PMI)
6970          && (lba != 0)) {
6971                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6972                                       /*sks_valid*/ 1,
6973                                       /*command*/ 1,
6974                                       /*field*/ 2,
6975                                       /*bit_valid*/ 0,
6976                                       /*bit*/ 0);
6977                 ctl_done((union ctl_io *)ctsio);
6978                 return (CTL_RETVAL_COMPLETE);
6979         }
6980
6981         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6982         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6983         ctsio->kern_rel_offset = 0;
6984         ctsio->kern_sg_entries = 0;
6985         ctsio->kern_data_len = min(sizeof(*data), alloc_len);
6986         ctsio->kern_total_len = ctsio->kern_data_len;
6987
6988         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6989         /* XXX KDM this may not be 512 bytes... */
6990         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6991         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6992         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6993         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6994                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6995
6996         ctl_set_success(ctsio);
6997         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6998         ctsio->be_move_done = ctl_config_move_done;
6999         ctl_datamove((union ctl_io *)ctsio);
7000         return (CTL_RETVAL_COMPLETE);
7001 }
7002
7003 int
7004 ctl_get_lba_status(struct ctl_scsiio *ctsio)
7005 {
7006         struct ctl_lun *lun = CTL_LUN(ctsio);
7007         struct scsi_get_lba_status *cdb;
7008         struct scsi_get_lba_status_data *data;
7009         struct ctl_lba_len_flags *lbalen;
7010         uint64_t lba;
7011         uint32_t alloc_len, total_len;
7012         int retval;
7013
7014         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7015
7016         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7017         lba = scsi_8btou64(cdb->addr);
7018         alloc_len = scsi_4btoul(cdb->alloc_len);
7019
7020         if (lba > lun->be_lun->maxlba) {
7021                 ctl_set_lba_out_of_range(ctsio, lba);
7022                 ctl_done((union ctl_io *)ctsio);
7023                 return (CTL_RETVAL_COMPLETE);
7024         }
7025
7026         total_len = sizeof(*data) + sizeof(data->descr[0]);
7027         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7028         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7029         ctsio->kern_rel_offset = 0;
7030         ctsio->kern_sg_entries = 0;
7031         ctsio->kern_data_len = min(total_len, alloc_len);
7032         ctsio->kern_total_len = ctsio->kern_data_len;
7033
7034         /* Fill dummy data in case backend can't tell anything. */
7035         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7036         scsi_u64to8b(lba, data->descr[0].addr);
7037         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7038             data->descr[0].length);
7039         data->descr[0].status = 0; /* Mapped or unknown. */
7040
7041         ctl_set_success(ctsio);
7042         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7043         ctsio->be_move_done = ctl_config_move_done;
7044
7045         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7046         lbalen->lba = lba;
7047         lbalen->len = total_len;
7048         lbalen->flags = 0;
7049         retval = lun->backend->config_read((union ctl_io *)ctsio);
7050         return (retval);
7051 }
7052
7053 int
7054 ctl_read_defect(struct ctl_scsiio *ctsio)
7055 {
7056         struct scsi_read_defect_data_10 *ccb10;
7057         struct scsi_read_defect_data_12 *ccb12;
7058         struct scsi_read_defect_data_hdr_10 *data10;
7059         struct scsi_read_defect_data_hdr_12 *data12;
7060         uint32_t alloc_len, data_len;
7061         uint8_t format;
7062
7063         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7064
7065         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7066                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7067                 format = ccb10->format;
7068                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7069                 data_len = sizeof(*data10);
7070         } else {
7071                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7072                 format = ccb12->format;
7073                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7074                 data_len = sizeof(*data12);
7075         }
7076         if (alloc_len == 0) {
7077                 ctl_set_success(ctsio);
7078                 ctl_done((union ctl_io *)ctsio);
7079                 return (CTL_RETVAL_COMPLETE);
7080         }
7081
7082         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7083         ctsio->kern_rel_offset = 0;
7084         ctsio->kern_sg_entries = 0;
7085         ctsio->kern_data_len = min(data_len, alloc_len);
7086         ctsio->kern_total_len = ctsio->kern_data_len;
7087
7088         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7089                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7090                     ctsio->kern_data_ptr;
7091                 data10->format = format;
7092                 scsi_ulto2b(0, data10->length);
7093         } else {
7094                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7095                     ctsio->kern_data_ptr;
7096                 data12->format = format;
7097                 scsi_ulto2b(0, data12->generation);
7098                 scsi_ulto4b(0, data12->length);
7099         }
7100
7101         ctl_set_success(ctsio);
7102         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7103         ctsio->be_move_done = ctl_config_move_done;
7104         ctl_datamove((union ctl_io *)ctsio);
7105         return (CTL_RETVAL_COMPLETE);
7106 }
7107
7108 int
7109 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7110 {
7111         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7112         struct ctl_lun *lun = CTL_LUN(ctsio);
7113         struct scsi_maintenance_in *cdb;
7114         int retval;
7115         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7116         int num_ha_groups, num_target_ports, shared_group;
7117         struct ctl_port *port;
7118         struct scsi_target_group_data *rtg_ptr;
7119         struct scsi_target_group_data_extended *rtg_ext_ptr;
7120         struct scsi_target_port_group_descriptor *tpg_desc;
7121
7122         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7123
7124         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7125         retval = CTL_RETVAL_COMPLETE;
7126
7127         switch (cdb->byte2 & STG_PDF_MASK) {
7128         case STG_PDF_LENGTH:
7129                 ext = 0;
7130                 break;
7131         case STG_PDF_EXTENDED:
7132                 ext = 1;
7133                 break;
7134         default:
7135                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7136                                       /*sks_valid*/ 1,
7137                                       /*command*/ 1,
7138                                       /*field*/ 2,
7139                                       /*bit_valid*/ 1,
7140                                       /*bit*/ 5);
7141                 ctl_done((union ctl_io *)ctsio);
7142                 return(retval);
7143         }
7144
7145         num_target_ports = 0;
7146         shared_group = (softc->is_single != 0);
7147         mtx_lock(&softc->ctl_lock);
7148         STAILQ_FOREACH(port, &softc->port_list, links) {
7149                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7150                         continue;
7151                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7152                         continue;
7153                 num_target_ports++;
7154                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7155                         shared_group = 1;
7156         }
7157         mtx_unlock(&softc->ctl_lock);
7158         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7159
7160         if (ext)
7161                 total_len = sizeof(struct scsi_target_group_data_extended);
7162         else
7163                 total_len = sizeof(struct scsi_target_group_data);
7164         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7165                 (shared_group + num_ha_groups) +
7166             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7167
7168         alloc_len = scsi_4btoul(cdb->length);
7169
7170         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7171         ctsio->kern_sg_entries = 0;
7172         ctsio->kern_rel_offset = 0;
7173         ctsio->kern_data_len = min(total_len, alloc_len);
7174         ctsio->kern_total_len = ctsio->kern_data_len;
7175
7176         if (ext) {
7177                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7178                     ctsio->kern_data_ptr;
7179                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7180                 rtg_ext_ptr->format_type = 0x10;
7181                 rtg_ext_ptr->implicit_transition_time = 0;
7182                 tpg_desc = &rtg_ext_ptr->groups[0];
7183         } else {
7184                 rtg_ptr = (struct scsi_target_group_data *)
7185                     ctsio->kern_data_ptr;
7186                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7187                 tpg_desc = &rtg_ptr->groups[0];
7188         }
7189
7190         mtx_lock(&softc->ctl_lock);
7191         pg = softc->port_min / softc->port_cnt;
7192         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7193                 /* Some shelf is known to be primary. */
7194                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7195                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7196                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7197                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7198                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7199                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7200                 else
7201                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7202                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7203                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7204                 } else {
7205                         ts = os;
7206                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7207                 }
7208         } else {
7209                 /* No known primary shelf. */
7210                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7211                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7212                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7213                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7214                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7215                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7216                 } else {
7217                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7218                 }
7219         }
7220         if (shared_group) {
7221                 tpg_desc->pref_state = ts;
7222                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7223                     TPG_U_SUP | TPG_T_SUP;
7224                 scsi_ulto2b(1, tpg_desc->target_port_group);
7225                 tpg_desc->status = TPG_IMPLICIT;
7226                 pc = 0;
7227                 STAILQ_FOREACH(port, &softc->port_list, links) {
7228                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7229                                 continue;
7230                         if (!softc->is_single &&
7231                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7232                                 continue;
7233                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7234                                 continue;
7235                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7236                             relative_target_port_identifier);
7237                         pc++;
7238                 }
7239                 tpg_desc->target_port_count = pc;
7240                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7241                     &tpg_desc->descriptors[pc];
7242         }
7243         for (g = 0; g < num_ha_groups; g++) {
7244                 tpg_desc->pref_state = (g == pg) ? ts : os;
7245                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7246                     TPG_U_SUP | TPG_T_SUP;
7247                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7248                 tpg_desc->status = TPG_IMPLICIT;
7249                 pc = 0;
7250                 STAILQ_FOREACH(port, &softc->port_list, links) {
7251                         if (port->targ_port < g * softc->port_cnt ||
7252                             port->targ_port >= (g + 1) * softc->port_cnt)
7253                                 continue;
7254                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7255                                 continue;
7256                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7257                                 continue;
7258                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7259                                 continue;
7260                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7261                             relative_target_port_identifier);
7262                         pc++;
7263                 }
7264                 tpg_desc->target_port_count = pc;
7265                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7266                     &tpg_desc->descriptors[pc];
7267         }
7268         mtx_unlock(&softc->ctl_lock);
7269
7270         ctl_set_success(ctsio);
7271         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7272         ctsio->be_move_done = ctl_config_move_done;
7273         ctl_datamove((union ctl_io *)ctsio);
7274         return(retval);
7275 }
7276
7277 int
7278 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7279 {
7280         struct ctl_lun *lun = CTL_LUN(ctsio);
7281         struct scsi_report_supported_opcodes *cdb;
7282         const struct ctl_cmd_entry *entry, *sentry;
7283         struct scsi_report_supported_opcodes_all *all;
7284         struct scsi_report_supported_opcodes_descr *descr;
7285         struct scsi_report_supported_opcodes_one *one;
7286         int retval;
7287         int alloc_len, total_len;
7288         int opcode, service_action, i, j, num;
7289
7290         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7291
7292         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7293         retval = CTL_RETVAL_COMPLETE;
7294
7295         opcode = cdb->requested_opcode;
7296         service_action = scsi_2btoul(cdb->requested_service_action);
7297         switch (cdb->options & RSO_OPTIONS_MASK) {
7298         case RSO_OPTIONS_ALL:
7299                 num = 0;
7300                 for (i = 0; i < 256; i++) {
7301                         entry = &ctl_cmd_table[i];
7302                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7303                                 for (j = 0; j < 32; j++) {
7304                                         sentry = &((const struct ctl_cmd_entry *)
7305                                             entry->execute)[j];
7306                                         if (ctl_cmd_applicable(
7307                                             lun->be_lun->lun_type, sentry))
7308                                                 num++;
7309                                 }
7310                         } else {
7311                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7312                                     entry))
7313                                         num++;
7314                         }
7315                 }
7316                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7317                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7318                 break;
7319         case RSO_OPTIONS_OC:
7320                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7321                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7322                                               /*sks_valid*/ 1,
7323                                               /*command*/ 1,
7324                                               /*field*/ 2,
7325                                               /*bit_valid*/ 1,
7326                                               /*bit*/ 2);
7327                         ctl_done((union ctl_io *)ctsio);
7328                         return (CTL_RETVAL_COMPLETE);
7329                 }
7330                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7331                 break;
7332         case RSO_OPTIONS_OC_SA:
7333                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7334                     service_action >= 32) {
7335                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7336                                               /*sks_valid*/ 1,
7337                                               /*command*/ 1,
7338                                               /*field*/ 2,
7339                                               /*bit_valid*/ 1,
7340                                               /*bit*/ 2);
7341                         ctl_done((union ctl_io *)ctsio);
7342                         return (CTL_RETVAL_COMPLETE);
7343                 }
7344                 /* FALLTHROUGH */
7345         case RSO_OPTIONS_OC_ASA:
7346                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7347                 break;
7348         default:
7349                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7350                                       /*sks_valid*/ 1,
7351                                       /*command*/ 1,
7352                                       /*field*/ 2,
7353                                       /*bit_valid*/ 1,
7354                                       /*bit*/ 2);
7355                 ctl_done((union ctl_io *)ctsio);
7356                 return (CTL_RETVAL_COMPLETE);
7357         }
7358
7359         alloc_len = scsi_4btoul(cdb->length);
7360
7361         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7362         ctsio->kern_sg_entries = 0;
7363         ctsio->kern_rel_offset = 0;
7364         ctsio->kern_data_len = min(total_len, alloc_len);
7365         ctsio->kern_total_len = ctsio->kern_data_len;
7366
7367         switch (cdb->options & RSO_OPTIONS_MASK) {
7368         case RSO_OPTIONS_ALL:
7369                 all = (struct scsi_report_supported_opcodes_all *)
7370                     ctsio->kern_data_ptr;
7371                 num = 0;
7372                 for (i = 0; i < 256; i++) {
7373                         entry = &ctl_cmd_table[i];
7374                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7375                                 for (j = 0; j < 32; j++) {
7376                                         sentry = &((const struct ctl_cmd_entry *)
7377                                             entry->execute)[j];
7378                                         if (!ctl_cmd_applicable(
7379                                             lun->be_lun->lun_type, sentry))
7380                                                 continue;
7381                                         descr = &all->descr[num++];
7382                                         descr->opcode = i;
7383                                         scsi_ulto2b(j, descr->service_action);
7384                                         descr->flags = RSO_SERVACTV;
7385                                         scsi_ulto2b(sentry->length,
7386                                             descr->cdb_length);
7387                                 }
7388                         } else {
7389                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7390                                     entry))
7391                                         continue;
7392                                 descr = &all->descr[num++];
7393                                 descr->opcode = i;
7394                                 scsi_ulto2b(0, descr->service_action);
7395                                 descr->flags = 0;
7396                                 scsi_ulto2b(entry->length, descr->cdb_length);
7397                         }
7398                 }
7399                 scsi_ulto4b(
7400                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7401                     all->length);
7402                 break;
7403         case RSO_OPTIONS_OC:
7404                 one = (struct scsi_report_supported_opcodes_one *)
7405                     ctsio->kern_data_ptr;
7406                 entry = &ctl_cmd_table[opcode];
7407                 goto fill_one;
7408         case RSO_OPTIONS_OC_SA:
7409                 one = (struct scsi_report_supported_opcodes_one *)
7410                     ctsio->kern_data_ptr;
7411                 entry = &ctl_cmd_table[opcode];
7412                 entry = &((const struct ctl_cmd_entry *)
7413                     entry->execute)[service_action];
7414 fill_one:
7415                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7416                         one->support = 3;
7417                         scsi_ulto2b(entry->length, one->cdb_length);
7418                         one->cdb_usage[0] = opcode;
7419                         memcpy(&one->cdb_usage[1], entry->usage,
7420                             entry->length - 1);
7421                 } else
7422                         one->support = 1;
7423                 break;
7424         case RSO_OPTIONS_OC_ASA:
7425                 one = (struct scsi_report_supported_opcodes_one *)
7426                     ctsio->kern_data_ptr;
7427                 entry = &ctl_cmd_table[opcode];
7428                 if (entry->flags & CTL_CMD_FLAG_SA5) {
7429                         entry = &((const struct ctl_cmd_entry *)
7430                             entry->execute)[service_action];
7431                 } else if (service_action != 0) {
7432                         one->support = 1;
7433                         break;
7434                 }
7435                 goto fill_one;
7436         }
7437
7438         ctl_set_success(ctsio);
7439         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7440         ctsio->be_move_done = ctl_config_move_done;
7441         ctl_datamove((union ctl_io *)ctsio);
7442         return(retval);
7443 }
7444
7445 int
7446 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7447 {
7448         struct scsi_report_supported_tmf *cdb;
7449         struct scsi_report_supported_tmf_ext_data *data;
7450         int retval;
7451         int alloc_len, total_len;
7452
7453         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7454
7455         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7456
7457         retval = CTL_RETVAL_COMPLETE;
7458
7459         if (cdb->options & RST_REPD)
7460                 total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7461         else
7462                 total_len = sizeof(struct scsi_report_supported_tmf_data);
7463         alloc_len = scsi_4btoul(cdb->length);
7464
7465         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7466         ctsio->kern_sg_entries = 0;
7467         ctsio->kern_rel_offset = 0;
7468         ctsio->kern_data_len = min(total_len, alloc_len);
7469         ctsio->kern_total_len = ctsio->kern_data_len;
7470
7471         data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7472         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7473             RST_TRS;
7474         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7475         data->length = total_len - 4;
7476
7477         ctl_set_success(ctsio);
7478         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7479         ctsio->be_move_done = ctl_config_move_done;
7480         ctl_datamove((union ctl_io *)ctsio);
7481         return (retval);
7482 }
7483
7484 int
7485 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7486 {
7487         struct scsi_report_timestamp *cdb;
7488         struct scsi_report_timestamp_data *data;
7489         struct timeval tv;
7490         int64_t timestamp;
7491         int retval;
7492         int alloc_len, total_len;
7493
7494         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7495
7496         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7497
7498         retval = CTL_RETVAL_COMPLETE;
7499
7500         total_len = sizeof(struct scsi_report_timestamp_data);
7501         alloc_len = scsi_4btoul(cdb->length);
7502
7503         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7504         ctsio->kern_sg_entries = 0;
7505         ctsio->kern_rel_offset = 0;
7506         ctsio->kern_data_len = min(total_len, alloc_len);
7507         ctsio->kern_total_len = ctsio->kern_data_len;
7508
7509         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7510         scsi_ulto2b(sizeof(*data) - 2, data->length);
7511         data->origin = RTS_ORIG_OUTSIDE;
7512         getmicrotime(&tv);
7513         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7514         scsi_ulto4b(timestamp >> 16, data->timestamp);
7515         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7516
7517         ctl_set_success(ctsio);
7518         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7519         ctsio->be_move_done = ctl_config_move_done;
7520         ctl_datamove((union ctl_io *)ctsio);
7521         return (retval);
7522 }
7523
7524 int
7525 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7526 {
7527         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7528         struct ctl_lun *lun = CTL_LUN(ctsio);
7529         struct scsi_per_res_in *cdb;
7530         int alloc_len, total_len = 0;
7531         /* struct scsi_per_res_in_rsrv in_data; */
7532         uint64_t key;
7533
7534         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7535
7536         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7537
7538         alloc_len = scsi_2btoul(cdb->length);
7539
7540 retry:
7541         mtx_lock(&lun->lun_lock);
7542         switch (cdb->action) {
7543         case SPRI_RK: /* read keys */
7544                 total_len = sizeof(struct scsi_per_res_in_keys) +
7545                         lun->pr_key_count *
7546                         sizeof(struct scsi_per_res_key);
7547                 break;
7548         case SPRI_RR: /* read reservation */
7549                 if (lun->flags & CTL_LUN_PR_RESERVED)
7550                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7551                 else
7552                         total_len = sizeof(struct scsi_per_res_in_header);
7553                 break;
7554         case SPRI_RC: /* report capabilities */
7555                 total_len = sizeof(struct scsi_per_res_cap);
7556                 break;
7557         case SPRI_RS: /* read full status */
7558                 total_len = sizeof(struct scsi_per_res_in_header) +
7559                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7560                     lun->pr_key_count;
7561                 break;
7562         default:
7563                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7564         }
7565         mtx_unlock(&lun->lun_lock);
7566
7567         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7568         ctsio->kern_rel_offset = 0;
7569         ctsio->kern_sg_entries = 0;
7570         ctsio->kern_data_len = min(total_len, alloc_len);
7571         ctsio->kern_total_len = ctsio->kern_data_len;
7572
7573         mtx_lock(&lun->lun_lock);
7574         switch (cdb->action) {
7575         case SPRI_RK: { // read keys
7576         struct scsi_per_res_in_keys *res_keys;
7577                 int i, key_count;
7578
7579                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7580
7581                 /*
7582                  * We had to drop the lock to allocate our buffer, which
7583                  * leaves time for someone to come in with another
7584                  * persistent reservation.  (That is unlikely, though,
7585                  * since this should be the only persistent reservation
7586                  * command active right now.)
7587                  */
7588                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7589                     (lun->pr_key_count *
7590                      sizeof(struct scsi_per_res_key)))){
7591                         mtx_unlock(&lun->lun_lock);
7592                         free(ctsio->kern_data_ptr, M_CTL);
7593                         printf("%s: reservation length changed, retrying\n",
7594                                __func__);
7595                         goto retry;
7596                 }
7597
7598                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7599
7600                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7601                              lun->pr_key_count, res_keys->header.length);
7602
7603                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7604                         if ((key = ctl_get_prkey(lun, i)) == 0)
7605                                 continue;
7606
7607                         /*
7608                          * We used lun->pr_key_count to calculate the
7609                          * size to allocate.  If it turns out the number of
7610                          * initiators with the registered flag set is
7611                          * larger than that (i.e. they haven't been kept in
7612                          * sync), we've got a problem.
7613                          */
7614                         if (key_count >= lun->pr_key_count) {
7615                                 key_count++;
7616                                 continue;
7617                         }
7618                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7619                         key_count++;
7620                 }
7621                 break;
7622         }
7623         case SPRI_RR: { // read reservation
7624                 struct scsi_per_res_in_rsrv *res;
7625                 int tmp_len, header_only;
7626
7627                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7628
7629                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7630
7631                 if (lun->flags & CTL_LUN_PR_RESERVED)
7632                 {
7633                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7634                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7635                                     res->header.length);
7636                         header_only = 0;
7637                 } else {
7638                         tmp_len = sizeof(struct scsi_per_res_in_header);
7639                         scsi_ulto4b(0, res->header.length);
7640                         header_only = 1;
7641                 }
7642
7643                 /*
7644                  * We had to drop the lock to allocate our buffer, which
7645                  * leaves time for someone to come in with another
7646                  * persistent reservation.  (That is unlikely, though,
7647                  * since this should be the only persistent reservation
7648                  * command active right now.)
7649                  */
7650                 if (tmp_len != total_len) {
7651                         mtx_unlock(&lun->lun_lock);
7652                         free(ctsio->kern_data_ptr, M_CTL);
7653                         printf("%s: reservation status changed, retrying\n",
7654                                __func__);
7655                         goto retry;
7656                 }
7657
7658                 /*
7659                  * No reservation held, so we're done.
7660                  */
7661                 if (header_only != 0)
7662                         break;
7663
7664                 /*
7665                  * If the registration is an All Registrants type, the key
7666                  * is 0, since it doesn't really matter.
7667                  */
7668                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7669                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7670                             res->data.reservation);
7671                 }
7672                 res->data.scopetype = lun->pr_res_type;
7673                 break;
7674         }
7675         case SPRI_RC:     //report capabilities
7676         {
7677                 struct scsi_per_res_cap *res_cap;
7678                 uint16_t type_mask;
7679
7680                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7681                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7682                 res_cap->flags1 = SPRI_CRH;
7683                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7684                 type_mask = SPRI_TM_WR_EX_AR |
7685                             SPRI_TM_EX_AC_RO |
7686                             SPRI_TM_WR_EX_RO |
7687                             SPRI_TM_EX_AC |
7688                             SPRI_TM_WR_EX |
7689                             SPRI_TM_EX_AC_AR;
7690                 scsi_ulto2b(type_mask, res_cap->type_mask);
7691                 break;
7692         }
7693         case SPRI_RS: { // read full status
7694                 struct scsi_per_res_in_full *res_status;
7695                 struct scsi_per_res_in_full_desc *res_desc;
7696                 struct ctl_port *port;
7697                 int i, len;
7698
7699                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7700
7701                 /*
7702                  * We had to drop the lock to allocate our buffer, which
7703                  * leaves time for someone to come in with another
7704                  * persistent reservation.  (That is unlikely, though,
7705                  * since this should be the only persistent reservation
7706                  * command active right now.)
7707                  */
7708                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7709                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7710                      lun->pr_key_count)){
7711                         mtx_unlock(&lun->lun_lock);
7712                         free(ctsio->kern_data_ptr, M_CTL);
7713                         printf("%s: reservation length changed, retrying\n",
7714                                __func__);
7715                         goto retry;
7716                 }
7717
7718                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7719
7720                 res_desc = &res_status->desc[0];
7721                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7722                         if ((key = ctl_get_prkey(lun, i)) == 0)
7723                                 continue;
7724
7725                         scsi_u64to8b(key, res_desc->res_key.key);
7726                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7727                             (lun->pr_res_idx == i ||
7728                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7729                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7730                                 res_desc->scopetype = lun->pr_res_type;
7731                         }
7732                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7733                             res_desc->rel_trgt_port_id);
7734                         len = 0;
7735                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7736                         if (port != NULL)
7737                                 len = ctl_create_iid(port,
7738                                     i % CTL_MAX_INIT_PER_PORT,
7739                                     res_desc->transport_id);
7740                         scsi_ulto4b(len, res_desc->additional_length);
7741                         res_desc = (struct scsi_per_res_in_full_desc *)
7742                             &res_desc->transport_id[len];
7743                 }
7744                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7745                     res_status->header.length);
7746                 break;
7747         }
7748         default:
7749                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7750         }
7751         mtx_unlock(&lun->lun_lock);
7752
7753         ctl_set_success(ctsio);
7754         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7755         ctsio->be_move_done = ctl_config_move_done;
7756         ctl_datamove((union ctl_io *)ctsio);
7757         return (CTL_RETVAL_COMPLETE);
7758 }
7759
7760 /*
7761  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7762  * it should return.
7763  */
7764 static int
7765 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7766                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7767                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7768                 struct scsi_per_res_out_parms* param)
7769 {
7770         union ctl_ha_msg persis_io;
7771         int i;
7772
7773         mtx_lock(&lun->lun_lock);
7774         if (sa_res_key == 0) {
7775                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7776                         /* validate scope and type */
7777                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7778                              SPR_LU_SCOPE) {
7779                                 mtx_unlock(&lun->lun_lock);
7780                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7781                                                       /*sks_valid*/ 1,
7782                                                       /*command*/ 1,
7783                                                       /*field*/ 2,
7784                                                       /*bit_valid*/ 1,
7785                                                       /*bit*/ 4);
7786                                 ctl_done((union ctl_io *)ctsio);
7787                                 return (1);
7788                         }
7789
7790                         if (type>8 || type==2 || type==4 || type==0) {
7791                                 mtx_unlock(&lun->lun_lock);
7792                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7793                                                       /*sks_valid*/ 1,
7794                                                       /*command*/ 1,
7795                                                       /*field*/ 2,
7796                                                       /*bit_valid*/ 1,
7797                                                       /*bit*/ 0);
7798                                 ctl_done((union ctl_io *)ctsio);
7799                                 return (1);
7800                         }
7801
7802                         /*
7803                          * Unregister everybody else and build UA for
7804                          * them
7805                          */
7806                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7807                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7808                                         continue;
7809
7810                                 ctl_clr_prkey(lun, i);
7811                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7812                         }
7813                         lun->pr_key_count = 1;
7814                         lun->pr_res_type = type;
7815                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7816                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7817                                 lun->pr_res_idx = residx;
7818                         lun->pr_generation++;
7819                         mtx_unlock(&lun->lun_lock);
7820
7821                         /* send msg to other side */
7822                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7823                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7824                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7825                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7826                         persis_io.pr.pr_info.res_type = type;
7827                         memcpy(persis_io.pr.pr_info.sa_res_key,
7828                                param->serv_act_res_key,
7829                                sizeof(param->serv_act_res_key));
7830                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7831                             sizeof(persis_io.pr), M_WAITOK);
7832                 } else {
7833                         /* not all registrants */
7834                         mtx_unlock(&lun->lun_lock);
7835                         free(ctsio->kern_data_ptr, M_CTL);
7836                         ctl_set_invalid_field(ctsio,
7837                                               /*sks_valid*/ 1,
7838                                               /*command*/ 0,
7839                                               /*field*/ 8,
7840                                               /*bit_valid*/ 0,
7841                                               /*bit*/ 0);
7842                         ctl_done((union ctl_io *)ctsio);
7843                         return (1);
7844                 }
7845         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7846                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7847                 int found = 0;
7848
7849                 if (res_key == sa_res_key) {
7850                         /* special case */
7851                         /*
7852                          * The spec implies this is not good but doesn't
7853                          * say what to do. There are two choices either
7854                          * generate a res conflict or check condition
7855                          * with illegal field in parameter data. Since
7856                          * that is what is done when the sa_res_key is
7857                          * zero I'll take that approach since this has
7858                          * to do with the sa_res_key.
7859                          */
7860                         mtx_unlock(&lun->lun_lock);
7861                         free(ctsio->kern_data_ptr, M_CTL);
7862                         ctl_set_invalid_field(ctsio,
7863                                               /*sks_valid*/ 1,
7864                                               /*command*/ 0,
7865                                               /*field*/ 8,
7866                                               /*bit_valid*/ 0,
7867                                               /*bit*/ 0);
7868                         ctl_done((union ctl_io *)ctsio);
7869                         return (1);
7870                 }
7871
7872                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7873                         if (ctl_get_prkey(lun, i) != sa_res_key)
7874                                 continue;
7875
7876                         found = 1;
7877                         ctl_clr_prkey(lun, i);
7878                         lun->pr_key_count--;
7879                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7880                 }
7881                 if (!found) {
7882                         mtx_unlock(&lun->lun_lock);
7883                         free(ctsio->kern_data_ptr, M_CTL);
7884                         ctl_set_reservation_conflict(ctsio);
7885                         ctl_done((union ctl_io *)ctsio);
7886                         return (CTL_RETVAL_COMPLETE);
7887                 }
7888                 lun->pr_generation++;
7889                 mtx_unlock(&lun->lun_lock);
7890
7891                 /* send msg to other side */
7892                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7893                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7894                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7895                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7896                 persis_io.pr.pr_info.res_type = type;
7897                 memcpy(persis_io.pr.pr_info.sa_res_key,
7898                        param->serv_act_res_key,
7899                        sizeof(param->serv_act_res_key));
7900                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7901                     sizeof(persis_io.pr), M_WAITOK);
7902         } else {
7903                 /* Reserved but not all registrants */
7904                 /* sa_res_key is res holder */
7905                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7906                         /* validate scope and type */
7907                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7908                              SPR_LU_SCOPE) {
7909                                 mtx_unlock(&lun->lun_lock);
7910                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7911                                                       /*sks_valid*/ 1,
7912                                                       /*command*/ 1,
7913                                                       /*field*/ 2,
7914                                                       /*bit_valid*/ 1,
7915                                                       /*bit*/ 4);
7916                                 ctl_done((union ctl_io *)ctsio);
7917                                 return (1);
7918                         }
7919
7920                         if (type>8 || type==2 || type==4 || type==0) {
7921                                 mtx_unlock(&lun->lun_lock);
7922                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7923                                                       /*sks_valid*/ 1,
7924                                                       /*command*/ 1,
7925                                                       /*field*/ 2,
7926                                                       /*bit_valid*/ 1,
7927                                                       /*bit*/ 0);
7928                                 ctl_done((union ctl_io *)ctsio);
7929                                 return (1);
7930                         }
7931
7932                         /*
7933                          * Do the following:
7934                          * if sa_res_key != res_key remove all
7935                          * registrants w/sa_res_key and generate UA
7936                          * for these registrants(Registrations
7937                          * Preempted) if it wasn't an exclusive
7938                          * reservation generate UA(Reservations
7939                          * Preempted) for all other registered nexuses
7940                          * if the type has changed. Establish the new
7941                          * reservation and holder. If res_key and
7942                          * sa_res_key are the same do the above
7943                          * except don't unregister the res holder.
7944                          */
7945
7946                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7947                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7948                                         continue;
7949
7950                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
7951                                         ctl_clr_prkey(lun, i);
7952                                         lun->pr_key_count--;
7953                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7954                                 } else if (type != lun->pr_res_type &&
7955                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
7956                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
7957                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
7958                                 }
7959                         }
7960                         lun->pr_res_type = type;
7961                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7962                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7963                                 lun->pr_res_idx = residx;
7964                         else
7965                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7966                         lun->pr_generation++;
7967                         mtx_unlock(&lun->lun_lock);
7968
7969                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7970                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7971                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7972                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7973                         persis_io.pr.pr_info.res_type = type;
7974                         memcpy(persis_io.pr.pr_info.sa_res_key,
7975                                param->serv_act_res_key,
7976                                sizeof(param->serv_act_res_key));
7977                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7978                             sizeof(persis_io.pr), M_WAITOK);
7979                 } else {
7980                         /*
7981                          * sa_res_key is not the res holder just
7982                          * remove registrants
7983                          */
7984                         int found=0;
7985
7986                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7987                                 if (sa_res_key != ctl_get_prkey(lun, i))
7988                                         continue;
7989
7990                                 found = 1;
7991                                 ctl_clr_prkey(lun, i);
7992                                 lun->pr_key_count--;
7993                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7994                         }
7995
7996                         if (!found) {
7997                                 mtx_unlock(&lun->lun_lock);
7998                                 free(ctsio->kern_data_ptr, M_CTL);
7999                                 ctl_set_reservation_conflict(ctsio);
8000                                 ctl_done((union ctl_io *)ctsio);
8001                                 return (1);
8002                         }
8003                         lun->pr_generation++;
8004                         mtx_unlock(&lun->lun_lock);
8005
8006                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8007                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8008                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8009                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8010                         persis_io.pr.pr_info.res_type = type;
8011                         memcpy(persis_io.pr.pr_info.sa_res_key,
8012                                param->serv_act_res_key,
8013                                sizeof(param->serv_act_res_key));
8014                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8015                             sizeof(persis_io.pr), M_WAITOK);
8016                 }
8017         }
8018         return (0);
8019 }
8020
8021 static void
8022 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8023 {
8024         uint64_t sa_res_key;
8025         int i;
8026
8027         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8028
8029         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8030          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8031          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8032                 if (sa_res_key == 0) {
8033                         /*
8034                          * Unregister everybody else and build UA for
8035                          * them
8036                          */
8037                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8038                                 if (i == msg->pr.pr_info.residx ||
8039                                     ctl_get_prkey(lun, i) == 0)
8040                                         continue;
8041
8042                                 ctl_clr_prkey(lun, i);
8043                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8044                         }
8045
8046                         lun->pr_key_count = 1;
8047                         lun->pr_res_type = msg->pr.pr_info.res_type;
8048                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8049                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8050                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8051                 } else {
8052                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8053                                 if (sa_res_key == ctl_get_prkey(lun, i))
8054                                         continue;
8055
8056                                 ctl_clr_prkey(lun, i);
8057                                 lun->pr_key_count--;
8058                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8059                         }
8060                 }
8061         } else {
8062                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8063                         if (i == msg->pr.pr_info.residx ||
8064                             ctl_get_prkey(lun, i) == 0)
8065                                 continue;
8066
8067                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8068                                 ctl_clr_prkey(lun, i);
8069                                 lun->pr_key_count--;
8070                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8071                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8072                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8073                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8074                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8075                         }
8076                 }
8077                 lun->pr_res_type = msg->pr.pr_info.res_type;
8078                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8079                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8080                         lun->pr_res_idx = msg->pr.pr_info.residx;
8081                 else
8082                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8083         }
8084         lun->pr_generation++;
8085
8086 }
8087
8088
8089 int
8090 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8091 {
8092         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8093         struct ctl_lun *lun = CTL_LUN(ctsio);
8094         int retval;
8095         u_int32_t param_len;
8096         struct scsi_per_res_out *cdb;
8097         struct scsi_per_res_out_parms* param;
8098         uint32_t residx;
8099         uint64_t res_key, sa_res_key, key;
8100         uint8_t type;
8101         union ctl_ha_msg persis_io;
8102         int    i;
8103
8104         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8105
8106         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8107         retval = CTL_RETVAL_COMPLETE;
8108
8109         /*
8110          * We only support whole-LUN scope.  The scope & type are ignored for
8111          * register, register and ignore existing key and clear.
8112          * We sometimes ignore scope and type on preempts too!!
8113          * Verify reservation type here as well.
8114          */
8115         type = cdb->scope_type & SPR_TYPE_MASK;
8116         if ((cdb->action == SPRO_RESERVE)
8117          || (cdb->action == SPRO_RELEASE)) {
8118                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8119                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8120                                               /*sks_valid*/ 1,
8121                                               /*command*/ 1,
8122                                               /*field*/ 2,
8123                                               /*bit_valid*/ 1,
8124                                               /*bit*/ 4);
8125                         ctl_done((union ctl_io *)ctsio);
8126                         return (CTL_RETVAL_COMPLETE);
8127                 }
8128
8129                 if (type>8 || type==2 || type==4 || type==0) {
8130                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8131                                               /*sks_valid*/ 1,
8132                                               /*command*/ 1,
8133                                               /*field*/ 2,
8134                                               /*bit_valid*/ 1,
8135                                               /*bit*/ 0);
8136                         ctl_done((union ctl_io *)ctsio);
8137                         return (CTL_RETVAL_COMPLETE);
8138                 }
8139         }
8140
8141         param_len = scsi_4btoul(cdb->length);
8142
8143         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8144                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8145                 ctsio->kern_data_len = param_len;
8146                 ctsio->kern_total_len = param_len;
8147                 ctsio->kern_rel_offset = 0;
8148                 ctsio->kern_sg_entries = 0;
8149                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8150                 ctsio->be_move_done = ctl_config_move_done;
8151                 ctl_datamove((union ctl_io *)ctsio);
8152
8153                 return (CTL_RETVAL_COMPLETE);
8154         }
8155
8156         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8157
8158         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8159         res_key = scsi_8btou64(param->res_key.key);
8160         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8161
8162         /*
8163          * Validate the reservation key here except for SPRO_REG_IGNO
8164          * This must be done for all other service actions
8165          */
8166         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8167                 mtx_lock(&lun->lun_lock);
8168                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8169                         if (res_key != key) {
8170                                 /*
8171                                  * The current key passed in doesn't match
8172                                  * the one the initiator previously
8173                                  * registered.
8174                                  */
8175                                 mtx_unlock(&lun->lun_lock);
8176                                 free(ctsio->kern_data_ptr, M_CTL);
8177                                 ctl_set_reservation_conflict(ctsio);
8178                                 ctl_done((union ctl_io *)ctsio);
8179                                 return (CTL_RETVAL_COMPLETE);
8180                         }
8181                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8182                         /*
8183                          * We are not registered
8184                          */
8185                         mtx_unlock(&lun->lun_lock);
8186                         free(ctsio->kern_data_ptr, M_CTL);
8187                         ctl_set_reservation_conflict(ctsio);
8188                         ctl_done((union ctl_io *)ctsio);
8189                         return (CTL_RETVAL_COMPLETE);
8190                 } else if (res_key != 0) {
8191                         /*
8192                          * We are not registered and trying to register but
8193                          * the register key isn't zero.
8194                          */
8195                         mtx_unlock(&lun->lun_lock);
8196                         free(ctsio->kern_data_ptr, M_CTL);
8197                         ctl_set_reservation_conflict(ctsio);
8198                         ctl_done((union ctl_io *)ctsio);
8199                         return (CTL_RETVAL_COMPLETE);
8200                 }
8201                 mtx_unlock(&lun->lun_lock);
8202         }
8203
8204         switch (cdb->action & SPRO_ACTION_MASK) {
8205         case SPRO_REGISTER:
8206         case SPRO_REG_IGNO: {
8207
8208 #if 0
8209                 printf("Registration received\n");
8210 #endif
8211
8212                 /*
8213                  * We don't support any of these options, as we report in
8214                  * the read capabilities request (see
8215                  * ctl_persistent_reserve_in(), above).
8216                  */
8217                 if ((param->flags & SPR_SPEC_I_PT)
8218                  || (param->flags & SPR_ALL_TG_PT)
8219                  || (param->flags & SPR_APTPL)) {
8220                         int bit_ptr;
8221
8222                         if (param->flags & SPR_APTPL)
8223                                 bit_ptr = 0;
8224                         else if (param->flags & SPR_ALL_TG_PT)
8225                                 bit_ptr = 2;
8226                         else /* SPR_SPEC_I_PT */
8227                                 bit_ptr = 3;
8228
8229                         free(ctsio->kern_data_ptr, M_CTL);
8230                         ctl_set_invalid_field(ctsio,
8231                                               /*sks_valid*/ 1,
8232                                               /*command*/ 0,
8233                                               /*field*/ 20,
8234                                               /*bit_valid*/ 1,
8235                                               /*bit*/ bit_ptr);
8236                         ctl_done((union ctl_io *)ctsio);
8237                         return (CTL_RETVAL_COMPLETE);
8238                 }
8239
8240                 mtx_lock(&lun->lun_lock);
8241
8242                 /*
8243                  * The initiator wants to clear the
8244                  * key/unregister.
8245                  */
8246                 if (sa_res_key == 0) {
8247                         if ((res_key == 0
8248                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8249                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8250                           && ctl_get_prkey(lun, residx) == 0)) {
8251                                 mtx_unlock(&lun->lun_lock);
8252                                 goto done;
8253                         }
8254
8255                         ctl_clr_prkey(lun, residx);
8256                         lun->pr_key_count--;
8257
8258                         if (residx == lun->pr_res_idx) {
8259                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8260                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8261
8262                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8263                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8264                                     lun->pr_key_count) {
8265                                         /*
8266                                          * If the reservation is a registrants
8267                                          * only type we need to generate a UA
8268                                          * for other registered inits.  The
8269                                          * sense code should be RESERVATIONS
8270                                          * RELEASED
8271                                          */
8272
8273                                         for (i = softc->init_min; i < softc->init_max; i++){
8274                                                 if (ctl_get_prkey(lun, i) == 0)
8275                                                         continue;
8276                                                 ctl_est_ua(lun, i,
8277                                                     CTL_UA_RES_RELEASE);
8278                                         }
8279                                 }
8280                                 lun->pr_res_type = 0;
8281                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8282                                 if (lun->pr_key_count==0) {
8283                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8284                                         lun->pr_res_type = 0;
8285                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8286                                 }
8287                         }
8288                         lun->pr_generation++;
8289                         mtx_unlock(&lun->lun_lock);
8290
8291                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8292                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8293                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8294                         persis_io.pr.pr_info.residx = residx;
8295                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8296                             sizeof(persis_io.pr), M_WAITOK);
8297                 } else /* sa_res_key != 0 */ {
8298
8299                         /*
8300                          * If we aren't registered currently then increment
8301                          * the key count and set the registered flag.
8302                          */
8303                         ctl_alloc_prkey(lun, residx);
8304                         if (ctl_get_prkey(lun, residx) == 0)
8305                                 lun->pr_key_count++;
8306                         ctl_set_prkey(lun, residx, sa_res_key);
8307                         lun->pr_generation++;
8308                         mtx_unlock(&lun->lun_lock);
8309
8310                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8311                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8312                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8313                         persis_io.pr.pr_info.residx = residx;
8314                         memcpy(persis_io.pr.pr_info.sa_res_key,
8315                                param->serv_act_res_key,
8316                                sizeof(param->serv_act_res_key));
8317                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8318                             sizeof(persis_io.pr), M_WAITOK);
8319                 }
8320
8321                 break;
8322         }
8323         case SPRO_RESERVE:
8324 #if 0
8325                 printf("Reserve executed type %d\n", type);
8326 #endif
8327                 mtx_lock(&lun->lun_lock);
8328                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8329                         /*
8330                          * if this isn't the reservation holder and it's
8331                          * not a "all registrants" type or if the type is
8332                          * different then we have a conflict
8333                          */
8334                         if ((lun->pr_res_idx != residx
8335                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8336                          || lun->pr_res_type != type) {
8337                                 mtx_unlock(&lun->lun_lock);
8338                                 free(ctsio->kern_data_ptr, M_CTL);
8339                                 ctl_set_reservation_conflict(ctsio);
8340                                 ctl_done((union ctl_io *)ctsio);
8341                                 return (CTL_RETVAL_COMPLETE);
8342                         }
8343                         mtx_unlock(&lun->lun_lock);
8344                 } else /* create a reservation */ {
8345                         /*
8346                          * If it's not an "all registrants" type record
8347                          * reservation holder
8348                          */
8349                         if (type != SPR_TYPE_WR_EX_AR
8350                          && type != SPR_TYPE_EX_AC_AR)
8351                                 lun->pr_res_idx = residx; /* Res holder */
8352                         else
8353                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8354
8355                         lun->flags |= CTL_LUN_PR_RESERVED;
8356                         lun->pr_res_type = type;
8357
8358                         mtx_unlock(&lun->lun_lock);
8359
8360                         /* send msg to other side */
8361                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8362                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8363                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8364                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8365                         persis_io.pr.pr_info.res_type = type;
8366                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8367                             sizeof(persis_io.pr), M_WAITOK);
8368                 }
8369                 break;
8370
8371         case SPRO_RELEASE:
8372                 mtx_lock(&lun->lun_lock);
8373                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8374                         /* No reservation exists return good status */
8375                         mtx_unlock(&lun->lun_lock);
8376                         goto done;
8377                 }
8378                 /*
8379                  * Is this nexus a reservation holder?
8380                  */
8381                 if (lun->pr_res_idx != residx
8382                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8383                         /*
8384                          * not a res holder return good status but
8385                          * do nothing
8386                          */
8387                         mtx_unlock(&lun->lun_lock);
8388                         goto done;
8389                 }
8390
8391                 if (lun->pr_res_type != type) {
8392                         mtx_unlock(&lun->lun_lock);
8393                         free(ctsio->kern_data_ptr, M_CTL);
8394                         ctl_set_illegal_pr_release(ctsio);
8395                         ctl_done((union ctl_io *)ctsio);
8396                         return (CTL_RETVAL_COMPLETE);
8397                 }
8398
8399                 /* okay to release */
8400                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8401                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8402                 lun->pr_res_type = 0;
8403
8404                 /*
8405                  * If this isn't an exclusive access reservation and NUAR
8406                  * is not set, generate UA for all other registrants.
8407                  */
8408                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8409                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8410                         for (i = softc->init_min; i < softc->init_max; i++) {
8411                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8412                                         continue;
8413                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8414                         }
8415                 }
8416                 mtx_unlock(&lun->lun_lock);
8417
8418                 /* Send msg to other side */
8419                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8420                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8421                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8422                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8423                      sizeof(persis_io.pr), M_WAITOK);
8424                 break;
8425
8426         case SPRO_CLEAR:
8427                 /* send msg to other side */
8428
8429                 mtx_lock(&lun->lun_lock);
8430                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8431                 lun->pr_res_type = 0;
8432                 lun->pr_key_count = 0;
8433                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8434
8435                 ctl_clr_prkey(lun, residx);
8436                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8437                         if (ctl_get_prkey(lun, i) != 0) {
8438                                 ctl_clr_prkey(lun, i);
8439                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8440                         }
8441                 lun->pr_generation++;
8442                 mtx_unlock(&lun->lun_lock);
8443
8444                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8445                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8446                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8447                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8448                      sizeof(persis_io.pr), M_WAITOK);
8449                 break;
8450
8451         case SPRO_PREEMPT:
8452         case SPRO_PRE_ABO: {
8453                 int nretval;
8454
8455                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8456                                           residx, ctsio, cdb, param);
8457                 if (nretval != 0)
8458                         return (CTL_RETVAL_COMPLETE);
8459                 break;
8460         }
8461         default:
8462                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8463         }
8464
8465 done:
8466         free(ctsio->kern_data_ptr, M_CTL);
8467         ctl_set_success(ctsio);
8468         ctl_done((union ctl_io *)ctsio);
8469
8470         return (retval);
8471 }
8472
8473 /*
8474  * This routine is for handling a message from the other SC pertaining to
8475  * persistent reserve out. All the error checking will have been done
8476  * so only perorming the action need be done here to keep the two
8477  * in sync.
8478  */
8479 static void
8480 ctl_hndl_per_res_out_on_other_sc(union ctl_io *io)
8481 {
8482         struct ctl_softc *softc = CTL_SOFTC(io);
8483         union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg;
8484         struct ctl_lun *lun;
8485         int i;
8486         uint32_t residx, targ_lun;
8487
8488         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8489         mtx_lock(&softc->ctl_lock);
8490         if (targ_lun >= CTL_MAX_LUNS ||
8491             (lun = softc->ctl_luns[targ_lun]) == NULL) {
8492                 mtx_unlock(&softc->ctl_lock);
8493                 return;
8494         }
8495         mtx_lock(&lun->lun_lock);
8496         mtx_unlock(&softc->ctl_lock);
8497         if (lun->flags & CTL_LUN_DISABLED) {
8498                 mtx_unlock(&lun->lun_lock);
8499                 return;
8500         }
8501         residx = ctl_get_initindex(&msg->hdr.nexus);
8502         switch(msg->pr.pr_info.action) {
8503         case CTL_PR_REG_KEY:
8504                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8505                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8506                         lun->pr_key_count++;
8507                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8508                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8509                 lun->pr_generation++;
8510                 break;
8511
8512         case CTL_PR_UNREG_KEY:
8513                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8514                 lun->pr_key_count--;
8515
8516                 /* XXX Need to see if the reservation has been released */
8517                 /* if so do we need to generate UA? */
8518                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8519                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8520                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8521
8522                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8523                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8524                             lun->pr_key_count) {
8525                                 /*
8526                                  * If the reservation is a registrants
8527                                  * only type we need to generate a UA
8528                                  * for other registered inits.  The
8529                                  * sense code should be RESERVATIONS
8530                                  * RELEASED
8531                                  */
8532
8533                                 for (i = softc->init_min; i < softc->init_max; i++) {
8534                                         if (ctl_get_prkey(lun, i) == 0)
8535                                                 continue;
8536
8537                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8538                                 }
8539                         }
8540                         lun->pr_res_type = 0;
8541                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8542                         if (lun->pr_key_count==0) {
8543                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8544                                 lun->pr_res_type = 0;
8545                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8546                         }
8547                 }
8548                 lun->pr_generation++;
8549                 break;
8550
8551         case CTL_PR_RESERVE:
8552                 lun->flags |= CTL_LUN_PR_RESERVED;
8553                 lun->pr_res_type = msg->pr.pr_info.res_type;
8554                 lun->pr_res_idx = msg->pr.pr_info.residx;
8555
8556                 break;
8557
8558         case CTL_PR_RELEASE:
8559                 /*
8560                  * If this isn't an exclusive access reservation and NUAR
8561                  * is not set, generate UA for all other registrants.
8562                  */
8563                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8564                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8565                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8566                         for (i = softc->init_min; i < softc->init_max; i++)
8567                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8568                                         continue;
8569                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8570                 }
8571
8572                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8573                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8574                 lun->pr_res_type = 0;
8575                 break;
8576
8577         case CTL_PR_PREEMPT:
8578                 ctl_pro_preempt_other(lun, msg);
8579                 break;
8580         case CTL_PR_CLEAR:
8581                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8582                 lun->pr_res_type = 0;
8583                 lun->pr_key_count = 0;
8584                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8585
8586                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8587                         if (ctl_get_prkey(lun, i) == 0)
8588                                 continue;
8589                         ctl_clr_prkey(lun, i);
8590                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8591                 }
8592                 lun->pr_generation++;
8593                 break;
8594         }
8595
8596         mtx_unlock(&lun->lun_lock);
8597 }
8598
8599 int
8600 ctl_read_write(struct ctl_scsiio *ctsio)
8601 {
8602         struct ctl_lun *lun = CTL_LUN(ctsio);
8603         struct ctl_lba_len_flags *lbalen;
8604         uint64_t lba;
8605         uint32_t num_blocks;
8606         int flags, retval;
8607         int isread;
8608
8609         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8610
8611         flags = 0;
8612         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8613               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8614         switch (ctsio->cdb[0]) {
8615         case READ_6:
8616         case WRITE_6: {
8617                 struct scsi_rw_6 *cdb;
8618
8619                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8620
8621                 lba = scsi_3btoul(cdb->addr);
8622                 /* only 5 bits are valid in the most significant address byte */
8623                 lba &= 0x1fffff;
8624                 num_blocks = cdb->length;
8625                 /*
8626                  * This is correct according to SBC-2.
8627                  */
8628                 if (num_blocks == 0)
8629                         num_blocks = 256;
8630                 break;
8631         }
8632         case READ_10:
8633         case WRITE_10: {
8634                 struct scsi_rw_10 *cdb;
8635
8636                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8637                 if (cdb->byte2 & SRW10_FUA)
8638                         flags |= CTL_LLF_FUA;
8639                 if (cdb->byte2 & SRW10_DPO)
8640                         flags |= CTL_LLF_DPO;
8641                 lba = scsi_4btoul(cdb->addr);
8642                 num_blocks = scsi_2btoul(cdb->length);
8643                 break;
8644         }
8645         case WRITE_VERIFY_10: {
8646                 struct scsi_write_verify_10 *cdb;
8647
8648                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8649                 flags |= CTL_LLF_FUA;
8650                 if (cdb->byte2 & SWV_DPO)
8651                         flags |= CTL_LLF_DPO;
8652                 lba = scsi_4btoul(cdb->addr);
8653                 num_blocks = scsi_2btoul(cdb->length);
8654                 break;
8655         }
8656         case READ_12:
8657         case WRITE_12: {
8658                 struct scsi_rw_12 *cdb;
8659
8660                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8661                 if (cdb->byte2 & SRW12_FUA)
8662                         flags |= CTL_LLF_FUA;
8663                 if (cdb->byte2 & SRW12_DPO)
8664                         flags |= CTL_LLF_DPO;
8665                 lba = scsi_4btoul(cdb->addr);
8666                 num_blocks = scsi_4btoul(cdb->length);
8667                 break;
8668         }
8669         case WRITE_VERIFY_12: {
8670                 struct scsi_write_verify_12 *cdb;
8671
8672                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8673                 flags |= CTL_LLF_FUA;
8674                 if (cdb->byte2 & SWV_DPO)
8675                         flags |= CTL_LLF_DPO;
8676                 lba = scsi_4btoul(cdb->addr);
8677                 num_blocks = scsi_4btoul(cdb->length);
8678                 break;
8679         }
8680         case READ_16:
8681         case WRITE_16: {
8682                 struct scsi_rw_16 *cdb;
8683
8684                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8685                 if (cdb->byte2 & SRW12_FUA)
8686                         flags |= CTL_LLF_FUA;
8687                 if (cdb->byte2 & SRW12_DPO)
8688                         flags |= CTL_LLF_DPO;
8689                 lba = scsi_8btou64(cdb->addr);
8690                 num_blocks = scsi_4btoul(cdb->length);
8691                 break;
8692         }
8693         case WRITE_ATOMIC_16: {
8694                 struct scsi_write_atomic_16 *cdb;
8695
8696                 if (lun->be_lun->atomicblock == 0) {
8697                         ctl_set_invalid_opcode(ctsio);
8698                         ctl_done((union ctl_io *)ctsio);
8699                         return (CTL_RETVAL_COMPLETE);
8700                 }
8701
8702                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8703                 if (cdb->byte2 & SRW12_FUA)
8704                         flags |= CTL_LLF_FUA;
8705                 if (cdb->byte2 & SRW12_DPO)
8706                         flags |= CTL_LLF_DPO;
8707                 lba = scsi_8btou64(cdb->addr);
8708                 num_blocks = scsi_2btoul(cdb->length);
8709                 if (num_blocks > lun->be_lun->atomicblock) {
8710                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8711                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8712                             /*bit*/ 0);
8713                         ctl_done((union ctl_io *)ctsio);
8714                         return (CTL_RETVAL_COMPLETE);
8715                 }
8716                 break;
8717         }
8718         case WRITE_VERIFY_16: {
8719                 struct scsi_write_verify_16 *cdb;
8720
8721                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8722                 flags |= CTL_LLF_FUA;
8723                 if (cdb->byte2 & SWV_DPO)
8724                         flags |= CTL_LLF_DPO;
8725                 lba = scsi_8btou64(cdb->addr);
8726                 num_blocks = scsi_4btoul(cdb->length);
8727                 break;
8728         }
8729         default:
8730                 /*
8731                  * We got a command we don't support.  This shouldn't
8732                  * happen, commands should be filtered out above us.
8733                  */
8734                 ctl_set_invalid_opcode(ctsio);
8735                 ctl_done((union ctl_io *)ctsio);
8736
8737                 return (CTL_RETVAL_COMPLETE);
8738                 break; /* NOTREACHED */
8739         }
8740
8741         /*
8742          * The first check is to make sure we're in bounds, the second
8743          * check is to catch wrap-around problems.  If the lba + num blocks
8744          * is less than the lba, then we've wrapped around and the block
8745          * range is invalid anyway.
8746          */
8747         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8748          || ((lba + num_blocks) < lba)) {
8749                 ctl_set_lba_out_of_range(ctsio,
8750                     MAX(lba, lun->be_lun->maxlba + 1));
8751                 ctl_done((union ctl_io *)ctsio);
8752                 return (CTL_RETVAL_COMPLETE);
8753         }
8754
8755         /*
8756          * According to SBC-3, a transfer length of 0 is not an error.
8757          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8758          * translates to 256 blocks for those commands.
8759          */
8760         if (num_blocks == 0) {
8761                 ctl_set_success(ctsio);
8762                 ctl_done((union ctl_io *)ctsio);
8763                 return (CTL_RETVAL_COMPLETE);
8764         }
8765
8766         /* Set FUA and/or DPO if caches are disabled. */
8767         if (isread) {
8768                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8769                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8770         } else {
8771                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8772                         flags |= CTL_LLF_FUA;
8773         }
8774
8775         lbalen = (struct ctl_lba_len_flags *)
8776             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8777         lbalen->lba = lba;
8778         lbalen->len = num_blocks;
8779         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8780
8781         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8782         ctsio->kern_rel_offset = 0;
8783
8784         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8785
8786         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8787         return (retval);
8788 }
8789
8790 static int
8791 ctl_cnw_cont(union ctl_io *io)
8792 {
8793         struct ctl_lun *lun = CTL_LUN(io);
8794         struct ctl_scsiio *ctsio;
8795         struct ctl_lba_len_flags *lbalen;
8796         int retval;
8797
8798         ctsio = &io->scsiio;
8799         ctsio->io_hdr.status = CTL_STATUS_NONE;
8800         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8801         lbalen = (struct ctl_lba_len_flags *)
8802             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8803         lbalen->flags &= ~CTL_LLF_COMPARE;
8804         lbalen->flags |= CTL_LLF_WRITE;
8805
8806         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8807         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8808         return (retval);
8809 }
8810
8811 int
8812 ctl_cnw(struct ctl_scsiio *ctsio)
8813 {
8814         struct ctl_lun *lun = CTL_LUN(ctsio);
8815         struct ctl_lba_len_flags *lbalen;
8816         uint64_t lba;
8817         uint32_t num_blocks;
8818         int flags, retval;
8819
8820         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8821
8822         flags = 0;
8823         switch (ctsio->cdb[0]) {
8824         case COMPARE_AND_WRITE: {
8825                 struct scsi_compare_and_write *cdb;
8826
8827                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8828                 if (cdb->byte2 & SRW10_FUA)
8829                         flags |= CTL_LLF_FUA;
8830                 if (cdb->byte2 & SRW10_DPO)
8831                         flags |= CTL_LLF_DPO;
8832                 lba = scsi_8btou64(cdb->addr);
8833                 num_blocks = cdb->length;
8834                 break;
8835         }
8836         default:
8837                 /*
8838                  * We got a command we don't support.  This shouldn't
8839                  * happen, commands should be filtered out above us.
8840                  */
8841                 ctl_set_invalid_opcode(ctsio);
8842                 ctl_done((union ctl_io *)ctsio);
8843
8844                 return (CTL_RETVAL_COMPLETE);
8845                 break; /* NOTREACHED */
8846         }
8847
8848         /*
8849          * The first check is to make sure we're in bounds, the second
8850          * check is to catch wrap-around problems.  If the lba + num blocks
8851          * is less than the lba, then we've wrapped around and the block
8852          * range is invalid anyway.
8853          */
8854         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8855          || ((lba + num_blocks) < lba)) {
8856                 ctl_set_lba_out_of_range(ctsio,
8857                     MAX(lba, lun->be_lun->maxlba + 1));
8858                 ctl_done((union ctl_io *)ctsio);
8859                 return (CTL_RETVAL_COMPLETE);
8860         }
8861
8862         /*
8863          * According to SBC-3, a transfer length of 0 is not an error.
8864          */
8865         if (num_blocks == 0) {
8866                 ctl_set_success(ctsio);
8867                 ctl_done((union ctl_io *)ctsio);
8868                 return (CTL_RETVAL_COMPLETE);
8869         }
8870
8871         /* Set FUA if write cache is disabled. */
8872         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8873                 flags |= CTL_LLF_FUA;
8874
8875         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8876         ctsio->kern_rel_offset = 0;
8877
8878         /*
8879          * Set the IO_CONT flag, so that if this I/O gets passed to
8880          * ctl_data_submit_done(), it'll get passed back to
8881          * ctl_ctl_cnw_cont() for further processing.
8882          */
8883         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8884         ctsio->io_cont = ctl_cnw_cont;
8885
8886         lbalen = (struct ctl_lba_len_flags *)
8887             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8888         lbalen->lba = lba;
8889         lbalen->len = num_blocks;
8890         lbalen->flags = CTL_LLF_COMPARE | flags;
8891
8892         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8893         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8894         return (retval);
8895 }
8896
8897 int
8898 ctl_verify(struct ctl_scsiio *ctsio)
8899 {
8900         struct ctl_lun *lun = CTL_LUN(ctsio);
8901         struct ctl_lba_len_flags *lbalen;
8902         uint64_t lba;
8903         uint32_t num_blocks;
8904         int bytchk, flags;
8905         int retval;
8906
8907         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8908
8909         bytchk = 0;
8910         flags = CTL_LLF_FUA;
8911         switch (ctsio->cdb[0]) {
8912         case VERIFY_10: {
8913                 struct scsi_verify_10 *cdb;
8914
8915                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8916                 if (cdb->byte2 & SVFY_BYTCHK)
8917                         bytchk = 1;
8918                 if (cdb->byte2 & SVFY_DPO)
8919                         flags |= CTL_LLF_DPO;
8920                 lba = scsi_4btoul(cdb->addr);
8921                 num_blocks = scsi_2btoul(cdb->length);
8922                 break;
8923         }
8924         case VERIFY_12: {
8925                 struct scsi_verify_12 *cdb;
8926
8927                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8928                 if (cdb->byte2 & SVFY_BYTCHK)
8929                         bytchk = 1;
8930                 if (cdb->byte2 & SVFY_DPO)
8931                         flags |= CTL_LLF_DPO;
8932                 lba = scsi_4btoul(cdb->addr);
8933                 num_blocks = scsi_4btoul(cdb->length);
8934                 break;
8935         }
8936         case VERIFY_16: {
8937                 struct scsi_rw_16 *cdb;
8938
8939                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8940                 if (cdb->byte2 & SVFY_BYTCHK)
8941                         bytchk = 1;
8942                 if (cdb->byte2 & SVFY_DPO)
8943                         flags |= CTL_LLF_DPO;
8944                 lba = scsi_8btou64(cdb->addr);
8945                 num_blocks = scsi_4btoul(cdb->length);
8946                 break;
8947         }
8948         default:
8949                 /*
8950                  * We got a command we don't support.  This shouldn't
8951                  * happen, commands should be filtered out above us.
8952                  */
8953                 ctl_set_invalid_opcode(ctsio);
8954                 ctl_done((union ctl_io *)ctsio);
8955                 return (CTL_RETVAL_COMPLETE);
8956         }
8957
8958         /*
8959          * The first check is to make sure we're in bounds, the second
8960          * check is to catch wrap-around problems.  If the lba + num blocks
8961          * is less than the lba, then we've wrapped around and the block
8962          * range is invalid anyway.
8963          */
8964         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8965          || ((lba + num_blocks) < lba)) {
8966                 ctl_set_lba_out_of_range(ctsio,
8967                     MAX(lba, lun->be_lun->maxlba + 1));
8968                 ctl_done((union ctl_io *)ctsio);
8969                 return (CTL_RETVAL_COMPLETE);
8970         }
8971
8972         /*
8973          * According to SBC-3, a transfer length of 0 is not an error.
8974          */
8975         if (num_blocks == 0) {
8976                 ctl_set_success(ctsio);
8977                 ctl_done((union ctl_io *)ctsio);
8978                 return (CTL_RETVAL_COMPLETE);
8979         }
8980
8981         lbalen = (struct ctl_lba_len_flags *)
8982             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8983         lbalen->lba = lba;
8984         lbalen->len = num_blocks;
8985         if (bytchk) {
8986                 lbalen->flags = CTL_LLF_COMPARE | flags;
8987                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8988         } else {
8989                 lbalen->flags = CTL_LLF_VERIFY | flags;
8990                 ctsio->kern_total_len = 0;
8991         }
8992         ctsio->kern_rel_offset = 0;
8993
8994         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
8995         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8996         return (retval);
8997 }
8998
8999 int
9000 ctl_report_luns(struct ctl_scsiio *ctsio)
9001 {
9002         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9003         struct ctl_port *port = CTL_PORT(ctsio);
9004         struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio);
9005         struct scsi_report_luns *cdb;
9006         struct scsi_report_luns_data *lun_data;
9007         int num_filled, num_luns, num_port_luns, retval;
9008         uint32_t alloc_len, lun_datalen;
9009         uint32_t initidx, targ_lun_id, lun_id;
9010
9011         retval = CTL_RETVAL_COMPLETE;
9012         cdb = (struct scsi_report_luns *)ctsio->cdb;
9013
9014         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9015
9016         num_luns = 0;
9017         num_port_luns = port->lun_map ? port->lun_map_size : CTL_MAX_LUNS;
9018         mtx_lock(&softc->ctl_lock);
9019         for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) {
9020                 if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX)
9021                         num_luns++;
9022         }
9023         mtx_unlock(&softc->ctl_lock);
9024
9025         switch (cdb->select_report) {
9026         case RPL_REPORT_DEFAULT:
9027         case RPL_REPORT_ALL:
9028         case RPL_REPORT_NONSUBSID:
9029                 break;
9030         case RPL_REPORT_WELLKNOWN:
9031         case RPL_REPORT_ADMIN:
9032         case RPL_REPORT_CONGLOM:
9033                 num_luns = 0;
9034                 break;
9035         default:
9036                 ctl_set_invalid_field(ctsio,
9037                                       /*sks_valid*/ 1,
9038                                       /*command*/ 1,
9039                                       /*field*/ 2,
9040                                       /*bit_valid*/ 0,
9041                                       /*bit*/ 0);
9042                 ctl_done((union ctl_io *)ctsio);
9043                 return (retval);
9044                 break; /* NOTREACHED */
9045         }
9046
9047         alloc_len = scsi_4btoul(cdb->length);
9048         /*
9049          * The initiator has to allocate at least 16 bytes for this request,
9050          * so he can at least get the header and the first LUN.  Otherwise
9051          * we reject the request (per SPC-3 rev 14, section 6.21).
9052          */
9053         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9054             sizeof(struct scsi_report_luns_lundata))) {
9055                 ctl_set_invalid_field(ctsio,
9056                                       /*sks_valid*/ 1,
9057                                       /*command*/ 1,
9058                                       /*field*/ 6,
9059                                       /*bit_valid*/ 0,
9060                                       /*bit*/ 0);
9061                 ctl_done((union ctl_io *)ctsio);
9062                 return (retval);
9063         }
9064
9065         lun_datalen = sizeof(*lun_data) +
9066                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9067
9068         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9069         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9070         ctsio->kern_sg_entries = 0;
9071
9072         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9073
9074         mtx_lock(&softc->ctl_lock);
9075         for (targ_lun_id = 0, num_filled = 0;
9076             targ_lun_id < num_port_luns && num_filled < num_luns;
9077             targ_lun_id++) {
9078                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9079                 if (lun_id == UINT32_MAX)
9080                         continue;
9081                 lun = softc->ctl_luns[lun_id];
9082                 if (lun == NULL)
9083                         continue;
9084
9085                 be64enc(lun_data->luns[num_filled++].lundata,
9086                     ctl_encode_lun(targ_lun_id));
9087
9088                 /*
9089                  * According to SPC-3, rev 14 section 6.21:
9090                  *
9091                  * "The execution of a REPORT LUNS command to any valid and
9092                  * installed logical unit shall clear the REPORTED LUNS DATA
9093                  * HAS CHANGED unit attention condition for all logical
9094                  * units of that target with respect to the requesting
9095                  * initiator. A valid and installed logical unit is one
9096                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9097                  * INQUIRY data (see 6.4.2)."
9098                  *
9099                  * If request_lun is NULL, the LUN this report luns command
9100                  * was issued to is either disabled or doesn't exist. In that
9101                  * case, we shouldn't clear any pending lun change unit
9102                  * attention.
9103                  */
9104                 if (request_lun != NULL) {
9105                         mtx_lock(&lun->lun_lock);
9106                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9107                         mtx_unlock(&lun->lun_lock);
9108                 }
9109         }
9110         mtx_unlock(&softc->ctl_lock);
9111
9112         /*
9113          * It's quite possible that we've returned fewer LUNs than we allocated
9114          * space for.  Trim it.
9115          */
9116         lun_datalen = sizeof(*lun_data) +
9117                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9118         ctsio->kern_rel_offset = 0;
9119         ctsio->kern_sg_entries = 0;
9120         ctsio->kern_data_len = min(lun_datalen, alloc_len);
9121         ctsio->kern_total_len = ctsio->kern_data_len;
9122
9123         /*
9124          * We set this to the actual data length, regardless of how much
9125          * space we actually have to return results.  If the user looks at
9126          * this value, he'll know whether or not he allocated enough space
9127          * and reissue the command if necessary.  We don't support well
9128          * known logical units, so if the user asks for that, return none.
9129          */
9130         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9131
9132         /*
9133          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9134          * this request.
9135          */
9136         ctl_set_success(ctsio);
9137         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9138         ctsio->be_move_done = ctl_config_move_done;
9139         ctl_datamove((union ctl_io *)ctsio);
9140         return (retval);
9141 }
9142
9143 int
9144 ctl_request_sense(struct ctl_scsiio *ctsio)
9145 {
9146         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9147         struct ctl_lun *lun = CTL_LUN(ctsio);
9148         struct scsi_request_sense *cdb;
9149         struct scsi_sense_data *sense_ptr, *ps;
9150         uint32_t initidx;
9151         int have_error;
9152         u_int sense_len = SSD_FULL_SIZE;
9153         scsi_sense_data_type sense_format;
9154         ctl_ua_type ua_type;
9155         uint8_t asc = 0, ascq = 0;
9156
9157         cdb = (struct scsi_request_sense *)ctsio->cdb;
9158
9159         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9160
9161         /*
9162          * Determine which sense format the user wants.
9163          */
9164         if (cdb->byte2 & SRS_DESC)
9165                 sense_format = SSD_TYPE_DESC;
9166         else
9167                 sense_format = SSD_TYPE_FIXED;
9168
9169         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9170         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9171         ctsio->kern_sg_entries = 0;
9172         ctsio->kern_rel_offset = 0;
9173
9174         /*
9175          * struct scsi_sense_data, which is currently set to 256 bytes, is
9176          * larger than the largest allowed value for the length field in the
9177          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9178          */
9179         ctsio->kern_data_len = cdb->length;
9180         ctsio->kern_total_len = cdb->length;
9181
9182         /*
9183          * If we don't have a LUN, we don't have any pending sense.
9184          */
9185         if (lun == NULL ||
9186             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9187              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9188                 /* "Logical unit not supported" */
9189                 ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9190                     /*current_error*/ 1,
9191                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9192                     /*asc*/ 0x25,
9193                     /*ascq*/ 0x00,
9194                     SSD_ELEM_NONE);
9195                 goto send;
9196         }
9197
9198         have_error = 0;
9199         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9200         /*
9201          * Check for pending sense, and then for pending unit attentions.
9202          * Pending sense gets returned first, then pending unit attentions.
9203          */
9204         mtx_lock(&lun->lun_lock);
9205         ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
9206         if (ps != NULL)
9207                 ps += initidx % CTL_MAX_INIT_PER_PORT;
9208         if (ps != NULL && ps->error_code != 0) {
9209                 scsi_sense_data_type stored_format;
9210
9211                 /*
9212                  * Check to see which sense format was used for the stored
9213                  * sense data.
9214                  */
9215                 stored_format = scsi_sense_type(ps);
9216
9217                 /*
9218                  * If the user requested a different sense format than the
9219                  * one we stored, then we need to convert it to the other
9220                  * format.  If we're going from descriptor to fixed format
9221                  * sense data, we may lose things in translation, depending
9222                  * on what options were used.
9223                  *
9224                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9225                  * for some reason we'll just copy it out as-is.
9226                  */
9227                 if ((stored_format == SSD_TYPE_FIXED)
9228                  && (sense_format == SSD_TYPE_DESC))
9229                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9230                             ps, (struct scsi_sense_data_desc *)sense_ptr);
9231                 else if ((stored_format == SSD_TYPE_DESC)
9232                       && (sense_format == SSD_TYPE_FIXED))
9233                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9234                             ps, (struct scsi_sense_data_fixed *)sense_ptr);
9235                 else
9236                         memcpy(sense_ptr, ps, sizeof(*sense_ptr));
9237
9238                 ps->error_code = 0;
9239                 have_error = 1;
9240         } else {
9241                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9242                     sense_format);
9243                 if (ua_type != CTL_UA_NONE)
9244                         have_error = 1;
9245         }
9246         if (have_error == 0) {
9247                 /*
9248                  * Report informational exception if have one and allowed.
9249                  */
9250                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9251                         asc = lun->ie_asc;
9252                         ascq = lun->ie_ascq;
9253                 }
9254                 ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9255                     /*current_error*/ 1,
9256                     /*sense_key*/ SSD_KEY_NO_SENSE,
9257                     /*asc*/ asc,
9258                     /*ascq*/ ascq,
9259                     SSD_ELEM_NONE);
9260         }
9261         mtx_unlock(&lun->lun_lock);
9262
9263 send:
9264         /*
9265          * We report the SCSI status as OK, since the status of the command
9266          * itself is OK.  We're reporting sense as parameter data.
9267          */
9268         ctl_set_success(ctsio);
9269         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9270         ctsio->be_move_done = ctl_config_move_done;
9271         ctl_datamove((union ctl_io *)ctsio);
9272         return (CTL_RETVAL_COMPLETE);
9273 }
9274
9275 int
9276 ctl_tur(struct ctl_scsiio *ctsio)
9277 {
9278
9279         CTL_DEBUG_PRINT(("ctl_tur\n"));
9280
9281         ctl_set_success(ctsio);
9282         ctl_done((union ctl_io *)ctsio);
9283
9284         return (CTL_RETVAL_COMPLETE);
9285 }
9286
9287 /*
9288  * SCSI VPD page 0x00, the Supported VPD Pages page.
9289  */
9290 static int
9291 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9292 {
9293         struct ctl_lun *lun = CTL_LUN(ctsio);
9294         struct scsi_vpd_supported_pages *pages;
9295         int sup_page_size;
9296         int p;
9297
9298         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9299             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9300         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9301         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9302         ctsio->kern_rel_offset = 0;
9303         ctsio->kern_sg_entries = 0;
9304         ctsio->kern_data_len = min(sup_page_size, alloc_len);
9305         ctsio->kern_total_len = ctsio->kern_data_len;
9306
9307         /*
9308          * The control device is always connected.  The disk device, on the
9309          * other hand, may not be online all the time.  Need to change this
9310          * to figure out whether the disk device is actually online or not.
9311          */
9312         if (lun != NULL)
9313                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9314                                 lun->be_lun->lun_type;
9315         else
9316                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9317
9318         p = 0;
9319         /* Supported VPD pages */
9320         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9321         /* Serial Number */
9322         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9323         /* Device Identification */
9324         pages->page_list[p++] = SVPD_DEVICE_ID;
9325         /* Extended INQUIRY Data */
9326         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9327         /* Mode Page Policy */
9328         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9329         /* SCSI Ports */
9330         pages->page_list[p++] = SVPD_SCSI_PORTS;
9331         /* Third-party Copy */
9332         pages->page_list[p++] = SVPD_SCSI_TPC;
9333         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9334                 /* Block limits */
9335                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9336                 /* Block Device Characteristics */
9337                 pages->page_list[p++] = SVPD_BDC;
9338                 /* Logical Block Provisioning */
9339                 pages->page_list[p++] = SVPD_LBP;
9340         }
9341         pages->length = p;
9342
9343         ctl_set_success(ctsio);
9344         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9345         ctsio->be_move_done = ctl_config_move_done;
9346         ctl_datamove((union ctl_io *)ctsio);
9347         return (CTL_RETVAL_COMPLETE);
9348 }
9349
9350 /*
9351  * SCSI VPD page 0x80, the Unit Serial Number page.
9352  */
9353 static int
9354 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9355 {
9356         struct ctl_lun *lun = CTL_LUN(ctsio);
9357         struct scsi_vpd_unit_serial_number *sn_ptr;
9358         int data_len;
9359
9360         data_len = 4 + CTL_SN_LEN;
9361         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9362         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9363         ctsio->kern_rel_offset = 0;
9364         ctsio->kern_sg_entries = 0;
9365         ctsio->kern_data_len = min(data_len, alloc_len);
9366         ctsio->kern_total_len = ctsio->kern_data_len;
9367
9368         /*
9369          * The control device is always connected.  The disk device, on the
9370          * other hand, may not be online all the time.  Need to change this
9371          * to figure out whether the disk device is actually online or not.
9372          */
9373         if (lun != NULL)
9374                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9375                                   lun->be_lun->lun_type;
9376         else
9377                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9378
9379         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9380         sn_ptr->length = CTL_SN_LEN;
9381         /*
9382          * If we don't have a LUN, we just leave the serial number as
9383          * all spaces.
9384          */
9385         if (lun != NULL) {
9386                 strncpy((char *)sn_ptr->serial_num,
9387                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9388         } else
9389                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9390
9391         ctl_set_success(ctsio);
9392         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9393         ctsio->be_move_done = ctl_config_move_done;
9394         ctl_datamove((union ctl_io *)ctsio);
9395         return (CTL_RETVAL_COMPLETE);
9396 }
9397
9398
9399 /*
9400  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9401  */
9402 static int
9403 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9404 {
9405         struct ctl_lun *lun = CTL_LUN(ctsio);
9406         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9407         int data_len;
9408
9409         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9410         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9411         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9412         ctsio->kern_sg_entries = 0;
9413         ctsio->kern_rel_offset = 0;
9414         ctsio->kern_data_len = min(data_len, alloc_len);
9415         ctsio->kern_total_len = ctsio->kern_data_len;
9416
9417         /*
9418          * The control device is always connected.  The disk device, on the
9419          * other hand, may not be online all the time.
9420          */
9421         if (lun != NULL)
9422                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9423                                      lun->be_lun->lun_type;
9424         else
9425                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9426         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9427         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9428         /*
9429          * We support head of queue, ordered and simple tags.
9430          */
9431         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9432         /*
9433          * Volatile cache supported.
9434          */
9435         eid_ptr->flags3 = SVPD_EID_V_SUP;
9436
9437         /*
9438          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9439          * attention for a particular IT nexus on all LUNs once we report
9440          * it to that nexus once.  This bit is required as of SPC-4.
9441          */
9442         eid_ptr->flags4 = SVPD_EID_LUICLR;
9443
9444         /*
9445          * We support revert to defaults (RTD) bit in MODE SELECT.
9446          */
9447         eid_ptr->flags5 = SVPD_EID_RTD_SUP;
9448
9449         /*
9450          * XXX KDM in order to correctly answer this, we would need
9451          * information from the SIM to determine how much sense data it
9452          * can send.  So this would really be a path inquiry field, most
9453          * likely.  This can be set to a maximum of 252 according to SPC-4,
9454          * but the hardware may or may not be able to support that much.
9455          * 0 just means that the maximum sense data length is not reported.
9456          */
9457         eid_ptr->max_sense_length = 0;
9458
9459         ctl_set_success(ctsio);
9460         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9461         ctsio->be_move_done = ctl_config_move_done;
9462         ctl_datamove((union ctl_io *)ctsio);
9463         return (CTL_RETVAL_COMPLETE);
9464 }
9465
9466 static int
9467 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9468 {
9469         struct ctl_lun *lun = CTL_LUN(ctsio);
9470         struct scsi_vpd_mode_page_policy *mpp_ptr;
9471         int data_len;
9472
9473         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9474             sizeof(struct scsi_vpd_mode_page_policy_descr);
9475
9476         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9477         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9478         ctsio->kern_rel_offset = 0;
9479         ctsio->kern_sg_entries = 0;
9480         ctsio->kern_data_len = min(data_len, alloc_len);
9481         ctsio->kern_total_len = ctsio->kern_data_len;
9482
9483         /*
9484          * The control device is always connected.  The disk device, on the
9485          * other hand, may not be online all the time.
9486          */
9487         if (lun != NULL)
9488                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9489                                      lun->be_lun->lun_type;
9490         else
9491                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9492         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9493         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9494         mpp_ptr->descr[0].page_code = 0x3f;
9495         mpp_ptr->descr[0].subpage_code = 0xff;
9496         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9497
9498         ctl_set_success(ctsio);
9499         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9500         ctsio->be_move_done = ctl_config_move_done;
9501         ctl_datamove((union ctl_io *)ctsio);
9502         return (CTL_RETVAL_COMPLETE);
9503 }
9504
9505 /*
9506  * SCSI VPD page 0x83, the Device Identification page.
9507  */
9508 static int
9509 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9510 {
9511         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9512         struct ctl_port *port = CTL_PORT(ctsio);
9513         struct ctl_lun *lun = CTL_LUN(ctsio);
9514         struct scsi_vpd_device_id *devid_ptr;
9515         struct scsi_vpd_id_descriptor *desc;
9516         int data_len, g;
9517         uint8_t proto;
9518
9519         data_len = sizeof(struct scsi_vpd_device_id) +
9520             sizeof(struct scsi_vpd_id_descriptor) +
9521                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9522             sizeof(struct scsi_vpd_id_descriptor) +
9523                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9524         if (lun && lun->lun_devid)
9525                 data_len += lun->lun_devid->len;
9526         if (port && port->port_devid)
9527                 data_len += port->port_devid->len;
9528         if (port && port->target_devid)
9529                 data_len += port->target_devid->len;
9530
9531         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9532         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9533         ctsio->kern_sg_entries = 0;
9534         ctsio->kern_rel_offset = 0;
9535         ctsio->kern_sg_entries = 0;
9536         ctsio->kern_data_len = min(data_len, alloc_len);
9537         ctsio->kern_total_len = ctsio->kern_data_len;
9538
9539         /*
9540          * The control device is always connected.  The disk device, on the
9541          * other hand, may not be online all the time.
9542          */
9543         if (lun != NULL)
9544                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9545                                      lun->be_lun->lun_type;
9546         else
9547                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9548         devid_ptr->page_code = SVPD_DEVICE_ID;
9549         scsi_ulto2b(data_len - 4, devid_ptr->length);
9550
9551         if (port && port->port_type == CTL_PORT_FC)
9552                 proto = SCSI_PROTO_FC << 4;
9553         else if (port && port->port_type == CTL_PORT_SAS)
9554                 proto = SCSI_PROTO_SAS << 4;
9555         else if (port && port->port_type == CTL_PORT_ISCSI)
9556                 proto = SCSI_PROTO_ISCSI << 4;
9557         else
9558                 proto = SCSI_PROTO_SPI << 4;
9559         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9560
9561         /*
9562          * We're using a LUN association here.  i.e., this device ID is a
9563          * per-LUN identifier.
9564          */
9565         if (lun && lun->lun_devid) {
9566                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9567                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9568                     lun->lun_devid->len);
9569         }
9570
9571         /*
9572          * This is for the WWPN which is a port association.
9573          */
9574         if (port && port->port_devid) {
9575                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9576                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9577                     port->port_devid->len);
9578         }
9579
9580         /*
9581          * This is for the Relative Target Port(type 4h) identifier
9582          */
9583         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9584         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9585             SVPD_ID_TYPE_RELTARG;
9586         desc->length = 4;
9587         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9588         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9589             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9590
9591         /*
9592          * This is for the Target Port Group(type 5h) identifier
9593          */
9594         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9595         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9596             SVPD_ID_TYPE_TPORTGRP;
9597         desc->length = 4;
9598         if (softc->is_single ||
9599             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9600                 g = 1;
9601         else
9602                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9603         scsi_ulto2b(g, &desc->identifier[2]);
9604         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9605             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9606
9607         /*
9608          * This is for the Target identifier
9609          */
9610         if (port && port->target_devid) {
9611                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9612         }
9613
9614         ctl_set_success(ctsio);
9615         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9616         ctsio->be_move_done = ctl_config_move_done;
9617         ctl_datamove((union ctl_io *)ctsio);
9618         return (CTL_RETVAL_COMPLETE);
9619 }
9620
9621 static int
9622 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9623 {
9624         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9625         struct ctl_lun *lun = CTL_LUN(ctsio);
9626         struct scsi_vpd_scsi_ports *sp;
9627         struct scsi_vpd_port_designation *pd;
9628         struct scsi_vpd_port_designation_cont *pdc;
9629         struct ctl_port *port;
9630         int data_len, num_target_ports, iid_len, id_len;
9631
9632         num_target_ports = 0;
9633         iid_len = 0;
9634         id_len = 0;
9635         mtx_lock(&softc->ctl_lock);
9636         STAILQ_FOREACH(port, &softc->port_list, links) {
9637                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9638                         continue;
9639                 if (lun != NULL &&
9640                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9641                         continue;
9642                 num_target_ports++;
9643                 if (port->init_devid)
9644                         iid_len += port->init_devid->len;
9645                 if (port->port_devid)
9646                         id_len += port->port_devid->len;
9647         }
9648         mtx_unlock(&softc->ctl_lock);
9649
9650         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9651             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9652              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9653         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9654         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9655         ctsio->kern_sg_entries = 0;
9656         ctsio->kern_rel_offset = 0;
9657         ctsio->kern_sg_entries = 0;
9658         ctsio->kern_data_len = min(data_len, alloc_len);
9659         ctsio->kern_total_len = ctsio->kern_data_len;
9660
9661         /*
9662          * The control device is always connected.  The disk device, on the
9663          * other hand, may not be online all the time.  Need to change this
9664          * to figure out whether the disk device is actually online or not.
9665          */
9666         if (lun != NULL)
9667                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9668                                   lun->be_lun->lun_type;
9669         else
9670                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9671
9672         sp->page_code = SVPD_SCSI_PORTS;
9673         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9674             sp->page_length);
9675         pd = &sp->design[0];
9676
9677         mtx_lock(&softc->ctl_lock);
9678         STAILQ_FOREACH(port, &softc->port_list, links) {
9679                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9680                         continue;
9681                 if (lun != NULL &&
9682                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9683                         continue;
9684                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9685                 if (port->init_devid) {
9686                         iid_len = port->init_devid->len;
9687                         memcpy(pd->initiator_transportid,
9688                             port->init_devid->data, port->init_devid->len);
9689                 } else
9690                         iid_len = 0;
9691                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9692                 pdc = (struct scsi_vpd_port_designation_cont *)
9693                     (&pd->initiator_transportid[iid_len]);
9694                 if (port->port_devid) {
9695                         id_len = port->port_devid->len;
9696                         memcpy(pdc->target_port_descriptors,
9697                             port->port_devid->data, port->port_devid->len);
9698                 } else
9699                         id_len = 0;
9700                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9701                 pd = (struct scsi_vpd_port_designation *)
9702                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9703         }
9704         mtx_unlock(&softc->ctl_lock);
9705
9706         ctl_set_success(ctsio);
9707         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9708         ctsio->be_move_done = ctl_config_move_done;
9709         ctl_datamove((union ctl_io *)ctsio);
9710         return (CTL_RETVAL_COMPLETE);
9711 }
9712
9713 static int
9714 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9715 {
9716         struct ctl_lun *lun = CTL_LUN(ctsio);
9717         struct scsi_vpd_block_limits *bl_ptr;
9718         uint64_t ival;
9719
9720         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9721         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9722         ctsio->kern_sg_entries = 0;
9723         ctsio->kern_rel_offset = 0;
9724         ctsio->kern_sg_entries = 0;
9725         ctsio->kern_data_len = min(sizeof(*bl_ptr), alloc_len);
9726         ctsio->kern_total_len = ctsio->kern_data_len;
9727
9728         /*
9729          * The control device is always connected.  The disk device, on the
9730          * other hand, may not be online all the time.  Need to change this
9731          * to figure out whether the disk device is actually online or not.
9732          */
9733         if (lun != NULL)
9734                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9735                                   lun->be_lun->lun_type;
9736         else
9737                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9738
9739         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9740         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9741         bl_ptr->max_cmp_write_len = 0xff;
9742         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9743         if (lun != NULL) {
9744                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9745                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9746                         ival = 0xffffffff;
9747                         ctl_get_opt_number(&lun->be_lun->options,
9748                             "unmap_max_lba", &ival);
9749                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9750                         ival = 0xffffffff;
9751                         ctl_get_opt_number(&lun->be_lun->options,
9752                             "unmap_max_descr", &ival);
9753                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9754                         if (lun->be_lun->ublockexp != 0) {
9755                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9756                                     bl_ptr->opt_unmap_grain);
9757                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9758                                     bl_ptr->unmap_grain_align);
9759                         }
9760                 }
9761                 scsi_ulto4b(lun->be_lun->atomicblock,
9762                     bl_ptr->max_atomic_transfer_length);
9763                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9764                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9765                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9766                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9767                 ival = UINT64_MAX;
9768                 ctl_get_opt_number(&lun->be_lun->options, "write_same_max_lba", &ival);
9769                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9770         }
9771
9772         ctl_set_success(ctsio);
9773         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9774         ctsio->be_move_done = ctl_config_move_done;
9775         ctl_datamove((union ctl_io *)ctsio);
9776         return (CTL_RETVAL_COMPLETE);
9777 }
9778
9779 static int
9780 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9781 {
9782         struct ctl_lun *lun = CTL_LUN(ctsio);
9783         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9784         const char *value;
9785         u_int i;
9786
9787         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9788         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9789         ctsio->kern_sg_entries = 0;
9790         ctsio->kern_rel_offset = 0;
9791         ctsio->kern_data_len = min(sizeof(*bdc_ptr), alloc_len);
9792         ctsio->kern_total_len = ctsio->kern_data_len;
9793
9794         /*
9795          * The control device is always connected.  The disk device, on the
9796          * other hand, may not be online all the time.  Need to change this
9797          * to figure out whether the disk device is actually online or not.
9798          */
9799         if (lun != NULL)
9800                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9801                                   lun->be_lun->lun_type;
9802         else
9803                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9804         bdc_ptr->page_code = SVPD_BDC;
9805         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9806         if (lun != NULL &&
9807             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9808                 i = strtol(value, NULL, 0);
9809         else
9810                 i = CTL_DEFAULT_ROTATION_RATE;
9811         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9812         if (lun != NULL &&
9813             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9814                 i = strtol(value, NULL, 0);
9815         else
9816                 i = 0;
9817         bdc_ptr->wab_wac_ff = (i & 0x0f);
9818         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9819
9820         ctl_set_success(ctsio);
9821         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9822         ctsio->be_move_done = ctl_config_move_done;
9823         ctl_datamove((union ctl_io *)ctsio);
9824         return (CTL_RETVAL_COMPLETE);
9825 }
9826
9827 static int
9828 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9829 {
9830         struct ctl_lun *lun = CTL_LUN(ctsio);
9831         struct scsi_vpd_logical_block_prov *lbp_ptr;
9832         const char *value;
9833
9834         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9835         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9836         ctsio->kern_sg_entries = 0;
9837         ctsio->kern_rel_offset = 0;
9838         ctsio->kern_data_len = min(sizeof(*lbp_ptr), alloc_len);
9839         ctsio->kern_total_len = ctsio->kern_data_len;
9840
9841         /*
9842          * The control device is always connected.  The disk device, on the
9843          * other hand, may not be online all the time.  Need to change this
9844          * to figure out whether the disk device is actually online or not.
9845          */
9846         if (lun != NULL)
9847                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9848                                   lun->be_lun->lun_type;
9849         else
9850                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9851
9852         lbp_ptr->page_code = SVPD_LBP;
9853         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9854         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
9855         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9856                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
9857                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
9858                 value = ctl_get_opt(&lun->be_lun->options, "provisioning_type");
9859                 if (value != NULL) {
9860                         if (strcmp(value, "resource") == 0)
9861                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
9862                         else if (strcmp(value, "thin") == 0)
9863                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
9864                 } else
9865                         lbp_ptr->prov_type = SVPD_LBP_THIN;
9866         }
9867
9868         ctl_set_success(ctsio);
9869         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9870         ctsio->be_move_done = ctl_config_move_done;
9871         ctl_datamove((union ctl_io *)ctsio);
9872         return (CTL_RETVAL_COMPLETE);
9873 }
9874
9875 /*
9876  * INQUIRY with the EVPD bit set.
9877  */
9878 static int
9879 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9880 {
9881         struct ctl_lun *lun = CTL_LUN(ctsio);
9882         struct scsi_inquiry *cdb;
9883         int alloc_len, retval;
9884
9885         cdb = (struct scsi_inquiry *)ctsio->cdb;
9886         alloc_len = scsi_2btoul(cdb->length);
9887
9888         switch (cdb->page_code) {
9889         case SVPD_SUPPORTED_PAGES:
9890                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9891                 break;
9892         case SVPD_UNIT_SERIAL_NUMBER:
9893                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9894                 break;
9895         case SVPD_DEVICE_ID:
9896                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9897                 break;
9898         case SVPD_EXTENDED_INQUIRY_DATA:
9899                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
9900                 break;
9901         case SVPD_MODE_PAGE_POLICY:
9902                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
9903                 break;
9904         case SVPD_SCSI_PORTS:
9905                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
9906                 break;
9907         case SVPD_SCSI_TPC:
9908                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
9909                 break;
9910         case SVPD_BLOCK_LIMITS:
9911                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9912                         goto err;
9913                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9914                 break;
9915         case SVPD_BDC:
9916                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9917                         goto err;
9918                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
9919                 break;
9920         case SVPD_LBP:
9921                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9922                         goto err;
9923                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9924                 break;
9925         default:
9926 err:
9927                 ctl_set_invalid_field(ctsio,
9928                                       /*sks_valid*/ 1,
9929                                       /*command*/ 1,
9930                                       /*field*/ 2,
9931                                       /*bit_valid*/ 0,
9932                                       /*bit*/ 0);
9933                 ctl_done((union ctl_io *)ctsio);
9934                 retval = CTL_RETVAL_COMPLETE;
9935                 break;
9936         }
9937
9938         return (retval);
9939 }
9940
9941 /*
9942  * Standard INQUIRY data.
9943  */
9944 static int
9945 ctl_inquiry_std(struct ctl_scsiio *ctsio)
9946 {
9947         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9948         struct ctl_port *port = CTL_PORT(ctsio);
9949         struct ctl_lun *lun = CTL_LUN(ctsio);
9950         struct scsi_inquiry_data *inq_ptr;
9951         struct scsi_inquiry *cdb;
9952         char *val;
9953         uint32_t alloc_len, data_len;
9954         ctl_port_type port_type;
9955
9956         port_type = port->port_type;
9957         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
9958                 port_type = CTL_PORT_SCSI;
9959
9960         cdb = (struct scsi_inquiry *)ctsio->cdb;
9961         alloc_len = scsi_2btoul(cdb->length);
9962
9963         /*
9964          * We malloc the full inquiry data size here and fill it
9965          * in.  If the user only asks for less, we'll give him
9966          * that much.
9967          */
9968         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
9969         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9970         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9971         ctsio->kern_sg_entries = 0;
9972         ctsio->kern_rel_offset = 0;
9973         ctsio->kern_data_len = min(data_len, alloc_len);
9974         ctsio->kern_total_len = ctsio->kern_data_len;
9975
9976         if (lun != NULL) {
9977                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
9978                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
9979                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9980                             lun->be_lun->lun_type;
9981                 } else {
9982                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
9983                             lun->be_lun->lun_type;
9984                 }
9985                 if (lun->flags & CTL_LUN_REMOVABLE)
9986                         inq_ptr->dev_qual2 |= SID_RMB;
9987         } else
9988                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9989
9990         /* RMB in byte 2 is 0 */
9991         inq_ptr->version = SCSI_REV_SPC5;
9992
9993         /*
9994          * According to SAM-3, even if a device only supports a single
9995          * level of LUN addressing, it should still set the HISUP bit:
9996          *
9997          * 4.9.1 Logical unit numbers overview
9998          *
9999          * All logical unit number formats described in this standard are
10000          * hierarchical in structure even when only a single level in that
10001          * hierarchy is used. The HISUP bit shall be set to one in the
10002          * standard INQUIRY data (see SPC-2) when any logical unit number
10003          * format described in this standard is used.  Non-hierarchical
10004          * formats are outside the scope of this standard.
10005          *
10006          * Therefore we set the HiSup bit here.
10007          *
10008          * The response format is 2, per SPC-3.
10009          */
10010         inq_ptr->response_format = SID_HiSup | 2;
10011
10012         inq_ptr->additional_length = data_len -
10013             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10014         CTL_DEBUG_PRINT(("additional_length = %d\n",
10015                          inq_ptr->additional_length));
10016
10017         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10018         if (port_type == CTL_PORT_SCSI)
10019                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10020         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10021         inq_ptr->flags = SID_CmdQue;
10022         if (port_type == CTL_PORT_SCSI)
10023                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10024
10025         /*
10026          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10027          * We have 8 bytes for the vendor name, and 16 bytes for the device
10028          * name and 4 bytes for the revision.
10029          */
10030         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10031             "vendor")) == NULL) {
10032                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10033         } else {
10034                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10035                 strncpy(inq_ptr->vendor, val,
10036                     min(sizeof(inq_ptr->vendor), strlen(val)));
10037         }
10038         if (lun == NULL) {
10039                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10040                     sizeof(inq_ptr->product));
10041         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10042                 switch (lun->be_lun->lun_type) {
10043                 case T_DIRECT:
10044                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10045                             sizeof(inq_ptr->product));
10046                         break;
10047                 case T_PROCESSOR:
10048                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10049                             sizeof(inq_ptr->product));
10050                         break;
10051                 case T_CDROM:
10052                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10053                             sizeof(inq_ptr->product));
10054                         break;
10055                 default:
10056                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10057                             sizeof(inq_ptr->product));
10058                         break;
10059                 }
10060         } else {
10061                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10062                 strncpy(inq_ptr->product, val,
10063                     min(sizeof(inq_ptr->product), strlen(val)));
10064         }
10065
10066         /*
10067          * XXX make this a macro somewhere so it automatically gets
10068          * incremented when we make changes.
10069          */
10070         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10071             "revision")) == NULL) {
10072                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10073         } else {
10074                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10075                 strncpy(inq_ptr->revision, val,
10076                     min(sizeof(inq_ptr->revision), strlen(val)));
10077         }
10078
10079         /*
10080          * For parallel SCSI, we support double transition and single
10081          * transition clocking.  We also support QAS (Quick Arbitration
10082          * and Selection) and Information Unit transfers on both the
10083          * control and array devices.
10084          */
10085         if (port_type == CTL_PORT_SCSI)
10086                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10087                                     SID_SPI_IUS;
10088
10089         /* SAM-6 (no version claimed) */
10090         scsi_ulto2b(0x00C0, inq_ptr->version1);
10091         /* SPC-5 (no version claimed) */
10092         scsi_ulto2b(0x05C0, inq_ptr->version2);
10093         if (port_type == CTL_PORT_FC) {
10094                 /* FCP-2 ANSI INCITS.350:2003 */
10095                 scsi_ulto2b(0x0917, inq_ptr->version3);
10096         } else if (port_type == CTL_PORT_SCSI) {
10097                 /* SPI-4 ANSI INCITS.362:200x */
10098                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10099         } else if (port_type == CTL_PORT_ISCSI) {
10100                 /* iSCSI (no version claimed) */
10101                 scsi_ulto2b(0x0960, inq_ptr->version3);
10102         } else if (port_type == CTL_PORT_SAS) {
10103                 /* SAS (no version claimed) */
10104                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10105         } else if (port_type == CTL_PORT_UMASS) {
10106                 /* USB Mass Storage Class Bulk-Only Transport, Revision 1.0 */
10107                 scsi_ulto2b(0x1730, inq_ptr->version3);
10108         }
10109
10110         if (lun == NULL) {
10111                 /* SBC-4 (no version claimed) */
10112                 scsi_ulto2b(0x0600, inq_ptr->version4);
10113         } else {
10114                 switch (lun->be_lun->lun_type) {
10115                 case T_DIRECT:
10116                         /* SBC-4 (no version claimed) */
10117                         scsi_ulto2b(0x0600, inq_ptr->version4);
10118                         break;
10119                 case T_PROCESSOR:
10120                         break;
10121                 case T_CDROM:
10122                         /* MMC-6 (no version claimed) */
10123                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10124                         break;
10125                 default:
10126                         break;
10127                 }
10128         }
10129
10130         ctl_set_success(ctsio);
10131         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10132         ctsio->be_move_done = ctl_config_move_done;
10133         ctl_datamove((union ctl_io *)ctsio);
10134         return (CTL_RETVAL_COMPLETE);
10135 }
10136
10137 int
10138 ctl_inquiry(struct ctl_scsiio *ctsio)
10139 {
10140         struct scsi_inquiry *cdb;
10141         int retval;
10142
10143         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10144
10145         cdb = (struct scsi_inquiry *)ctsio->cdb;
10146         if (cdb->byte2 & SI_EVPD)
10147                 retval = ctl_inquiry_evpd(ctsio);
10148         else if (cdb->page_code == 0)
10149                 retval = ctl_inquiry_std(ctsio);
10150         else {
10151                 ctl_set_invalid_field(ctsio,
10152                                       /*sks_valid*/ 1,
10153                                       /*command*/ 1,
10154                                       /*field*/ 2,
10155                                       /*bit_valid*/ 0,
10156                                       /*bit*/ 0);
10157                 ctl_done((union ctl_io *)ctsio);
10158                 return (CTL_RETVAL_COMPLETE);
10159         }
10160
10161         return (retval);
10162 }
10163
10164 int
10165 ctl_get_config(struct ctl_scsiio *ctsio)
10166 {
10167         struct ctl_lun *lun = CTL_LUN(ctsio);
10168         struct scsi_get_config_header *hdr;
10169         struct scsi_get_config_feature *feature;
10170         struct scsi_get_config *cdb;
10171         uint32_t alloc_len, data_len;
10172         int rt, starting;
10173
10174         cdb = (struct scsi_get_config *)ctsio->cdb;
10175         rt = (cdb->rt & SGC_RT_MASK);
10176         starting = scsi_2btoul(cdb->starting_feature);
10177         alloc_len = scsi_2btoul(cdb->length);
10178
10179         data_len = sizeof(struct scsi_get_config_header) +
10180             sizeof(struct scsi_get_config_feature) + 8 +
10181             sizeof(struct scsi_get_config_feature) + 8 +
10182             sizeof(struct scsi_get_config_feature) + 4 +
10183             sizeof(struct scsi_get_config_feature) + 4 +
10184             sizeof(struct scsi_get_config_feature) + 8 +
10185             sizeof(struct scsi_get_config_feature) +
10186             sizeof(struct scsi_get_config_feature) + 4 +
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         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10193         ctsio->kern_sg_entries = 0;
10194         ctsio->kern_rel_offset = 0;
10195
10196         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10197         if (lun->flags & CTL_LUN_NO_MEDIA)
10198                 scsi_ulto2b(0x0000, hdr->current_profile);
10199         else
10200                 scsi_ulto2b(0x0010, hdr->current_profile);
10201         feature = (struct scsi_get_config_feature *)(hdr + 1);
10202
10203         if (starting > 0x003b)
10204                 goto done;
10205         if (starting > 0x003a)
10206                 goto f3b;
10207         if (starting > 0x002b)
10208                 goto f3a;
10209         if (starting > 0x002a)
10210                 goto f2b;
10211         if (starting > 0x001f)
10212                 goto f2a;
10213         if (starting > 0x001e)
10214                 goto f1f;
10215         if (starting > 0x001d)
10216                 goto f1e;
10217         if (starting > 0x0010)
10218                 goto f1d;
10219         if (starting > 0x0003)
10220                 goto f10;
10221         if (starting > 0x0002)
10222                 goto f3;
10223         if (starting > 0x0001)
10224                 goto f2;
10225         if (starting > 0x0000)
10226                 goto f1;
10227
10228         /* Profile List */
10229         scsi_ulto2b(0x0000, feature->feature_code);
10230         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10231         feature->add_length = 8;
10232         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10233         feature->feature_data[2] = 0x00;
10234         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10235         feature->feature_data[6] = 0x01;
10236         feature = (struct scsi_get_config_feature *)
10237             &feature->feature_data[feature->add_length];
10238
10239 f1:     /* Core */
10240         scsi_ulto2b(0x0001, feature->feature_code);
10241         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10242         feature->add_length = 8;
10243         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10244         feature->feature_data[4] = 0x03;
10245         feature = (struct scsi_get_config_feature *)
10246             &feature->feature_data[feature->add_length];
10247
10248 f2:     /* Morphing */
10249         scsi_ulto2b(0x0002, feature->feature_code);
10250         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10251         feature->add_length = 4;
10252         feature->feature_data[0] = 0x02;
10253         feature = (struct scsi_get_config_feature *)
10254             &feature->feature_data[feature->add_length];
10255
10256 f3:     /* Removable Medium */
10257         scsi_ulto2b(0x0003, feature->feature_code);
10258         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10259         feature->add_length = 4;
10260         feature->feature_data[0] = 0x39;
10261         feature = (struct scsi_get_config_feature *)
10262             &feature->feature_data[feature->add_length];
10263
10264         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10265                 goto done;
10266
10267 f10:    /* Random Read */
10268         scsi_ulto2b(0x0010, feature->feature_code);
10269         feature->flags = 0x00;
10270         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10271                 feature->flags |= SGC_F_CURRENT;
10272         feature->add_length = 8;
10273         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10274         scsi_ulto2b(1, &feature->feature_data[4]);
10275         feature->feature_data[6] = 0x00;
10276         feature = (struct scsi_get_config_feature *)
10277             &feature->feature_data[feature->add_length];
10278
10279 f1d:    /* Multi-Read */
10280         scsi_ulto2b(0x001D, feature->feature_code);
10281         feature->flags = 0x00;
10282         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10283                 feature->flags |= SGC_F_CURRENT;
10284         feature->add_length = 0;
10285         feature = (struct scsi_get_config_feature *)
10286             &feature->feature_data[feature->add_length];
10287
10288 f1e:    /* CD Read */
10289         scsi_ulto2b(0x001E, feature->feature_code);
10290         feature->flags = 0x00;
10291         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10292                 feature->flags |= SGC_F_CURRENT;
10293         feature->add_length = 4;
10294         feature->feature_data[0] = 0x00;
10295         feature = (struct scsi_get_config_feature *)
10296             &feature->feature_data[feature->add_length];
10297
10298 f1f:    /* DVD Read */
10299         scsi_ulto2b(0x001F, feature->feature_code);
10300         feature->flags = 0x08;
10301         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10302                 feature->flags |= SGC_F_CURRENT;
10303         feature->add_length = 4;
10304         feature->feature_data[0] = 0x01;
10305         feature->feature_data[2] = 0x03;
10306         feature = (struct scsi_get_config_feature *)
10307             &feature->feature_data[feature->add_length];
10308
10309 f2a:    /* DVD+RW */
10310         scsi_ulto2b(0x002A, feature->feature_code);
10311         feature->flags = 0x04;
10312         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10313                 feature->flags |= SGC_F_CURRENT;
10314         feature->add_length = 4;
10315         feature->feature_data[0] = 0x00;
10316         feature->feature_data[1] = 0x00;
10317         feature = (struct scsi_get_config_feature *)
10318             &feature->feature_data[feature->add_length];
10319
10320 f2b:    /* DVD+R */
10321         scsi_ulto2b(0x002B, feature->feature_code);
10322         feature->flags = 0x00;
10323         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10324                 feature->flags |= SGC_F_CURRENT;
10325         feature->add_length = 4;
10326         feature->feature_data[0] = 0x00;
10327         feature = (struct scsi_get_config_feature *)
10328             &feature->feature_data[feature->add_length];
10329
10330 f3a:    /* DVD+RW Dual Layer */
10331         scsi_ulto2b(0x003A, feature->feature_code);
10332         feature->flags = 0x00;
10333         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10334                 feature->flags |= SGC_F_CURRENT;
10335         feature->add_length = 4;
10336         feature->feature_data[0] = 0x00;
10337         feature->feature_data[1] = 0x00;
10338         feature = (struct scsi_get_config_feature *)
10339             &feature->feature_data[feature->add_length];
10340
10341 f3b:    /* DVD+R Dual Layer */
10342         scsi_ulto2b(0x003B, feature->feature_code);
10343         feature->flags = 0x00;
10344         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10345                 feature->flags |= SGC_F_CURRENT;
10346         feature->add_length = 4;
10347         feature->feature_data[0] = 0x00;
10348         feature = (struct scsi_get_config_feature *)
10349             &feature->feature_data[feature->add_length];
10350
10351 done:
10352         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10353         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10354                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10355                 if (scsi_2btoul(feature->feature_code) == starting)
10356                         feature = (struct scsi_get_config_feature *)
10357                             &feature->feature_data[feature->add_length];
10358                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10359         }
10360         scsi_ulto4b(data_len - 4, hdr->data_length);
10361         ctsio->kern_data_len = min(data_len, alloc_len);
10362         ctsio->kern_total_len = ctsio->kern_data_len;
10363
10364         ctl_set_success(ctsio);
10365         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10366         ctsio->be_move_done = ctl_config_move_done;
10367         ctl_datamove((union ctl_io *)ctsio);
10368         return (CTL_RETVAL_COMPLETE);
10369 }
10370
10371 int
10372 ctl_get_event_status(struct ctl_scsiio *ctsio)
10373 {
10374         struct scsi_get_event_status_header *hdr;
10375         struct scsi_get_event_status *cdb;
10376         uint32_t alloc_len, data_len;
10377         int notif_class;
10378
10379         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10380         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10381                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10382                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10383                 ctl_done((union ctl_io *)ctsio);
10384                 return (CTL_RETVAL_COMPLETE);
10385         }
10386         notif_class = cdb->notif_class;
10387         alloc_len = scsi_2btoul(cdb->length);
10388
10389         data_len = sizeof(struct scsi_get_event_status_header);
10390         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10391         ctsio->kern_sg_entries = 0;
10392         ctsio->kern_rel_offset = 0;
10393         ctsio->kern_data_len = min(data_len, alloc_len);
10394         ctsio->kern_total_len = ctsio->kern_data_len;
10395
10396         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10397         scsi_ulto2b(0, hdr->descr_length);
10398         hdr->nea_class = SGESN_NEA;
10399         hdr->supported_class = 0;
10400
10401         ctl_set_success(ctsio);
10402         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10403         ctsio->be_move_done = ctl_config_move_done;
10404         ctl_datamove((union ctl_io *)ctsio);
10405         return (CTL_RETVAL_COMPLETE);
10406 }
10407
10408 int
10409 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10410 {
10411         struct scsi_mechanism_status_header *hdr;
10412         struct scsi_mechanism_status *cdb;
10413         uint32_t alloc_len, data_len;
10414
10415         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10416         alloc_len = scsi_2btoul(cdb->length);
10417
10418         data_len = sizeof(struct scsi_mechanism_status_header);
10419         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10420         ctsio->kern_sg_entries = 0;
10421         ctsio->kern_rel_offset = 0;
10422         ctsio->kern_data_len = min(data_len, alloc_len);
10423         ctsio->kern_total_len = ctsio->kern_data_len;
10424
10425         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10426         hdr->state1 = 0x00;
10427         hdr->state2 = 0xe0;
10428         scsi_ulto3b(0, hdr->lba);
10429         hdr->slots_num = 0;
10430         scsi_ulto2b(0, hdr->slots_length);
10431
10432         ctl_set_success(ctsio);
10433         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10434         ctsio->be_move_done = ctl_config_move_done;
10435         ctl_datamove((union ctl_io *)ctsio);
10436         return (CTL_RETVAL_COMPLETE);
10437 }
10438
10439 static void
10440 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10441 {
10442
10443         lba += 150;
10444         buf[0] = 0;
10445         buf[1] = bin2bcd((lba / 75) / 60);
10446         buf[2] = bin2bcd((lba / 75) % 60);
10447         buf[3] = bin2bcd(lba % 75);
10448 }
10449
10450 int
10451 ctl_read_toc(struct ctl_scsiio *ctsio)
10452 {
10453         struct ctl_lun *lun = CTL_LUN(ctsio);
10454         struct scsi_read_toc_hdr *hdr;
10455         struct scsi_read_toc_type01_descr *descr;
10456         struct scsi_read_toc *cdb;
10457         uint32_t alloc_len, data_len;
10458         int format, msf;
10459
10460         cdb = (struct scsi_read_toc *)ctsio->cdb;
10461         msf = (cdb->byte2 & CD_MSF) != 0;
10462         format = cdb->format;
10463         alloc_len = scsi_2btoul(cdb->data_len);
10464
10465         data_len = sizeof(struct scsi_read_toc_hdr);
10466         if (format == 0)
10467                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10468         else
10469                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10470         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10471         ctsio->kern_sg_entries = 0;
10472         ctsio->kern_rel_offset = 0;
10473         ctsio->kern_data_len = min(data_len, alloc_len);
10474         ctsio->kern_total_len = ctsio->kern_data_len;
10475
10476         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10477         if (format == 0) {
10478                 scsi_ulto2b(0x12, hdr->data_length);
10479                 hdr->first = 1;
10480                 hdr->last = 1;
10481                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10482                 descr->addr_ctl = 0x14;
10483                 descr->track_number = 1;
10484                 if (msf)
10485                         ctl_ultomsf(0, descr->track_start);
10486                 else
10487                         scsi_ulto4b(0, descr->track_start);
10488                 descr++;
10489                 descr->addr_ctl = 0x14;
10490                 descr->track_number = 0xaa;
10491                 if (msf)
10492                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10493                 else
10494                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10495         } else {
10496                 scsi_ulto2b(0x0a, hdr->data_length);
10497                 hdr->first = 1;
10498                 hdr->last = 1;
10499                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10500                 descr->addr_ctl = 0x14;
10501                 descr->track_number = 1;
10502                 if (msf)
10503                         ctl_ultomsf(0, descr->track_start);
10504                 else
10505                         scsi_ulto4b(0, descr->track_start);
10506         }
10507
10508         ctl_set_success(ctsio);
10509         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10510         ctsio->be_move_done = ctl_config_move_done;
10511         ctl_datamove((union ctl_io *)ctsio);
10512         return (CTL_RETVAL_COMPLETE);
10513 }
10514
10515 /*
10516  * For known CDB types, parse the LBA and length.
10517  */
10518 static int
10519 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10520 {
10521         if (io->io_hdr.io_type != CTL_IO_SCSI)
10522                 return (1);
10523
10524         switch (io->scsiio.cdb[0]) {
10525         case COMPARE_AND_WRITE: {
10526                 struct scsi_compare_and_write *cdb;
10527
10528                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10529
10530                 *lba = scsi_8btou64(cdb->addr);
10531                 *len = cdb->length;
10532                 break;
10533         }
10534         case READ_6:
10535         case WRITE_6: {
10536                 struct scsi_rw_6 *cdb;
10537
10538                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10539
10540                 *lba = scsi_3btoul(cdb->addr);
10541                 /* only 5 bits are valid in the most significant address byte */
10542                 *lba &= 0x1fffff;
10543                 *len = cdb->length;
10544                 break;
10545         }
10546         case READ_10:
10547         case WRITE_10: {
10548                 struct scsi_rw_10 *cdb;
10549
10550                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10551
10552                 *lba = scsi_4btoul(cdb->addr);
10553                 *len = scsi_2btoul(cdb->length);
10554                 break;
10555         }
10556         case WRITE_VERIFY_10: {
10557                 struct scsi_write_verify_10 *cdb;
10558
10559                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10560
10561                 *lba = scsi_4btoul(cdb->addr);
10562                 *len = scsi_2btoul(cdb->length);
10563                 break;
10564         }
10565         case READ_12:
10566         case WRITE_12: {
10567                 struct scsi_rw_12 *cdb;
10568
10569                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10570
10571                 *lba = scsi_4btoul(cdb->addr);
10572                 *len = scsi_4btoul(cdb->length);
10573                 break;
10574         }
10575         case WRITE_VERIFY_12: {
10576                 struct scsi_write_verify_12 *cdb;
10577
10578                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10579
10580                 *lba = scsi_4btoul(cdb->addr);
10581                 *len = scsi_4btoul(cdb->length);
10582                 break;
10583         }
10584         case READ_16:
10585         case WRITE_16: {
10586                 struct scsi_rw_16 *cdb;
10587
10588                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10589
10590                 *lba = scsi_8btou64(cdb->addr);
10591                 *len = scsi_4btoul(cdb->length);
10592                 break;
10593         }
10594         case WRITE_ATOMIC_16: {
10595                 struct scsi_write_atomic_16 *cdb;
10596
10597                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10598
10599                 *lba = scsi_8btou64(cdb->addr);
10600                 *len = scsi_2btoul(cdb->length);
10601                 break;
10602         }
10603         case WRITE_VERIFY_16: {
10604                 struct scsi_write_verify_16 *cdb;
10605
10606                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10607
10608                 *lba = scsi_8btou64(cdb->addr);
10609                 *len = scsi_4btoul(cdb->length);
10610                 break;
10611         }
10612         case WRITE_SAME_10: {
10613                 struct scsi_write_same_10 *cdb;
10614
10615                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10616
10617                 *lba = scsi_4btoul(cdb->addr);
10618                 *len = scsi_2btoul(cdb->length);
10619                 break;
10620         }
10621         case WRITE_SAME_16: {
10622                 struct scsi_write_same_16 *cdb;
10623
10624                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10625
10626                 *lba = scsi_8btou64(cdb->addr);
10627                 *len = scsi_4btoul(cdb->length);
10628                 break;
10629         }
10630         case VERIFY_10: {
10631                 struct scsi_verify_10 *cdb;
10632
10633                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10634
10635                 *lba = scsi_4btoul(cdb->addr);
10636                 *len = scsi_2btoul(cdb->length);
10637                 break;
10638         }
10639         case VERIFY_12: {
10640                 struct scsi_verify_12 *cdb;
10641
10642                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10643
10644                 *lba = scsi_4btoul(cdb->addr);
10645                 *len = scsi_4btoul(cdb->length);
10646                 break;
10647         }
10648         case VERIFY_16: {
10649                 struct scsi_verify_16 *cdb;
10650
10651                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10652
10653                 *lba = scsi_8btou64(cdb->addr);
10654                 *len = scsi_4btoul(cdb->length);
10655                 break;
10656         }
10657         case UNMAP: {
10658                 *lba = 0;
10659                 *len = UINT64_MAX;
10660                 break;
10661         }
10662         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10663                 struct scsi_get_lba_status *cdb;
10664
10665                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10666                 *lba = scsi_8btou64(cdb->addr);
10667                 *len = UINT32_MAX;
10668                 break;
10669         }
10670         default:
10671                 return (1);
10672                 break; /* NOTREACHED */
10673         }
10674
10675         return (0);
10676 }
10677
10678 static ctl_action
10679 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10680     bool seq)
10681 {
10682         uint64_t endlba1, endlba2;
10683
10684         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10685         endlba2 = lba2 + len2 - 1;
10686
10687         if ((endlba1 < lba2) || (endlba2 < lba1))
10688                 return (CTL_ACTION_PASS);
10689         else
10690                 return (CTL_ACTION_BLOCK);
10691 }
10692
10693 static int
10694 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10695 {
10696         struct ctl_ptr_len_flags *ptrlen;
10697         struct scsi_unmap_desc *buf, *end, *range;
10698         uint64_t lba;
10699         uint32_t len;
10700
10701         /* If not UNMAP -- go other way. */
10702         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10703             io->scsiio.cdb[0] != UNMAP)
10704                 return (CTL_ACTION_ERROR);
10705
10706         /* If UNMAP without data -- block and wait for data. */
10707         ptrlen = (struct ctl_ptr_len_flags *)
10708             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10709         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10710             ptrlen->ptr == NULL)
10711                 return (CTL_ACTION_BLOCK);
10712
10713         /* UNMAP with data -- check for collision. */
10714         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10715         end = buf + ptrlen->len / sizeof(*buf);
10716         for (range = buf; range < end; range++) {
10717                 lba = scsi_8btou64(range->lba);
10718                 len = scsi_4btoul(range->length);
10719                 if ((lba < lba2 + len2) && (lba + len > lba2))
10720                         return (CTL_ACTION_BLOCK);
10721         }
10722         return (CTL_ACTION_PASS);
10723 }
10724
10725 static ctl_action
10726 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10727 {
10728         uint64_t lba1, lba2;
10729         uint64_t len1, len2;
10730         int retval;
10731
10732         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10733                 return (CTL_ACTION_ERROR);
10734
10735         retval = ctl_extent_check_unmap(io1, lba2, len2);
10736         if (retval != CTL_ACTION_ERROR)
10737                 return (retval);
10738
10739         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10740                 return (CTL_ACTION_ERROR);
10741
10742         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10743                 seq = FALSE;
10744         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10745 }
10746
10747 static ctl_action
10748 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10749 {
10750         uint64_t lba1, lba2;
10751         uint64_t len1, len2;
10752
10753         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10754                 return (CTL_ACTION_PASS);
10755         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10756                 return (CTL_ACTION_ERROR);
10757         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10758                 return (CTL_ACTION_ERROR);
10759
10760         if (lba1 + len1 == lba2)
10761                 return (CTL_ACTION_BLOCK);
10762         return (CTL_ACTION_PASS);
10763 }
10764
10765 static ctl_action
10766 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10767     union ctl_io *ooa_io)
10768 {
10769         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10770         const ctl_serialize_action *serialize_row;
10771
10772         /*
10773          * The initiator attempted multiple untagged commands at the same
10774          * time.  Can't do that.
10775          */
10776         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10777          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10778          && ((pending_io->io_hdr.nexus.targ_port ==
10779               ooa_io->io_hdr.nexus.targ_port)
10780           && (pending_io->io_hdr.nexus.initid ==
10781               ooa_io->io_hdr.nexus.initid))
10782          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10783               CTL_FLAG_STATUS_SENT)) == 0))
10784                 return (CTL_ACTION_OVERLAP);
10785
10786         /*
10787          * The initiator attempted to send multiple tagged commands with
10788          * the same ID.  (It's fine if different initiators have the same
10789          * tag ID.)
10790          *
10791          * Even if all of those conditions are true, we don't kill the I/O
10792          * if the command ahead of us has been aborted.  We won't end up
10793          * sending it to the FETD, and it's perfectly legal to resend a
10794          * command with the same tag number as long as the previous
10795          * instance of this tag number has been aborted somehow.
10796          */
10797         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10798          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10799          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10800          && ((pending_io->io_hdr.nexus.targ_port ==
10801               ooa_io->io_hdr.nexus.targ_port)
10802           && (pending_io->io_hdr.nexus.initid ==
10803               ooa_io->io_hdr.nexus.initid))
10804          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10805               CTL_FLAG_STATUS_SENT)) == 0))
10806                 return (CTL_ACTION_OVERLAP_TAG);
10807
10808         /*
10809          * If we get a head of queue tag, SAM-3 says that we should
10810          * immediately execute it.
10811          *
10812          * What happens if this command would normally block for some other
10813          * reason?  e.g. a request sense with a head of queue tag
10814          * immediately after a write.  Normally that would block, but this
10815          * will result in its getting executed immediately...
10816          *
10817          * We currently return "pass" instead of "skip", so we'll end up
10818          * going through the rest of the queue to check for overlapped tags.
10819          *
10820          * XXX KDM check for other types of blockage first??
10821          */
10822         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10823                 return (CTL_ACTION_PASS);
10824
10825         /*
10826          * Ordered tags have to block until all items ahead of them
10827          * have completed.  If we get called with an ordered tag, we always
10828          * block, if something else is ahead of us in the queue.
10829          */
10830         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10831                 return (CTL_ACTION_BLOCK);
10832
10833         /*
10834          * Simple tags get blocked until all head of queue and ordered tags
10835          * ahead of them have completed.  I'm lumping untagged commands in
10836          * with simple tags here.  XXX KDM is that the right thing to do?
10837          */
10838         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10839           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10840          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10841           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10842                 return (CTL_ACTION_BLOCK);
10843
10844         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10845         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
10846             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
10847              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
10848              pending_io->scsiio.cdb[1], pending_io));
10849         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10850         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
10851                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
10852         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
10853             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
10854              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
10855              ooa_io->scsiio.cdb[1], ooa_io));
10856
10857         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10858
10859         switch (serialize_row[pending_entry->seridx]) {
10860         case CTL_SER_BLOCK:
10861                 return (CTL_ACTION_BLOCK);
10862         case CTL_SER_EXTENT:
10863                 return (ctl_extent_check(ooa_io, pending_io,
10864                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10865         case CTL_SER_EXTENTOPT:
10866                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10867                     SCP_QUEUE_ALG_UNRESTRICTED)
10868                         return (ctl_extent_check(ooa_io, pending_io,
10869                             (lun->be_lun &&
10870                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10871                 return (CTL_ACTION_PASS);
10872         case CTL_SER_EXTENTSEQ:
10873                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10874                         return (ctl_extent_check_seq(ooa_io, pending_io));
10875                 return (CTL_ACTION_PASS);
10876         case CTL_SER_PASS:
10877                 return (CTL_ACTION_PASS);
10878         case CTL_SER_BLOCKOPT:
10879                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10880                     SCP_QUEUE_ALG_UNRESTRICTED)
10881                         return (CTL_ACTION_BLOCK);
10882                 return (CTL_ACTION_PASS);
10883         case CTL_SER_SKIP:
10884                 return (CTL_ACTION_SKIP);
10885         default:
10886                 panic("%s: Invalid serialization value %d for %d => %d",
10887                     __func__, serialize_row[pending_entry->seridx],
10888                     pending_entry->seridx, ooa_entry->seridx);
10889         }
10890
10891         return (CTL_ACTION_ERROR);
10892 }
10893
10894 /*
10895  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10896  * Assumptions:
10897  * - pending_io is generally either incoming, or on the blocked queue
10898  * - starting I/O is the I/O we want to start the check with.
10899  */
10900 static ctl_action
10901 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10902               union ctl_io *starting_io)
10903 {
10904         union ctl_io *ooa_io;
10905         ctl_action action;
10906
10907         mtx_assert(&lun->lun_lock, MA_OWNED);
10908
10909         /*
10910          * Run back along the OOA queue, starting with the current
10911          * blocked I/O and going through every I/O before it on the
10912          * queue.  If starting_io is NULL, we'll just end up returning
10913          * CTL_ACTION_PASS.
10914          */
10915         for (ooa_io = starting_io; ooa_io != NULL;
10916              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10917              ooa_links)){
10918
10919                 /*
10920                  * This routine just checks to see whether
10921                  * cur_blocked is blocked by ooa_io, which is ahead
10922                  * of it in the queue.  It doesn't queue/dequeue
10923                  * cur_blocked.
10924                  */
10925                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10926                 switch (action) {
10927                 case CTL_ACTION_BLOCK:
10928                 case CTL_ACTION_OVERLAP:
10929                 case CTL_ACTION_OVERLAP_TAG:
10930                 case CTL_ACTION_SKIP:
10931                 case CTL_ACTION_ERROR:
10932                         return (action);
10933                         break; /* NOTREACHED */
10934                 case CTL_ACTION_PASS:
10935                         break;
10936                 default:
10937                         panic("%s: Invalid action %d\n", __func__, action);
10938                 }
10939         }
10940
10941         return (CTL_ACTION_PASS);
10942 }
10943
10944 /*
10945  * Assumptions:
10946  * - An I/O has just completed, and has been removed from the per-LUN OOA
10947  *   queue, so some items on the blocked queue may now be unblocked.
10948  */
10949 static int
10950 ctl_check_blocked(struct ctl_lun *lun)
10951 {
10952         struct ctl_softc *softc = lun->ctl_softc;
10953         union ctl_io *cur_blocked, *next_blocked;
10954
10955         mtx_assert(&lun->lun_lock, MA_OWNED);
10956
10957         /*
10958          * Run forward from the head of the blocked queue, checking each
10959          * entry against the I/Os prior to it on the OOA queue to see if
10960          * there is still any blockage.
10961          *
10962          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10963          * with our removing a variable on it while it is traversing the
10964          * list.
10965          */
10966         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10967              cur_blocked != NULL; cur_blocked = next_blocked) {
10968                 union ctl_io *prev_ooa;
10969                 ctl_action action;
10970
10971                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10972                                                           blocked_links);
10973
10974                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10975                                                       ctl_ooaq, ooa_links);
10976
10977                 /*
10978                  * If cur_blocked happens to be the first item in the OOA
10979                  * queue now, prev_ooa will be NULL, and the action
10980                  * returned will just be CTL_ACTION_PASS.
10981                  */
10982                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10983
10984                 switch (action) {
10985                 case CTL_ACTION_BLOCK:
10986                         /* Nothing to do here, still blocked */
10987                         break;
10988                 case CTL_ACTION_OVERLAP:
10989                 case CTL_ACTION_OVERLAP_TAG:
10990                         /*
10991                          * This shouldn't happen!  In theory we've already
10992                          * checked this command for overlap...
10993                          */
10994                         break;
10995                 case CTL_ACTION_PASS:
10996                 case CTL_ACTION_SKIP: {
10997                         const struct ctl_cmd_entry *entry;
10998
10999                         /*
11000                          * The skip case shouldn't happen, this transaction
11001                          * should have never made it onto the blocked queue.
11002                          */
11003                         /*
11004                          * This I/O is no longer blocked, we can remove it
11005                          * from the blocked queue.  Since this is a TAILQ
11006                          * (doubly linked list), we can do O(1) removals
11007                          * from any place on the list.
11008                          */
11009                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11010                                      blocked_links);
11011                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11012
11013                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11014                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11015                                 /*
11016                                  * Need to send IO back to original side to
11017                                  * run
11018                                  */
11019                                 union ctl_ha_msg msg_info;
11020
11021                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11022                                 msg_info.hdr.original_sc =
11023                                         cur_blocked->io_hdr.original_sc;
11024                                 msg_info.hdr.serializing_sc = cur_blocked;
11025                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11026                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11027                                     sizeof(msg_info.hdr), M_NOWAIT);
11028                                 break;
11029                         }
11030                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11031
11032                         /*
11033                          * Check this I/O for LUN state changes that may
11034                          * have happened while this command was blocked.
11035                          * The LUN state may have been changed by a command
11036                          * ahead of us in the queue, so we need to re-check
11037                          * for any states that can be caused by SCSI
11038                          * commands.
11039                          */
11040                         if (ctl_scsiio_lun_check(lun, entry,
11041                                                  &cur_blocked->scsiio) == 0) {
11042                                 cur_blocked->io_hdr.flags |=
11043                                                       CTL_FLAG_IS_WAS_ON_RTR;
11044                                 ctl_enqueue_rtr(cur_blocked);
11045                         } else
11046                                 ctl_done(cur_blocked);
11047                         break;
11048                 }
11049                 default:
11050                         /*
11051                          * This probably shouldn't happen -- we shouldn't
11052                          * get CTL_ACTION_ERROR, or anything else.
11053                          */
11054                         break;
11055                 }
11056         }
11057
11058         return (CTL_RETVAL_COMPLETE);
11059 }
11060
11061 /*
11062  * This routine (with one exception) checks LUN flags that can be set by
11063  * commands ahead of us in the OOA queue.  These flags have to be checked
11064  * when a command initially comes in, and when we pull a command off the
11065  * blocked queue and are preparing to execute it.  The reason we have to
11066  * check these flags for commands on the blocked queue is that the LUN
11067  * state may have been changed by a command ahead of us while we're on the
11068  * blocked queue.
11069  *
11070  * Ordering is somewhat important with these checks, so please pay
11071  * careful attention to the placement of any new checks.
11072  */
11073 static int
11074 ctl_scsiio_lun_check(struct ctl_lun *lun,
11075     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11076 {
11077         struct ctl_softc *softc = lun->ctl_softc;
11078         int retval;
11079         uint32_t residx;
11080
11081         retval = 0;
11082
11083         mtx_assert(&lun->lun_lock, MA_OWNED);
11084
11085         /*
11086          * If this shelf is a secondary shelf controller, we may have to
11087          * reject some commands disallowed by HA mode and link state.
11088          */
11089         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11090                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11091                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11092                         ctl_set_lun_unavail(ctsio);
11093                         retval = 1;
11094                         goto bailout;
11095                 }
11096                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11097                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11098                         ctl_set_lun_transit(ctsio);
11099                         retval = 1;
11100                         goto bailout;
11101                 }
11102                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11103                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11104                         ctl_set_lun_standby(ctsio);
11105                         retval = 1;
11106                         goto bailout;
11107                 }
11108
11109                 /* The rest of checks are only done on executing side */
11110                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11111                         goto bailout;
11112         }
11113
11114         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11115                 if (lun->be_lun &&
11116                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11117                         ctl_set_hw_write_protected(ctsio);
11118                         retval = 1;
11119                         goto bailout;
11120                 }
11121                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11122                         ctl_set_sense(ctsio, /*current_error*/ 1,
11123                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11124                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11125                         retval = 1;
11126                         goto bailout;
11127                 }
11128         }
11129
11130         /*
11131          * Check for a reservation conflict.  If this command isn't allowed
11132          * even on reserved LUNs, and if this initiator isn't the one who
11133          * reserved us, reject the command with a reservation conflict.
11134          */
11135         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11136         if ((lun->flags & CTL_LUN_RESERVED)
11137          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11138                 if (lun->res_idx != residx) {
11139                         ctl_set_reservation_conflict(ctsio);
11140                         retval = 1;
11141                         goto bailout;
11142                 }
11143         }
11144
11145         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11146             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11147                 /* No reservation or command is allowed. */;
11148         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11149             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11150              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11151              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11152                 /* The command is allowed for Write Exclusive resv. */;
11153         } else {
11154                 /*
11155                  * if we aren't registered or it's a res holder type
11156                  * reservation and this isn't the res holder then set a
11157                  * conflict.
11158                  */
11159                 if (ctl_get_prkey(lun, residx) == 0 ||
11160                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11161                         ctl_set_reservation_conflict(ctsio);
11162                         retval = 1;
11163                         goto bailout;
11164                 }
11165         }
11166
11167         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11168                 if (lun->flags & CTL_LUN_EJECTED)
11169                         ctl_set_lun_ejected(ctsio);
11170                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11171                         if (lun->flags & CTL_LUN_REMOVABLE)
11172                                 ctl_set_lun_no_media(ctsio);
11173                         else
11174                                 ctl_set_lun_int_reqd(ctsio);
11175                 } else if (lun->flags & CTL_LUN_STOPPED)
11176                         ctl_set_lun_stopped(ctsio);
11177                 else
11178                         goto bailout;
11179                 retval = 1;
11180                 goto bailout;
11181         }
11182
11183 bailout:
11184         return (retval);
11185 }
11186
11187 static void
11188 ctl_failover_io(union ctl_io *io, int have_lock)
11189 {
11190         ctl_set_busy(&io->scsiio);
11191         ctl_done(io);
11192 }
11193
11194 static void
11195 ctl_failover_lun(union ctl_io *rio)
11196 {
11197         struct ctl_softc *softc = CTL_SOFTC(rio);
11198         struct ctl_lun *lun;
11199         struct ctl_io_hdr *io, *next_io;
11200         uint32_t targ_lun;
11201
11202         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11203         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11204
11205         /* Find and lock the LUN. */
11206         mtx_lock(&softc->ctl_lock);
11207         if (targ_lun > CTL_MAX_LUNS ||
11208             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11209                 mtx_unlock(&softc->ctl_lock);
11210                 return;
11211         }
11212         mtx_lock(&lun->lun_lock);
11213         mtx_unlock(&softc->ctl_lock);
11214         if (lun->flags & CTL_LUN_DISABLED) {
11215                 mtx_unlock(&lun->lun_lock);
11216                 return;
11217         }
11218
11219         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11220                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11221                         /* We are master */
11222                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11223                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11224                                         io->flags |= CTL_FLAG_ABORT;
11225                                         io->flags |= CTL_FLAG_FAILOVER;
11226                                 } else { /* This can be only due to DATAMOVE */
11227                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11228                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11229                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11230                                         io->port_status = 31340;
11231                                         ctl_enqueue_isc((union ctl_io *)io);
11232                                 }
11233                         }
11234                         /* We are slave */
11235                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11236                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11237                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11238                                         io->flags |= CTL_FLAG_FAILOVER;
11239                                 } else {
11240                                         ctl_set_busy(&((union ctl_io *)io)->
11241                                             scsiio);
11242                                         ctl_done((union ctl_io *)io);
11243                                 }
11244                         }
11245                 }
11246         } else { /* SERIALIZE modes */
11247                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11248                     next_io) {
11249                         /* We are master */
11250                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11251                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11252                                     blocked_links);
11253                                 io->flags &= ~CTL_FLAG_BLOCKED;
11254                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11255                                 ctl_free_io((union ctl_io *)io);
11256                         }
11257                 }
11258                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11259                         /* We are master */
11260                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11261                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11262                                 ctl_free_io((union ctl_io *)io);
11263                         }
11264                         /* We are slave */
11265                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11266                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11267                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11268                                         ctl_set_busy(&((union ctl_io *)io)->
11269                                             scsiio);
11270                                         ctl_done((union ctl_io *)io);
11271                                 }
11272                         }
11273                 }
11274                 ctl_check_blocked(lun);
11275         }
11276         mtx_unlock(&lun->lun_lock);
11277 }
11278
11279 static int
11280 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11281 {
11282         struct ctl_lun *lun;
11283         const struct ctl_cmd_entry *entry;
11284         uint32_t initidx, targ_lun;
11285         int retval = 0;
11286
11287         lun = NULL;
11288         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11289         if (targ_lun < CTL_MAX_LUNS)
11290                 lun = softc->ctl_luns[targ_lun];
11291         if (lun) {
11292                 /*
11293                  * If the LUN is invalid, pretend that it doesn't exist.
11294                  * It will go away as soon as all pending I/O has been
11295                  * completed.
11296                  */
11297                 mtx_lock(&lun->lun_lock);
11298                 if (lun->flags & CTL_LUN_DISABLED) {
11299                         mtx_unlock(&lun->lun_lock);
11300                         lun = NULL;
11301                 }
11302         }
11303         CTL_LUN(ctsio) = lun;
11304         if (lun) {
11305                 CTL_BACKEND_LUN(ctsio) = lun->be_lun;
11306
11307                 /*
11308                  * Every I/O goes into the OOA queue for a particular LUN,
11309                  * and stays there until completion.
11310                  */
11311 #ifdef CTL_TIME_IO
11312                 if (TAILQ_EMPTY(&lun->ooa_queue))
11313                         lun->idle_time += getsbinuptime() - lun->last_busy;
11314 #endif
11315                 TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11316         }
11317
11318         /* Get command entry and return error if it is unsuppotyed. */
11319         entry = ctl_validate_command(ctsio);
11320         if (entry == NULL) {
11321                 if (lun)
11322                         mtx_unlock(&lun->lun_lock);
11323                 return (retval);
11324         }
11325
11326         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11327         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11328
11329         /*
11330          * Check to see whether we can send this command to LUNs that don't
11331          * exist.  This should pretty much only be the case for inquiry
11332          * and request sense.  Further checks, below, really require having
11333          * a LUN, so we can't really check the command anymore.  Just put
11334          * it on the rtr queue.
11335          */
11336         if (lun == NULL) {
11337                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11338                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11339                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11340                         return (retval);
11341                 }
11342
11343                 ctl_set_unsupported_lun(ctsio);
11344                 ctl_done((union ctl_io *)ctsio);
11345                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11346                 return (retval);
11347         } else {
11348                 /*
11349                  * Make sure we support this particular command on this LUN.
11350                  * e.g., we don't support writes to the control LUN.
11351                  */
11352                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11353                         mtx_unlock(&lun->lun_lock);
11354                         ctl_set_invalid_opcode(ctsio);
11355                         ctl_done((union ctl_io *)ctsio);
11356                         return (retval);
11357                 }
11358         }
11359
11360         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11361
11362         /*
11363          * If we've got a request sense, it'll clear the contingent
11364          * allegiance condition.  Otherwise, if we have a CA condition for
11365          * this initiator, clear it, because it sent down a command other
11366          * than request sense.
11367          */
11368         if (ctsio->cdb[0] != REQUEST_SENSE) {
11369                 struct scsi_sense_data *ps;
11370
11371                 ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
11372                 if (ps != NULL)
11373                         ps[initidx % CTL_MAX_INIT_PER_PORT].error_code = 0;
11374         }
11375
11376         /*
11377          * If the command has this flag set, it handles its own unit
11378          * attention reporting, we shouldn't do anything.  Otherwise we
11379          * check for any pending unit attentions, and send them back to the
11380          * initiator.  We only do this when a command initially comes in,
11381          * not when we pull it off the blocked queue.
11382          *
11383          * According to SAM-3, section 5.3.2, the order that things get
11384          * presented back to the host is basically unit attentions caused
11385          * by some sort of reset event, busy status, reservation conflicts
11386          * or task set full, and finally any other status.
11387          *
11388          * One issue here is that some of the unit attentions we report
11389          * don't fall into the "reset" category (e.g. "reported luns data
11390          * has changed").  So reporting it here, before the reservation
11391          * check, may be technically wrong.  I guess the only thing to do
11392          * would be to check for and report the reset events here, and then
11393          * check for the other unit attention types after we check for a
11394          * reservation conflict.
11395          *
11396          * XXX KDM need to fix this
11397          */
11398         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11399                 ctl_ua_type ua_type;
11400                 u_int sense_len = 0;
11401
11402                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11403                     &sense_len, SSD_TYPE_NONE);
11404                 if (ua_type != CTL_UA_NONE) {
11405                         mtx_unlock(&lun->lun_lock);
11406                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11407                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11408                         ctsio->sense_len = sense_len;
11409                         ctl_done((union ctl_io *)ctsio);
11410                         return (retval);
11411                 }
11412         }
11413
11414
11415         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11416                 mtx_unlock(&lun->lun_lock);
11417                 ctl_done((union ctl_io *)ctsio);
11418                 return (retval);
11419         }
11420
11421         /*
11422          * XXX CHD this is where we want to send IO to other side if
11423          * this LUN is secondary on this SC. We will need to make a copy
11424          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11425          * the copy we send as FROM_OTHER.
11426          * We also need to stuff the address of the original IO so we can
11427          * find it easily. Something similar will need be done on the other
11428          * side so when we are done we can find the copy.
11429          */
11430         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11431             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11432             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11433                 union ctl_ha_msg msg_info;
11434                 int isc_retval;
11435
11436                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11437                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11438                 mtx_unlock(&lun->lun_lock);
11439
11440                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11441                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11442                 msg_info.hdr.serializing_sc = NULL;
11443                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11444                 msg_info.scsi.tag_num = ctsio->tag_num;
11445                 msg_info.scsi.tag_type = ctsio->tag_type;
11446                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11447                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11448
11449                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11450                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11451                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11452                         ctl_set_busy(ctsio);
11453                         ctl_done((union ctl_io *)ctsio);
11454                         return (retval);
11455                 }
11456                 return (retval);
11457         }
11458
11459         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11460                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11461                               ctl_ooaq, ooa_links))) {
11462         case CTL_ACTION_BLOCK:
11463                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11464                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11465                                   blocked_links);
11466                 mtx_unlock(&lun->lun_lock);
11467                 return (retval);
11468         case CTL_ACTION_PASS:
11469         case CTL_ACTION_SKIP:
11470                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11471                 mtx_unlock(&lun->lun_lock);
11472                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11473                 break;
11474         case CTL_ACTION_OVERLAP:
11475                 mtx_unlock(&lun->lun_lock);
11476                 ctl_set_overlapped_cmd(ctsio);
11477                 ctl_done((union ctl_io *)ctsio);
11478                 break;
11479         case CTL_ACTION_OVERLAP_TAG:
11480                 mtx_unlock(&lun->lun_lock);
11481                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11482                 ctl_done((union ctl_io *)ctsio);
11483                 break;
11484         case CTL_ACTION_ERROR:
11485         default:
11486                 mtx_unlock(&lun->lun_lock);
11487                 ctl_set_internal_failure(ctsio,
11488                                          /*sks_valid*/ 0,
11489                                          /*retry_count*/ 0);
11490                 ctl_done((union ctl_io *)ctsio);
11491                 break;
11492         }
11493         return (retval);
11494 }
11495
11496 const struct ctl_cmd_entry *
11497 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11498 {
11499         const struct ctl_cmd_entry *entry;
11500         int service_action;
11501
11502         entry = &ctl_cmd_table[ctsio->cdb[0]];
11503         if (sa)
11504                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11505         if (entry->flags & CTL_CMD_FLAG_SA5) {
11506                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11507                 entry = &((const struct ctl_cmd_entry *)
11508                     entry->execute)[service_action];
11509         }
11510         return (entry);
11511 }
11512
11513 const struct ctl_cmd_entry *
11514 ctl_validate_command(struct ctl_scsiio *ctsio)
11515 {
11516         const struct ctl_cmd_entry *entry;
11517         int i, sa;
11518         uint8_t diff;
11519
11520         entry = ctl_get_cmd_entry(ctsio, &sa);
11521         if (entry->execute == NULL) {
11522                 if (sa)
11523                         ctl_set_invalid_field(ctsio,
11524                                               /*sks_valid*/ 1,
11525                                               /*command*/ 1,
11526                                               /*field*/ 1,
11527                                               /*bit_valid*/ 1,
11528                                               /*bit*/ 4);
11529                 else
11530                         ctl_set_invalid_opcode(ctsio);
11531                 ctl_done((union ctl_io *)ctsio);
11532                 return (NULL);
11533         }
11534         KASSERT(entry->length > 0,
11535             ("Not defined length for command 0x%02x/0x%02x",
11536              ctsio->cdb[0], ctsio->cdb[1]));
11537         for (i = 1; i < entry->length; i++) {
11538                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11539                 if (diff == 0)
11540                         continue;
11541                 ctl_set_invalid_field(ctsio,
11542                                       /*sks_valid*/ 1,
11543                                       /*command*/ 1,
11544                                       /*field*/ i,
11545                                       /*bit_valid*/ 1,
11546                                       /*bit*/ fls(diff) - 1);
11547                 ctl_done((union ctl_io *)ctsio);
11548                 return (NULL);
11549         }
11550         return (entry);
11551 }
11552
11553 static int
11554 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11555 {
11556
11557         switch (lun_type) {
11558         case T_DIRECT:
11559                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11560                         return (0);
11561                 break;
11562         case T_PROCESSOR:
11563                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11564                         return (0);
11565                 break;
11566         case T_CDROM:
11567                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11568                         return (0);
11569                 break;
11570         default:
11571                 return (0);
11572         }
11573         return (1);
11574 }
11575
11576 static int
11577 ctl_scsiio(struct ctl_scsiio *ctsio)
11578 {
11579         int retval;
11580         const struct ctl_cmd_entry *entry;
11581
11582         retval = CTL_RETVAL_COMPLETE;
11583
11584         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11585
11586         entry = ctl_get_cmd_entry(ctsio, NULL);
11587
11588         /*
11589          * If this I/O has been aborted, just send it straight to
11590          * ctl_done() without executing it.
11591          */
11592         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11593                 ctl_done((union ctl_io *)ctsio);
11594                 goto bailout;
11595         }
11596
11597         /*
11598          * All the checks should have been handled by ctl_scsiio_precheck().
11599          * We should be clear now to just execute the I/O.
11600          */
11601         retval = entry->execute(ctsio);
11602
11603 bailout:
11604         return (retval);
11605 }
11606
11607 static int
11608 ctl_target_reset(union ctl_io *io)
11609 {
11610         struct ctl_softc *softc = CTL_SOFTC(io);
11611         struct ctl_port *port = CTL_PORT(io);
11612         struct ctl_lun *lun;
11613         uint32_t initidx;
11614         ctl_ua_type ua_type;
11615
11616         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11617                 union ctl_ha_msg msg_info;
11618
11619                 msg_info.hdr.nexus = io->io_hdr.nexus;
11620                 msg_info.task.task_action = io->taskio.task_action;
11621                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11622                 msg_info.hdr.original_sc = NULL;
11623                 msg_info.hdr.serializing_sc = NULL;
11624                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11625                     sizeof(msg_info.task), M_WAITOK);
11626         }
11627
11628         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11629         if (io->taskio.task_action == CTL_TASK_TARGET_RESET)
11630                 ua_type = CTL_UA_TARG_RESET;
11631         else
11632                 ua_type = CTL_UA_BUS_RESET;
11633         mtx_lock(&softc->ctl_lock);
11634         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11635                 if (port != NULL &&
11636                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
11637                         continue;
11638                 ctl_do_lun_reset(lun, initidx, ua_type);
11639         }
11640         mtx_unlock(&softc->ctl_lock);
11641         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11642         return (0);
11643 }
11644
11645 /*
11646  * The LUN should always be set.  The I/O is optional, and is used to
11647  * distinguish between I/Os sent by this initiator, and by other
11648  * initiators.  We set unit attention for initiators other than this one.
11649  * SAM-3 is vague on this point.  It does say that a unit attention should
11650  * be established for other initiators when a LUN is reset (see section
11651  * 5.7.3), but it doesn't specifically say that the unit attention should
11652  * be established for this particular initiator when a LUN is reset.  Here
11653  * is the relevant text, from SAM-3 rev 8:
11654  *
11655  * 5.7.2 When a SCSI initiator port aborts its own tasks
11656  *
11657  * When a SCSI initiator port causes its own task(s) to be aborted, no
11658  * notification that the task(s) have been aborted shall be returned to
11659  * the SCSI initiator port other than the completion response for the
11660  * command or task management function action that caused the task(s) to
11661  * be aborted and notification(s) associated with related effects of the
11662  * action (e.g., a reset unit attention condition).
11663  *
11664  * XXX KDM for now, we're setting unit attention for all initiators.
11665  */
11666 static void
11667 ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua_type)
11668 {
11669         union ctl_io *xio;
11670         int i;
11671
11672         mtx_lock(&lun->lun_lock);
11673         /* Abort tasks. */
11674         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11675              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11676                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11677         }
11678         /* Clear CA. */
11679         for (i = 0; i < CTL_MAX_PORTS; i++) {
11680                 free(lun->pending_sense[i], M_CTL);
11681                 lun->pending_sense[i] = NULL;
11682         }
11683         /* Clear reservation. */
11684         lun->flags &= ~CTL_LUN_RESERVED;
11685         /* Clear prevent media removal. */
11686         if (lun->prevent) {
11687                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
11688                         ctl_clear_mask(lun->prevent, i);
11689                 lun->prevent_count = 0;
11690         }
11691         /* Clear TPC status */
11692         ctl_tpc_lun_clear(lun, -1);
11693         /* Establish UA. */
11694 #if 0
11695         ctl_est_ua_all(lun, initidx, ua_type);
11696 #else
11697         ctl_est_ua_all(lun, -1, ua_type);
11698 #endif
11699         mtx_unlock(&lun->lun_lock);
11700 }
11701
11702 static int
11703 ctl_lun_reset(union ctl_io *io)
11704 {
11705         struct ctl_softc *softc = CTL_SOFTC(io);
11706         struct ctl_lun *lun;
11707         uint32_t targ_lun, initidx;
11708
11709         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11710         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11711         mtx_lock(&softc->ctl_lock);
11712         if (targ_lun >= CTL_MAX_LUNS ||
11713             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11714                 mtx_unlock(&softc->ctl_lock);
11715                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11716                 return (1);
11717         }
11718         ctl_do_lun_reset(lun, initidx, CTL_UA_LUN_RESET);
11719         mtx_unlock(&softc->ctl_lock);
11720         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11721
11722         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11723                 union ctl_ha_msg msg_info;
11724
11725                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11726                 msg_info.hdr.nexus = io->io_hdr.nexus;
11727                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11728                 msg_info.hdr.original_sc = NULL;
11729                 msg_info.hdr.serializing_sc = NULL;
11730                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11731                     sizeof(msg_info.task), M_WAITOK);
11732         }
11733         return (0);
11734 }
11735
11736 static void
11737 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11738     int other_sc)
11739 {
11740         union ctl_io *xio;
11741
11742         mtx_assert(&lun->lun_lock, MA_OWNED);
11743
11744         /*
11745          * Run through the OOA queue and attempt to find the given I/O.
11746          * The target port, initiator ID, tag type and tag number have to
11747          * match the values that we got from the initiator.  If we have an
11748          * untagged command to abort, simply abort the first untagged command
11749          * we come to.  We only allow one untagged command at a time of course.
11750          */
11751         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11752              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11753
11754                 if ((targ_port == UINT32_MAX ||
11755                      targ_port == xio->io_hdr.nexus.targ_port) &&
11756                     (init_id == UINT32_MAX ||
11757                      init_id == xio->io_hdr.nexus.initid)) {
11758                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11759                             init_id != xio->io_hdr.nexus.initid)
11760                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11761                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11762                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11763                                 union ctl_ha_msg msg_info;
11764
11765                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11766                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11767                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11768                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11769                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11770                                 msg_info.hdr.original_sc = NULL;
11771                                 msg_info.hdr.serializing_sc = NULL;
11772                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11773                                     sizeof(msg_info.task), M_NOWAIT);
11774                         }
11775                 }
11776         }
11777 }
11778
11779 static int
11780 ctl_abort_task_set(union ctl_io *io)
11781 {
11782         struct ctl_softc *softc = CTL_SOFTC(io);
11783         struct ctl_lun *lun;
11784         uint32_t targ_lun;
11785
11786         /*
11787          * Look up the LUN.
11788          */
11789         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11790         mtx_lock(&softc->ctl_lock);
11791         if (targ_lun >= CTL_MAX_LUNS ||
11792             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11793                 mtx_unlock(&softc->ctl_lock);
11794                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11795                 return (1);
11796         }
11797
11798         mtx_lock(&lun->lun_lock);
11799         mtx_unlock(&softc->ctl_lock);
11800         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11801                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11802                     io->io_hdr.nexus.initid,
11803                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11804         } else { /* CTL_TASK_CLEAR_TASK_SET */
11805                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11806                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11807         }
11808         mtx_unlock(&lun->lun_lock);
11809         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11810         return (0);
11811 }
11812
11813 static void
11814 ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
11815     ctl_ua_type ua_type)
11816 {
11817         struct ctl_lun *lun;
11818         struct scsi_sense_data *ps;
11819         uint32_t p, i;
11820
11821         p = initidx / CTL_MAX_INIT_PER_PORT;
11822         i = initidx % CTL_MAX_INIT_PER_PORT;
11823         mtx_lock(&softc->ctl_lock);
11824         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11825                 mtx_lock(&lun->lun_lock);
11826                 /* Abort tasks. */
11827                 ctl_abort_tasks_lun(lun, p, i, 1);
11828                 /* Clear CA. */
11829                 ps = lun->pending_sense[p];
11830                 if (ps != NULL)
11831                         ps[i].error_code = 0;
11832                 /* Clear reservation. */
11833                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11834                         lun->flags &= ~CTL_LUN_RESERVED;
11835                 /* Clear prevent media removal. */
11836                 if (lun->prevent && ctl_is_set(lun->prevent, initidx)) {
11837                         ctl_clear_mask(lun->prevent, initidx);
11838                         lun->prevent_count--;
11839                 }
11840                 /* Clear TPC status */
11841                 ctl_tpc_lun_clear(lun, initidx);
11842                 /* Establish UA. */
11843                 ctl_est_ua(lun, initidx, ua_type);
11844                 mtx_unlock(&lun->lun_lock);
11845         }
11846         mtx_unlock(&softc->ctl_lock);
11847 }
11848
11849 static int
11850 ctl_i_t_nexus_reset(union ctl_io *io)
11851 {
11852         struct ctl_softc *softc = CTL_SOFTC(io);
11853         uint32_t initidx;
11854
11855         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11856                 union ctl_ha_msg msg_info;
11857
11858                 msg_info.hdr.nexus = io->io_hdr.nexus;
11859                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
11860                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11861                 msg_info.hdr.original_sc = NULL;
11862                 msg_info.hdr.serializing_sc = NULL;
11863                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11864                     sizeof(msg_info.task), M_WAITOK);
11865         }
11866
11867         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11868         ctl_i_t_nexus_loss(softc, initidx, CTL_UA_I_T_NEXUS_LOSS);
11869         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11870         return (0);
11871 }
11872
11873 static int
11874 ctl_abort_task(union ctl_io *io)
11875 {
11876         struct ctl_softc *softc = CTL_SOFTC(io);
11877         union ctl_io *xio;
11878         struct ctl_lun *lun;
11879 #if 0
11880         struct sbuf sb;
11881         char printbuf[128];
11882 #endif
11883         int found;
11884         uint32_t targ_lun;
11885
11886         found = 0;
11887
11888         /*
11889          * Look up the LUN.
11890          */
11891         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11892         mtx_lock(&softc->ctl_lock);
11893         if (targ_lun >= CTL_MAX_LUNS ||
11894             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11895                 mtx_unlock(&softc->ctl_lock);
11896                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11897                 return (1);
11898         }
11899
11900 #if 0
11901         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11902                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11903 #endif
11904
11905         mtx_lock(&lun->lun_lock);
11906         mtx_unlock(&softc->ctl_lock);
11907         /*
11908          * Run through the OOA queue and attempt to find the given I/O.
11909          * The target port, initiator ID, tag type and tag number have to
11910          * match the values that we got from the initiator.  If we have an
11911          * untagged command to abort, simply abort the first untagged command
11912          * we come to.  We only allow one untagged command at a time of course.
11913          */
11914         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11915              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11916 #if 0
11917                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11918
11919                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11920                             lun->lun, xio->scsiio.tag_num,
11921                             xio->scsiio.tag_type,
11922                             (xio->io_hdr.blocked_links.tqe_prev
11923                             == NULL) ? "" : " BLOCKED",
11924                             (xio->io_hdr.flags &
11925                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11926                             (xio->io_hdr.flags &
11927                             CTL_FLAG_ABORT) ? " ABORT" : "",
11928                             (xio->io_hdr.flags &
11929                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11930                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11931                 sbuf_finish(&sb);
11932                 printf("%s\n", sbuf_data(&sb));
11933 #endif
11934
11935                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11936                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11937                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11938                         continue;
11939
11940                 /*
11941                  * If the abort says that the task is untagged, the
11942                  * task in the queue must be untagged.  Otherwise,
11943                  * we just check to see whether the tag numbers
11944                  * match.  This is because the QLogic firmware
11945                  * doesn't pass back the tag type in an abort
11946                  * request.
11947                  */
11948 #if 0
11949                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11950                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11951                  || (xio->scsiio.tag_num == io->taskio.tag_num))
11952 #endif
11953                 /*
11954                  * XXX KDM we've got problems with FC, because it
11955                  * doesn't send down a tag type with aborts.  So we
11956                  * can only really go by the tag number...
11957                  * This may cause problems with parallel SCSI.
11958                  * Need to figure that out!!
11959                  */
11960                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
11961                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11962                         found = 1;
11963                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
11964                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11965                                 union ctl_ha_msg msg_info;
11966
11967                                 msg_info.hdr.nexus = io->io_hdr.nexus;
11968                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11969                                 msg_info.task.tag_num = io->taskio.tag_num;
11970                                 msg_info.task.tag_type = io->taskio.tag_type;
11971                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11972                                 msg_info.hdr.original_sc = NULL;
11973                                 msg_info.hdr.serializing_sc = NULL;
11974 #if 0
11975                                 printf("Sent Abort to other side\n");
11976 #endif
11977                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11978                                     sizeof(msg_info.task), M_NOWAIT);
11979                         }
11980 #if 0
11981                         printf("ctl_abort_task: found I/O to abort\n");
11982 #endif
11983                 }
11984         }
11985         mtx_unlock(&lun->lun_lock);
11986
11987         if (found == 0) {
11988                 /*
11989                  * This isn't really an error.  It's entirely possible for
11990                  * the abort and command completion to cross on the wire.
11991                  * This is more of an informative/diagnostic error.
11992                  */
11993 #if 0
11994                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11995                        "%u:%u:%u tag %d type %d\n",
11996                        io->io_hdr.nexus.initid,
11997                        io->io_hdr.nexus.targ_port,
11998                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11999                        io->taskio.tag_type);
12000 #endif
12001         }
12002         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12003         return (0);
12004 }
12005
12006 static int
12007 ctl_query_task(union ctl_io *io, int task_set)
12008 {
12009         struct ctl_softc *softc = CTL_SOFTC(io);
12010         union ctl_io *xio;
12011         struct ctl_lun *lun;
12012         int found = 0;
12013         uint32_t targ_lun;
12014
12015         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12016         mtx_lock(&softc->ctl_lock);
12017         if (targ_lun >= CTL_MAX_LUNS ||
12018             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12019                 mtx_unlock(&softc->ctl_lock);
12020                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12021                 return (1);
12022         }
12023         mtx_lock(&lun->lun_lock);
12024         mtx_unlock(&softc->ctl_lock);
12025         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12026              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12027
12028                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12029                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12030                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12031                         continue;
12032
12033                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12034                         found = 1;
12035                         break;
12036                 }
12037         }
12038         mtx_unlock(&lun->lun_lock);
12039         if (found)
12040                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12041         else
12042                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12043         return (0);
12044 }
12045
12046 static int
12047 ctl_query_async_event(union ctl_io *io)
12048 {
12049         struct ctl_softc *softc = CTL_SOFTC(io);
12050         struct ctl_lun *lun;
12051         ctl_ua_type ua;
12052         uint32_t targ_lun, initidx;
12053
12054         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12055         mtx_lock(&softc->ctl_lock);
12056         if (targ_lun >= CTL_MAX_LUNS ||
12057             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12058                 mtx_unlock(&softc->ctl_lock);
12059                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12060                 return (1);
12061         }
12062         mtx_lock(&lun->lun_lock);
12063         mtx_unlock(&softc->ctl_lock);
12064         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12065         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12066         mtx_unlock(&lun->lun_lock);
12067         if (ua != CTL_UA_NONE)
12068                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12069         else
12070                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12071         return (0);
12072 }
12073
12074 static void
12075 ctl_run_task(union ctl_io *io)
12076 {
12077         int retval = 1;
12078
12079         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12080         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12081             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12082         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12083         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12084         switch (io->taskio.task_action) {
12085         case CTL_TASK_ABORT_TASK:
12086                 retval = ctl_abort_task(io);
12087                 break;
12088         case CTL_TASK_ABORT_TASK_SET:
12089         case CTL_TASK_CLEAR_TASK_SET:
12090                 retval = ctl_abort_task_set(io);
12091                 break;
12092         case CTL_TASK_CLEAR_ACA:
12093                 break;
12094         case CTL_TASK_I_T_NEXUS_RESET:
12095                 retval = ctl_i_t_nexus_reset(io);
12096                 break;
12097         case CTL_TASK_LUN_RESET:
12098                 retval = ctl_lun_reset(io);
12099                 break;
12100         case CTL_TASK_TARGET_RESET:
12101         case CTL_TASK_BUS_RESET:
12102                 retval = ctl_target_reset(io);
12103                 break;
12104         case CTL_TASK_PORT_LOGIN:
12105                 break;
12106         case CTL_TASK_PORT_LOGOUT:
12107                 break;
12108         case CTL_TASK_QUERY_TASK:
12109                 retval = ctl_query_task(io, 0);
12110                 break;
12111         case CTL_TASK_QUERY_TASK_SET:
12112                 retval = ctl_query_task(io, 1);
12113                 break;
12114         case CTL_TASK_QUERY_ASYNC_EVENT:
12115                 retval = ctl_query_async_event(io);
12116                 break;
12117         default:
12118                 printf("%s: got unknown task management event %d\n",
12119                        __func__, io->taskio.task_action);
12120                 break;
12121         }
12122         if (retval == 0)
12123                 io->io_hdr.status = CTL_SUCCESS;
12124         else
12125                 io->io_hdr.status = CTL_ERROR;
12126         ctl_done(io);
12127 }
12128
12129 /*
12130  * For HA operation.  Handle commands that come in from the other
12131  * controller.
12132  */
12133 static void
12134 ctl_handle_isc(union ctl_io *io)
12135 {
12136         struct ctl_softc *softc = CTL_SOFTC(io);
12137         struct ctl_lun *lun;
12138         const struct ctl_cmd_entry *entry;
12139         uint32_t targ_lun;
12140
12141         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12142         switch (io->io_hdr.msg_type) {
12143         case CTL_MSG_SERIALIZE:
12144                 ctl_serialize_other_sc_cmd(&io->scsiio);
12145                 break;
12146         case CTL_MSG_R2R:               /* Only used in SER_ONLY mode. */
12147                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12148                 if (targ_lun >= CTL_MAX_LUNS ||
12149                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12150                         ctl_done(io);
12151                         break;
12152                 }
12153                 mtx_lock(&lun->lun_lock);
12154                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
12155                         mtx_unlock(&lun->lun_lock);
12156                         ctl_done(io);
12157                         break;
12158                 }
12159                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12160                 mtx_unlock(&lun->lun_lock);
12161                 ctl_enqueue_rtr(io);
12162                 break;
12163         case CTL_MSG_FINISH_IO:
12164                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12165                         ctl_done(io);
12166                         break;
12167                 }
12168                 if (targ_lun >= CTL_MAX_LUNS ||
12169                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12170                         ctl_free_io(io);
12171                         break;
12172                 }
12173                 mtx_lock(&lun->lun_lock);
12174                 TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12175                 ctl_check_blocked(lun);
12176                 mtx_unlock(&lun->lun_lock);
12177                 ctl_free_io(io);
12178                 break;
12179         case CTL_MSG_PERS_ACTION:
12180                 ctl_hndl_per_res_out_on_other_sc(io);
12181                 ctl_free_io(io);
12182                 break;
12183         case CTL_MSG_BAD_JUJU:
12184                 ctl_done(io);
12185                 break;
12186         case CTL_MSG_DATAMOVE:          /* Only used in XFER mode */
12187                 ctl_datamove_remote(io);
12188                 break;
12189         case CTL_MSG_DATAMOVE_DONE:     /* Only used in XFER mode */
12190                 io->scsiio.be_move_done(io);
12191                 break;
12192         case CTL_MSG_FAILOVER:
12193                 ctl_failover_lun(io);
12194                 ctl_free_io(io);
12195                 break;
12196         default:
12197                 printf("%s: Invalid message type %d\n",
12198                        __func__, io->io_hdr.msg_type);
12199                 ctl_free_io(io);
12200                 break;
12201         }
12202
12203 }
12204
12205
12206 /*
12207  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12208  * there is no match.
12209  */
12210 static ctl_lun_error_pattern
12211 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12212 {
12213         const struct ctl_cmd_entry *entry;
12214         ctl_lun_error_pattern filtered_pattern, pattern;
12215
12216         pattern = desc->error_pattern;
12217
12218         /*
12219          * XXX KDM we need more data passed into this function to match a
12220          * custom pattern, and we actually need to implement custom pattern
12221          * matching.
12222          */
12223         if (pattern & CTL_LUN_PAT_CMD)
12224                 return (CTL_LUN_PAT_CMD);
12225
12226         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12227                 return (CTL_LUN_PAT_ANY);
12228
12229         entry = ctl_get_cmd_entry(ctsio, NULL);
12230
12231         filtered_pattern = entry->pattern & pattern;
12232
12233         /*
12234          * If the user requested specific flags in the pattern (e.g.
12235          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12236          * flags.
12237          *
12238          * If the user did not specify any flags, it doesn't matter whether
12239          * or not the command supports the flags.
12240          */
12241         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12242              (pattern & ~CTL_LUN_PAT_MASK))
12243                 return (CTL_LUN_PAT_NONE);
12244
12245         /*
12246          * If the user asked for a range check, see if the requested LBA
12247          * range overlaps with this command's LBA range.
12248          */
12249         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12250                 uint64_t lba1;
12251                 uint64_t len1;
12252                 ctl_action action;
12253                 int retval;
12254
12255                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12256                 if (retval != 0)
12257                         return (CTL_LUN_PAT_NONE);
12258
12259                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12260                                               desc->lba_range.len, FALSE);
12261                 /*
12262                  * A "pass" means that the LBA ranges don't overlap, so
12263                  * this doesn't match the user's range criteria.
12264                  */
12265                 if (action == CTL_ACTION_PASS)
12266                         return (CTL_LUN_PAT_NONE);
12267         }
12268
12269         return (filtered_pattern);
12270 }
12271
12272 static void
12273 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12274 {
12275         struct ctl_error_desc *desc, *desc2;
12276
12277         mtx_assert(&lun->lun_lock, MA_OWNED);
12278
12279         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12280                 ctl_lun_error_pattern pattern;
12281                 /*
12282                  * Check to see whether this particular command matches
12283                  * the pattern in the descriptor.
12284                  */
12285                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12286                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12287                         continue;
12288
12289                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12290                 case CTL_LUN_INJ_ABORTED:
12291                         ctl_set_aborted(&io->scsiio);
12292                         break;
12293                 case CTL_LUN_INJ_MEDIUM_ERR:
12294                         ctl_set_medium_error(&io->scsiio,
12295                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12296                              CTL_FLAG_DATA_OUT);
12297                         break;
12298                 case CTL_LUN_INJ_UA:
12299                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12300                          * OCCURRED */
12301                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12302                         break;
12303                 case CTL_LUN_INJ_CUSTOM:
12304                         /*
12305                          * We're assuming the user knows what he is doing.
12306                          * Just copy the sense information without doing
12307                          * checks.
12308                          */
12309                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12310                               MIN(sizeof(desc->custom_sense),
12311                                   sizeof(io->scsiio.sense_data)));
12312                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12313                         io->scsiio.sense_len = SSD_FULL_SIZE;
12314                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12315                         break;
12316                 case CTL_LUN_INJ_NONE:
12317                 default:
12318                         /*
12319                          * If this is an error injection type we don't know
12320                          * about, clear the continuous flag (if it is set)
12321                          * so it will get deleted below.
12322                          */
12323                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12324                         break;
12325                 }
12326                 /*
12327                  * By default, each error injection action is a one-shot
12328                  */
12329                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12330                         continue;
12331
12332                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12333
12334                 free(desc, M_CTL);
12335         }
12336 }
12337
12338 #ifdef CTL_IO_DELAY
12339 static void
12340 ctl_datamove_timer_wakeup(void *arg)
12341 {
12342         union ctl_io *io;
12343
12344         io = (union ctl_io *)arg;
12345
12346         ctl_datamove(io);
12347 }
12348 #endif /* CTL_IO_DELAY */
12349
12350 void
12351 ctl_datamove(union ctl_io *io)
12352 {
12353         void (*fe_datamove)(union ctl_io *io);
12354
12355         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12356
12357         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12358
12359         /* No data transferred yet.  Frontend must update this when done. */
12360         io->scsiio.kern_data_resid = io->scsiio.kern_data_len;
12361
12362 #ifdef CTL_TIME_IO
12363         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12364                 char str[256];
12365                 char path_str[64];
12366                 struct sbuf sb;
12367
12368                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12369                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12370
12371                 sbuf_cat(&sb, path_str);
12372                 switch (io->io_hdr.io_type) {
12373                 case CTL_IO_SCSI:
12374                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12375                         sbuf_printf(&sb, "\n");
12376                         sbuf_cat(&sb, path_str);
12377                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12378                                     io->scsiio.tag_num, io->scsiio.tag_type);
12379                         break;
12380                 case CTL_IO_TASK:
12381                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12382                                     "Tag Type: %d\n", io->taskio.task_action,
12383                                     io->taskio.tag_num, io->taskio.tag_type);
12384                         break;
12385                 default:
12386                         panic("%s: Invalid CTL I/O type %d\n",
12387                             __func__, io->io_hdr.io_type);
12388                 }
12389                 sbuf_cat(&sb, path_str);
12390                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12391                             (intmax_t)time_uptime - io->io_hdr.start_time);
12392                 sbuf_finish(&sb);
12393                 printf("%s", sbuf_data(&sb));
12394         }
12395 #endif /* CTL_TIME_IO */
12396
12397 #ifdef CTL_IO_DELAY
12398         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12399                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12400         } else {
12401                 struct ctl_lun *lun;
12402
12403                 lun = CTL_LUN(io);
12404                 if ((lun != NULL)
12405                  && (lun->delay_info.datamove_delay > 0)) {
12406
12407                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12408                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12409                         callout_reset(&io->io_hdr.delay_callout,
12410                                       lun->delay_info.datamove_delay * hz,
12411                                       ctl_datamove_timer_wakeup, io);
12412                         if (lun->delay_info.datamove_type ==
12413                             CTL_DELAY_TYPE_ONESHOT)
12414                                 lun->delay_info.datamove_delay = 0;
12415                         return;
12416                 }
12417         }
12418 #endif
12419
12420         /*
12421          * This command has been aborted.  Set the port status, so we fail
12422          * the data move.
12423          */
12424         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12425                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12426                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12427                        io->io_hdr.nexus.targ_port,
12428                        io->io_hdr.nexus.targ_lun);
12429                 io->io_hdr.port_status = 31337;
12430                 /*
12431                  * Note that the backend, in this case, will get the
12432                  * callback in its context.  In other cases it may get
12433                  * called in the frontend's interrupt thread context.
12434                  */
12435                 io->scsiio.be_move_done(io);
12436                 return;
12437         }
12438
12439         /* Don't confuse frontend with zero length data move. */
12440         if (io->scsiio.kern_data_len == 0) {
12441                 io->scsiio.be_move_done(io);
12442                 return;
12443         }
12444
12445         fe_datamove = CTL_PORT(io)->fe_datamove;
12446         fe_datamove(io);
12447 }
12448
12449 static void
12450 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12451 {
12452         union ctl_ha_msg msg;
12453 #ifdef CTL_TIME_IO
12454         struct bintime cur_bt;
12455 #endif
12456
12457         memset(&msg, 0, sizeof(msg));
12458         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12459         msg.hdr.original_sc = io;
12460         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12461         msg.hdr.nexus = io->io_hdr.nexus;
12462         msg.hdr.status = io->io_hdr.status;
12463         msg.scsi.kern_data_resid = io->scsiio.kern_data_resid;
12464         msg.scsi.tag_num = io->scsiio.tag_num;
12465         msg.scsi.tag_type = io->scsiio.tag_type;
12466         msg.scsi.scsi_status = io->scsiio.scsi_status;
12467         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12468                io->scsiio.sense_len);
12469         msg.scsi.sense_len = io->scsiio.sense_len;
12470         msg.scsi.port_status = io->io_hdr.port_status;
12471         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12472         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12473                 ctl_failover_io(io, /*have_lock*/ have_lock);
12474                 return;
12475         }
12476         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12477             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12478             msg.scsi.sense_len, M_WAITOK);
12479
12480 #ifdef CTL_TIME_IO
12481         getbinuptime(&cur_bt);
12482         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12483         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12484 #endif
12485         io->io_hdr.num_dmas++;
12486 }
12487
12488 /*
12489  * The DMA to the remote side is done, now we need to tell the other side
12490  * we're done so it can continue with its data movement.
12491  */
12492 static void
12493 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12494 {
12495         union ctl_io *io;
12496         uint32_t i;
12497
12498         io = rq->context;
12499
12500         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12501                 printf("%s: ISC DMA write failed with error %d", __func__,
12502                        rq->ret);
12503                 ctl_set_internal_failure(&io->scsiio,
12504                                          /*sks_valid*/ 1,
12505                                          /*retry_count*/ rq->ret);
12506         }
12507
12508         ctl_dt_req_free(rq);
12509
12510         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12511                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12512         free(io->io_hdr.remote_sglist, M_CTL);
12513         io->io_hdr.remote_sglist = NULL;
12514         io->io_hdr.local_sglist = NULL;
12515
12516         /*
12517          * The data is in local and remote memory, so now we need to send
12518          * status (good or back) back to the other side.
12519          */
12520         ctl_send_datamove_done(io, /*have_lock*/ 0);
12521 }
12522
12523 /*
12524  * We've moved the data from the host/controller into local memory.  Now we
12525  * need to push it over to the remote controller's memory.
12526  */
12527 static int
12528 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12529 {
12530         int retval;
12531
12532         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12533                                           ctl_datamove_remote_write_cb);
12534         return (retval);
12535 }
12536
12537 static void
12538 ctl_datamove_remote_write(union ctl_io *io)
12539 {
12540         int retval;
12541         void (*fe_datamove)(union ctl_io *io);
12542
12543         /*
12544          * - Get the data from the host/HBA into local memory.
12545          * - DMA memory from the local controller to the remote controller.
12546          * - Send status back to the remote controller.
12547          */
12548
12549         retval = ctl_datamove_remote_sgl_setup(io);
12550         if (retval != 0)
12551                 return;
12552
12553         /* Switch the pointer over so the FETD knows what to do */
12554         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12555
12556         /*
12557          * Use a custom move done callback, since we need to send completion
12558          * back to the other controller, not to the backend on this side.
12559          */
12560         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12561
12562         fe_datamove = CTL_PORT(io)->fe_datamove;
12563         fe_datamove(io);
12564 }
12565
12566 static int
12567 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12568 {
12569 #if 0
12570         char str[256];
12571         char path_str[64];
12572         struct sbuf sb;
12573 #endif
12574         uint32_t i;
12575
12576         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12577                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12578         free(io->io_hdr.remote_sglist, M_CTL);
12579         io->io_hdr.remote_sglist = NULL;
12580         io->io_hdr.local_sglist = NULL;
12581
12582 #if 0
12583         scsi_path_string(io, path_str, sizeof(path_str));
12584         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12585         sbuf_cat(&sb, path_str);
12586         scsi_command_string(&io->scsiio, NULL, &sb);
12587         sbuf_printf(&sb, "\n");
12588         sbuf_cat(&sb, path_str);
12589         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12590                     io->scsiio.tag_num, io->scsiio.tag_type);
12591         sbuf_cat(&sb, path_str);
12592         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12593                     io->io_hdr.flags, io->io_hdr.status);
12594         sbuf_finish(&sb);
12595         printk("%s", sbuf_data(&sb));
12596 #endif
12597
12598
12599         /*
12600          * The read is done, now we need to send status (good or bad) back
12601          * to the other side.
12602          */
12603         ctl_send_datamove_done(io, /*have_lock*/ 0);
12604
12605         return (0);
12606 }
12607
12608 static void
12609 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12610 {
12611         union ctl_io *io;
12612         void (*fe_datamove)(union ctl_io *io);
12613
12614         io = rq->context;
12615
12616         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12617                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12618                        rq->ret);
12619                 ctl_set_internal_failure(&io->scsiio,
12620                                          /*sks_valid*/ 1,
12621                                          /*retry_count*/ rq->ret);
12622         }
12623
12624         ctl_dt_req_free(rq);
12625
12626         /* Switch the pointer over so the FETD knows what to do */
12627         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12628
12629         /*
12630          * Use a custom move done callback, since we need to send completion
12631          * back to the other controller, not to the backend on this side.
12632          */
12633         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12634
12635         /* XXX KDM add checks like the ones in ctl_datamove? */
12636
12637         fe_datamove = CTL_PORT(io)->fe_datamove;
12638         fe_datamove(io);
12639 }
12640
12641 static int
12642 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12643 {
12644         struct ctl_sg_entry *local_sglist;
12645         uint32_t len_to_go;
12646         int retval;
12647         int i;
12648
12649         retval = 0;
12650         local_sglist = io->io_hdr.local_sglist;
12651         len_to_go = io->scsiio.kern_data_len;
12652
12653         /*
12654          * The difficult thing here is that the size of the various
12655          * S/G segments may be different than the size from the
12656          * remote controller.  That'll make it harder when DMAing
12657          * the data back to the other side.
12658          */
12659         for (i = 0; len_to_go > 0; i++) {
12660                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12661                 local_sglist[i].addr =
12662                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12663
12664                 len_to_go -= local_sglist[i].len;
12665         }
12666         /*
12667          * Reset the number of S/G entries accordingly.  The original
12668          * number of S/G entries is available in rem_sg_entries.
12669          */
12670         io->scsiio.kern_sg_entries = i;
12671
12672 #if 0
12673         printf("%s: kern_sg_entries = %d\n", __func__,
12674                io->scsiio.kern_sg_entries);
12675         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12676                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12677                        local_sglist[i].addr, local_sglist[i].len);
12678 #endif
12679
12680         return (retval);
12681 }
12682
12683 static int
12684 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12685                          ctl_ha_dt_cb callback)
12686 {
12687         struct ctl_ha_dt_req *rq;
12688         struct ctl_sg_entry *remote_sglist, *local_sglist;
12689         uint32_t local_used, remote_used, total_used;
12690         int i, j, isc_ret;
12691
12692         rq = ctl_dt_req_alloc();
12693
12694         /*
12695          * If we failed to allocate the request, and if the DMA didn't fail
12696          * anyway, set busy status.  This is just a resource allocation
12697          * failure.
12698          */
12699         if ((rq == NULL)
12700          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12701              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12702                 ctl_set_busy(&io->scsiio);
12703
12704         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12705             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12706
12707                 if (rq != NULL)
12708                         ctl_dt_req_free(rq);
12709
12710                 /*
12711                  * The data move failed.  We need to return status back
12712                  * to the other controller.  No point in trying to DMA
12713                  * data to the remote controller.
12714                  */
12715
12716                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12717
12718                 return (1);
12719         }
12720
12721         local_sglist = io->io_hdr.local_sglist;
12722         remote_sglist = io->io_hdr.remote_sglist;
12723         local_used = 0;
12724         remote_used = 0;
12725         total_used = 0;
12726
12727         /*
12728          * Pull/push the data over the wire from/to the other controller.
12729          * This takes into account the possibility that the local and
12730          * remote sglists may not be identical in terms of the size of
12731          * the elements and the number of elements.
12732          *
12733          * One fundamental assumption here is that the length allocated for
12734          * both the local and remote sglists is identical.  Otherwise, we've
12735          * essentially got a coding error of some sort.
12736          */
12737         isc_ret = CTL_HA_STATUS_SUCCESS;
12738         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12739                 uint32_t cur_len;
12740                 uint8_t *tmp_ptr;
12741
12742                 rq->command = command;
12743                 rq->context = io;
12744
12745                 /*
12746                  * Both pointers should be aligned.  But it is possible
12747                  * that the allocation length is not.  They should both
12748                  * also have enough slack left over at the end, though,
12749                  * to round up to the next 8 byte boundary.
12750                  */
12751                 cur_len = MIN(local_sglist[i].len - local_used,
12752                               remote_sglist[j].len - remote_used);
12753                 rq->size = cur_len;
12754
12755                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12756                 tmp_ptr += local_used;
12757
12758 #if 0
12759                 /* Use physical addresses when talking to ISC hardware */
12760                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12761                         /* XXX KDM use busdma */
12762                         rq->local = vtophys(tmp_ptr);
12763                 } else
12764                         rq->local = tmp_ptr;
12765 #else
12766                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12767                     ("HA does not support BUS_ADDR"));
12768                 rq->local = tmp_ptr;
12769 #endif
12770
12771                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12772                 tmp_ptr += remote_used;
12773                 rq->remote = tmp_ptr;
12774
12775                 rq->callback = NULL;
12776
12777                 local_used += cur_len;
12778                 if (local_used >= local_sglist[i].len) {
12779                         i++;
12780                         local_used = 0;
12781                 }
12782
12783                 remote_used += cur_len;
12784                 if (remote_used >= remote_sglist[j].len) {
12785                         j++;
12786                         remote_used = 0;
12787                 }
12788                 total_used += cur_len;
12789
12790                 if (total_used >= io->scsiio.kern_data_len)
12791                         rq->callback = callback;
12792
12793 #if 0
12794                 printf("%s: %s: local %p remote %p size %d\n", __func__,
12795                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12796                        rq->local, rq->remote, rq->size);
12797 #endif
12798
12799                 isc_ret = ctl_dt_single(rq);
12800                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
12801                         break;
12802         }
12803         if (isc_ret != CTL_HA_STATUS_WAIT) {
12804                 rq->ret = isc_ret;
12805                 callback(rq);
12806         }
12807
12808         return (0);
12809 }
12810
12811 static void
12812 ctl_datamove_remote_read(union ctl_io *io)
12813 {
12814         int retval;
12815         uint32_t i;
12816
12817         /*
12818          * This will send an error to the other controller in the case of a
12819          * failure.
12820          */
12821         retval = ctl_datamove_remote_sgl_setup(io);
12822         if (retval != 0)
12823                 return;
12824
12825         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12826                                           ctl_datamove_remote_read_cb);
12827         if (retval != 0) {
12828                 /*
12829                  * Make sure we free memory if there was an error..  The
12830                  * ctl_datamove_remote_xfer() function will send the
12831                  * datamove done message, or call the callback with an
12832                  * error if there is a problem.
12833                  */
12834                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12835                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12836                 free(io->io_hdr.remote_sglist, M_CTL);
12837                 io->io_hdr.remote_sglist = NULL;
12838                 io->io_hdr.local_sglist = NULL;
12839         }
12840 }
12841
12842 /*
12843  * Process a datamove request from the other controller.  This is used for
12844  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12845  * first.  Once that is complete, the data gets DMAed into the remote
12846  * controller's memory.  For reads, we DMA from the remote controller's
12847  * memory into our memory first, and then move it out to the FETD.
12848  */
12849 static void
12850 ctl_datamove_remote(union ctl_io *io)
12851 {
12852
12853         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12854
12855         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12856                 ctl_failover_io(io, /*have_lock*/ 0);
12857                 return;
12858         }
12859
12860         /*
12861          * Note that we look for an aborted I/O here, but don't do some of
12862          * the other checks that ctl_datamove() normally does.
12863          * We don't need to run the datamove delay code, since that should
12864          * have been done if need be on the other controller.
12865          */
12866         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12867                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
12868                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12869                        io->io_hdr.nexus.targ_port,
12870                        io->io_hdr.nexus.targ_lun);
12871                 io->io_hdr.port_status = 31338;
12872                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12873                 return;
12874         }
12875
12876         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
12877                 ctl_datamove_remote_write(io);
12878         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
12879                 ctl_datamove_remote_read(io);
12880         else {
12881                 io->io_hdr.port_status = 31339;
12882                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12883         }
12884 }
12885
12886 static void
12887 ctl_process_done(union ctl_io *io)
12888 {
12889         struct ctl_softc *softc = CTL_SOFTC(io);
12890         struct ctl_port *port = CTL_PORT(io);
12891         struct ctl_lun *lun = CTL_LUN(io);
12892         void (*fe_done)(union ctl_io *io);
12893         union ctl_ha_msg msg;
12894
12895         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12896         fe_done = port->fe_done;
12897
12898 #ifdef CTL_TIME_IO
12899         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12900                 char str[256];
12901                 char path_str[64];
12902                 struct sbuf sb;
12903
12904                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12905                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12906
12907                 sbuf_cat(&sb, path_str);
12908                 switch (io->io_hdr.io_type) {
12909                 case CTL_IO_SCSI:
12910                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12911                         sbuf_printf(&sb, "\n");
12912                         sbuf_cat(&sb, path_str);
12913                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12914                                     io->scsiio.tag_num, io->scsiio.tag_type);
12915                         break;
12916                 case CTL_IO_TASK:
12917                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12918                                     "Tag Type: %d\n", io->taskio.task_action,
12919                                     io->taskio.tag_num, io->taskio.tag_type);
12920                         break;
12921                 default:
12922                         panic("%s: Invalid CTL I/O type %d\n",
12923                             __func__, io->io_hdr.io_type);
12924                 }
12925                 sbuf_cat(&sb, path_str);
12926                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12927                             (intmax_t)time_uptime - io->io_hdr.start_time);
12928                 sbuf_finish(&sb);
12929                 printf("%s", sbuf_data(&sb));
12930         }
12931 #endif /* CTL_TIME_IO */
12932
12933         switch (io->io_hdr.io_type) {
12934         case CTL_IO_SCSI:
12935                 break;
12936         case CTL_IO_TASK:
12937                 if (ctl_debug & CTL_DEBUG_INFO)
12938                         ctl_io_error_print(io, NULL);
12939                 fe_done(io);
12940                 return;
12941         default:
12942                 panic("%s: Invalid CTL I/O type %d\n",
12943                     __func__, io->io_hdr.io_type);
12944         }
12945
12946         if (lun == NULL) {
12947                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12948                                  io->io_hdr.nexus.targ_mapped_lun));
12949                 goto bailout;
12950         }
12951
12952         mtx_lock(&lun->lun_lock);
12953
12954         /*
12955          * Check to see if we have any informational exception and status
12956          * of this command can be modified to report it in form of either
12957          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
12958          */
12959         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
12960             io->io_hdr.status == CTL_SUCCESS &&
12961             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
12962                 uint8_t mrie = lun->MODE_IE.mrie;
12963                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
12964                     (lun->MODE_VER.byte3 & SMS_VER_PER));
12965                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
12966                      mrie == SIEP_MRIE_REC_UNCOND ||
12967                      mrie == SIEP_MRIE_NO_SENSE) &&
12968                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
12969                      CTL_CMD_FLAG_NO_SENSE) == 0) {
12970                         ctl_set_sense(&io->scsiio,
12971                               /*current_error*/ 1,
12972                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
12973                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
12974                               /*asc*/ lun->ie_asc,
12975                               /*ascq*/ lun->ie_ascq,
12976                               SSD_ELEM_NONE);
12977                         lun->ie_reported = 1;
12978                 }
12979         } else if (lun->ie_reported < 0)
12980                 lun->ie_reported = 0;
12981
12982         /*
12983          * Check to see if we have any errors to inject here.  We only
12984          * inject errors for commands that don't already have errors set.
12985          */
12986         if (!STAILQ_EMPTY(&lun->error_list) &&
12987             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
12988             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
12989                 ctl_inject_error(lun, io);
12990
12991         /*
12992          * XXX KDM how do we treat commands that aren't completed
12993          * successfully?
12994          *
12995          * XXX KDM should we also track I/O latency?
12996          */
12997         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
12998             io->io_hdr.io_type == CTL_IO_SCSI) {
12999                 int type;
13000 #ifdef CTL_TIME_IO
13001                 struct bintime bt;
13002
13003                 getbinuptime(&bt);
13004                 bintime_sub(&bt, &io->io_hdr.start_bt);
13005 #endif
13006                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13007                     CTL_FLAG_DATA_IN)
13008                         type = CTL_STATS_READ;
13009                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13010                     CTL_FLAG_DATA_OUT)
13011                         type = CTL_STATS_WRITE;
13012                 else
13013                         type = CTL_STATS_NO_IO;
13014
13015 #ifdef CTL_LEGACY_STATS
13016                 uint32_t targ_port = port->targ_port;
13017                 lun->legacy_stats.ports[targ_port].bytes[type] +=
13018                     io->scsiio.kern_total_len;
13019                 lun->legacy_stats.ports[targ_port].operations[type] ++;
13020                 lun->legacy_stats.ports[targ_port].num_dmas[type] +=
13021                     io->io_hdr.num_dmas;
13022 #ifdef CTL_TIME_IO
13023                 bintime_add(&lun->legacy_stats.ports[targ_port].dma_time[type],
13024                    &io->io_hdr.dma_bt);
13025                 bintime_add(&lun->legacy_stats.ports[targ_port].time[type],
13026                     &bt);
13027 #endif
13028 #endif /* CTL_LEGACY_STATS */
13029
13030                 lun->stats.bytes[type] += io->scsiio.kern_total_len;
13031                 lun->stats.operations[type] ++;
13032                 lun->stats.dmas[type] += io->io_hdr.num_dmas;
13033 #ifdef CTL_TIME_IO
13034                 bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt);
13035                 bintime_add(&lun->stats.time[type], &bt);
13036 #endif
13037
13038                 mtx_lock(&port->port_lock);
13039                 port->stats.bytes[type] += io->scsiio.kern_total_len;
13040                 port->stats.operations[type] ++;
13041                 port->stats.dmas[type] += io->io_hdr.num_dmas;
13042 #ifdef CTL_TIME_IO
13043                 bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt);
13044                 bintime_add(&port->stats.time[type], &bt);
13045 #endif
13046                 mtx_unlock(&port->port_lock);
13047         }
13048
13049         /*
13050          * Remove this from the OOA queue.
13051          */
13052         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13053 #ifdef CTL_TIME_IO
13054         if (TAILQ_EMPTY(&lun->ooa_queue))
13055                 lun->last_busy = getsbinuptime();
13056 #endif
13057
13058         /*
13059          * Run through the blocked queue on this LUN and see if anything
13060          * has become unblocked, now that this transaction is done.
13061          */
13062         ctl_check_blocked(lun);
13063
13064         /*
13065          * If the LUN has been invalidated, free it if there is nothing
13066          * left on its OOA queue.
13067          */
13068         if ((lun->flags & CTL_LUN_INVALID)
13069          && TAILQ_EMPTY(&lun->ooa_queue)) {
13070                 mtx_unlock(&lun->lun_lock);
13071                 mtx_lock(&softc->ctl_lock);
13072                 ctl_free_lun(lun);
13073                 mtx_unlock(&softc->ctl_lock);
13074         } else
13075                 mtx_unlock(&lun->lun_lock);
13076
13077 bailout:
13078
13079         /*
13080          * If this command has been aborted, make sure we set the status
13081          * properly.  The FETD is responsible for freeing the I/O and doing
13082          * whatever it needs to do to clean up its state.
13083          */
13084         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13085                 ctl_set_task_aborted(&io->scsiio);
13086
13087         /*
13088          * If enabled, print command error status.
13089          */
13090         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13091             (ctl_debug & CTL_DEBUG_INFO) != 0)
13092                 ctl_io_error_print(io, NULL);
13093
13094         /*
13095          * Tell the FETD or the other shelf controller we're done with this
13096          * command.  Note that only SCSI commands get to this point.  Task
13097          * management commands are completed above.
13098          */
13099         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13100             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13101                 memset(&msg, 0, sizeof(msg));
13102                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13103                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13104                 msg.hdr.nexus = io->io_hdr.nexus;
13105                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13106                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13107                     M_WAITOK);
13108         }
13109
13110         fe_done(io);
13111 }
13112
13113 /*
13114  * Front end should call this if it doesn't do autosense.  When the request
13115  * sense comes back in from the initiator, we'll dequeue this and send it.
13116  */
13117 int
13118 ctl_queue_sense(union ctl_io *io)
13119 {
13120         struct ctl_softc *softc = CTL_SOFTC(io);
13121         struct ctl_port *port = CTL_PORT(io);
13122         struct ctl_lun *lun;
13123         struct scsi_sense_data *ps;
13124         uint32_t initidx, p, targ_lun;
13125
13126         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13127
13128         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13129
13130         /*
13131          * LUN lookup will likely move to the ctl_work_thread() once we
13132          * have our new queueing infrastructure (that doesn't put things on
13133          * a per-LUN queue initially).  That is so that we can handle
13134          * things like an INQUIRY to a LUN that we don't have enabled.  We
13135          * can't deal with that right now.
13136          * If we don't have a LUN for this, just toss the sense information.
13137          */
13138         mtx_lock(&softc->ctl_lock);
13139         if (targ_lun >= CTL_MAX_LUNS ||
13140             (lun = softc->ctl_luns[targ_lun]) == NULL) {
13141                 mtx_unlock(&softc->ctl_lock);
13142                 goto bailout;
13143         }
13144         mtx_lock(&lun->lun_lock);
13145         mtx_unlock(&softc->ctl_lock);
13146
13147         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13148         p = initidx / CTL_MAX_INIT_PER_PORT;
13149         if (lun->pending_sense[p] == NULL) {
13150                 lun->pending_sense[p] = malloc(sizeof(*ps) * CTL_MAX_INIT_PER_PORT,
13151                     M_CTL, M_NOWAIT | M_ZERO);
13152         }
13153         if ((ps = lun->pending_sense[p]) != NULL) {
13154                 ps += initidx % CTL_MAX_INIT_PER_PORT;
13155                 memset(ps, 0, sizeof(*ps));
13156                 memcpy(ps, &io->scsiio.sense_data, io->scsiio.sense_len);
13157         }
13158         mtx_unlock(&lun->lun_lock);
13159
13160 bailout:
13161         ctl_free_io(io);
13162         return (CTL_RETVAL_COMPLETE);
13163 }
13164
13165 /*
13166  * Primary command inlet from frontend ports.  All SCSI and task I/O
13167  * requests must go through this function.
13168  */
13169 int
13170 ctl_queue(union ctl_io *io)
13171 {
13172         struct ctl_port *port = CTL_PORT(io);
13173
13174         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13175
13176 #ifdef CTL_TIME_IO
13177         io->io_hdr.start_time = time_uptime;
13178         getbinuptime(&io->io_hdr.start_bt);
13179 #endif /* CTL_TIME_IO */
13180
13181         /* Map FE-specific LUN ID into global one. */
13182         io->io_hdr.nexus.targ_mapped_lun =
13183             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13184
13185         switch (io->io_hdr.io_type) {
13186         case CTL_IO_SCSI:
13187         case CTL_IO_TASK:
13188                 if (ctl_debug & CTL_DEBUG_CDB)
13189                         ctl_io_print(io);
13190                 ctl_enqueue_incoming(io);
13191                 break;
13192         default:
13193                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13194                 return (EINVAL);
13195         }
13196
13197         return (CTL_RETVAL_COMPLETE);
13198 }
13199
13200 #ifdef CTL_IO_DELAY
13201 static void
13202 ctl_done_timer_wakeup(void *arg)
13203 {
13204         union ctl_io *io;
13205
13206         io = (union ctl_io *)arg;
13207         ctl_done(io);
13208 }
13209 #endif /* CTL_IO_DELAY */
13210
13211 void
13212 ctl_serseq_done(union ctl_io *io)
13213 {
13214         struct ctl_lun *lun = CTL_LUN(io);;
13215
13216         if (lun->be_lun == NULL ||
13217             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13218                 return;
13219         mtx_lock(&lun->lun_lock);
13220         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13221         ctl_check_blocked(lun);
13222         mtx_unlock(&lun->lun_lock);
13223 }
13224
13225 void
13226 ctl_done(union ctl_io *io)
13227 {
13228
13229         /*
13230          * Enable this to catch duplicate completion issues.
13231          */
13232 #if 0
13233         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13234                 printf("%s: type %d msg %d cdb %x iptl: "
13235                        "%u:%u:%u tag 0x%04x "
13236                        "flag %#x status %x\n",
13237                         __func__,
13238                         io->io_hdr.io_type,
13239                         io->io_hdr.msg_type,
13240                         io->scsiio.cdb[0],
13241                         io->io_hdr.nexus.initid,
13242                         io->io_hdr.nexus.targ_port,
13243                         io->io_hdr.nexus.targ_lun,
13244                         (io->io_hdr.io_type ==
13245                         CTL_IO_TASK) ?
13246                         io->taskio.tag_num :
13247                         io->scsiio.tag_num,
13248                         io->io_hdr.flags,
13249                         io->io_hdr.status);
13250         } else
13251                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13252 #endif
13253
13254         /*
13255          * This is an internal copy of an I/O, and should not go through
13256          * the normal done processing logic.
13257          */
13258         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13259                 return;
13260
13261 #ifdef CTL_IO_DELAY
13262         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13263                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13264         } else {
13265                 struct ctl_lun *lun = CTL_LUN(io);
13266
13267                 if ((lun != NULL)
13268                  && (lun->delay_info.done_delay > 0)) {
13269
13270                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13271                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13272                         callout_reset(&io->io_hdr.delay_callout,
13273                                       lun->delay_info.done_delay * hz,
13274                                       ctl_done_timer_wakeup, io);
13275                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13276                                 lun->delay_info.done_delay = 0;
13277                         return;
13278                 }
13279         }
13280 #endif /* CTL_IO_DELAY */
13281
13282         ctl_enqueue_done(io);
13283 }
13284
13285 static void
13286 ctl_work_thread(void *arg)
13287 {
13288         struct ctl_thread *thr = (struct ctl_thread *)arg;
13289         struct ctl_softc *softc = thr->ctl_softc;
13290         union ctl_io *io;
13291         int retval;
13292
13293         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13294
13295         while (!softc->shutdown) {
13296                 /*
13297                  * We handle the queues in this order:
13298                  * - ISC
13299                  * - done queue (to free up resources, unblock other commands)
13300                  * - RtR queue
13301                  * - incoming queue
13302                  *
13303                  * If those queues are empty, we break out of the loop and
13304                  * go to sleep.
13305                  */
13306                 mtx_lock(&thr->queue_lock);
13307                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13308                 if (io != NULL) {
13309                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13310                         mtx_unlock(&thr->queue_lock);
13311                         ctl_handle_isc(io);
13312                         continue;
13313                 }
13314                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13315                 if (io != NULL) {
13316                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13317                         /* clear any blocked commands, call fe_done */
13318                         mtx_unlock(&thr->queue_lock);
13319                         ctl_process_done(io);
13320                         continue;
13321                 }
13322                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13323                 if (io != NULL) {
13324                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13325                         mtx_unlock(&thr->queue_lock);
13326                         if (io->io_hdr.io_type == CTL_IO_TASK)
13327                                 ctl_run_task(io);
13328                         else
13329                                 ctl_scsiio_precheck(softc, &io->scsiio);
13330                         continue;
13331                 }
13332                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13333                 if (io != NULL) {
13334                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13335                         mtx_unlock(&thr->queue_lock);
13336                         retval = ctl_scsiio(&io->scsiio);
13337                         if (retval != CTL_RETVAL_COMPLETE)
13338                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13339                         continue;
13340                 }
13341
13342                 /* Sleep until we have something to do. */
13343                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13344         }
13345         thr->thread = NULL;
13346         kthread_exit();
13347 }
13348
13349 static void
13350 ctl_lun_thread(void *arg)
13351 {
13352         struct ctl_softc *softc = (struct ctl_softc *)arg;
13353         struct ctl_be_lun *be_lun;
13354
13355         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13356
13357         while (!softc->shutdown) {
13358                 mtx_lock(&softc->ctl_lock);
13359                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13360                 if (be_lun != NULL) {
13361                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13362                         mtx_unlock(&softc->ctl_lock);
13363                         ctl_create_lun(be_lun);
13364                         continue;
13365                 }
13366
13367                 /* Sleep until we have something to do. */
13368                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13369                     PDROP | PRIBIO, "-", 0);
13370         }
13371         softc->lun_thread = NULL;
13372         kthread_exit();
13373 }
13374
13375 static void
13376 ctl_thresh_thread(void *arg)
13377 {
13378         struct ctl_softc *softc = (struct ctl_softc *)arg;
13379         struct ctl_lun *lun;
13380         struct ctl_logical_block_provisioning_page *page;
13381         const char *attr;
13382         union ctl_ha_msg msg;
13383         uint64_t thres, val;
13384         int i, e, set;
13385
13386         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13387
13388         while (!softc->shutdown) {
13389                 mtx_lock(&softc->ctl_lock);
13390                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13391                         if ((lun->flags & CTL_LUN_DISABLED) ||
13392                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13393                             lun->backend->lun_attr == NULL)
13394                                 continue;
13395                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13396                             softc->ha_mode == CTL_HA_MODE_XFER)
13397                                 continue;
13398                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13399                                 continue;
13400                         e = 0;
13401                         page = &lun->MODE_LBP;
13402                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13403                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13404                                         continue;
13405                                 thres = scsi_4btoul(page->descr[i].count);
13406                                 thres <<= CTL_LBP_EXPONENT;
13407                                 switch (page->descr[i].resource) {
13408                                 case 0x01:
13409                                         attr = "blocksavail";
13410                                         break;
13411                                 case 0x02:
13412                                         attr = "blocksused";
13413                                         break;
13414                                 case 0xf1:
13415                                         attr = "poolblocksavail";
13416                                         break;
13417                                 case 0xf2:
13418                                         attr = "poolblocksused";
13419                                         break;
13420                                 default:
13421                                         continue;
13422                                 }
13423                                 mtx_unlock(&softc->ctl_lock); // XXX
13424                                 val = lun->backend->lun_attr(
13425                                     lun->be_lun->be_lun, attr);
13426                                 mtx_lock(&softc->ctl_lock);
13427                                 if (val == UINT64_MAX)
13428                                         continue;
13429                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13430                                     == SLBPPD_ARMING_INC)
13431                                         e = (val >= thres);
13432                                 else
13433                                         e = (val <= thres);
13434                                 if (e)
13435                                         break;
13436                         }
13437                         mtx_lock(&lun->lun_lock);
13438                         if (e) {
13439                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13440                                     (uint8_t *)page, lun->ua_tpt_info);
13441                                 if (lun->lasttpt == 0 ||
13442                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13443                                         lun->lasttpt = time_uptime;
13444                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13445                                         set = 1;
13446                                 } else
13447                                         set = 0;
13448                         } else {
13449                                 lun->lasttpt = 0;
13450                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13451                                 set = -1;
13452                         }
13453                         mtx_unlock(&lun->lun_lock);
13454                         if (set != 0 &&
13455                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13456                                 /* Send msg to other side. */
13457                                 bzero(&msg.ua, sizeof(msg.ua));
13458                                 msg.hdr.msg_type = CTL_MSG_UA;
13459                                 msg.hdr.nexus.initid = -1;
13460                                 msg.hdr.nexus.targ_port = -1;
13461                                 msg.hdr.nexus.targ_lun = lun->lun;
13462                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13463                                 msg.ua.ua_all = 1;
13464                                 msg.ua.ua_set = (set > 0);
13465                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13466                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13467                                 mtx_unlock(&softc->ctl_lock); // XXX
13468                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13469                                     sizeof(msg.ua), M_WAITOK);
13470                                 mtx_lock(&softc->ctl_lock);
13471                         }
13472                 }
13473                 mtx_sleep(&softc->thresh_thread, &softc->ctl_lock,
13474                     PDROP | PRIBIO, "-", CTL_LBP_PERIOD * hz);
13475         }
13476         softc->thresh_thread = NULL;
13477         kthread_exit();
13478 }
13479
13480 static void
13481 ctl_enqueue_incoming(union ctl_io *io)
13482 {
13483         struct ctl_softc *softc = CTL_SOFTC(io);
13484         struct ctl_thread *thr;
13485         u_int idx;
13486
13487         idx = (io->io_hdr.nexus.targ_port * 127 +
13488                io->io_hdr.nexus.initid) % worker_threads;
13489         thr = &softc->threads[idx];
13490         mtx_lock(&thr->queue_lock);
13491         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13492         mtx_unlock(&thr->queue_lock);
13493         wakeup(thr);
13494 }
13495
13496 static void
13497 ctl_enqueue_rtr(union ctl_io *io)
13498 {
13499         struct ctl_softc *softc = CTL_SOFTC(io);
13500         struct ctl_thread *thr;
13501
13502         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13503         mtx_lock(&thr->queue_lock);
13504         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13505         mtx_unlock(&thr->queue_lock);
13506         wakeup(thr);
13507 }
13508
13509 static void
13510 ctl_enqueue_done(union ctl_io *io)
13511 {
13512         struct ctl_softc *softc = CTL_SOFTC(io);
13513         struct ctl_thread *thr;
13514
13515         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13516         mtx_lock(&thr->queue_lock);
13517         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13518         mtx_unlock(&thr->queue_lock);
13519         wakeup(thr);
13520 }
13521
13522 static void
13523 ctl_enqueue_isc(union ctl_io *io)
13524 {
13525         struct ctl_softc *softc = CTL_SOFTC(io);
13526         struct ctl_thread *thr;
13527
13528         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13529         mtx_lock(&thr->queue_lock);
13530         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13531         mtx_unlock(&thr->queue_lock);
13532         wakeup(thr);
13533 }
13534
13535 /*
13536  *  vim: ts=8
13537  */